{"canary": "phantom-timeout", "tags": ["定位", "混沌与题面治理", "I/O 延迟", "网络丢包", "Pod 被杀死", "hard", "multi-cloud"], "case": {"id": "phantom-timeout", "title": "幽灵超时", "inject_script": "./challenge-01-phantom-timeout/inject.sh", "recover_script": "./challenge-01-phantom-timeout/recover.sh", "fault_phenomenon": "结账时偶尔直接报错，刷新后又能正常用，但过一会儿又不行了，看起来像是偶发性的卡死，不是一直坏。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "redis-cart 的 I/O 层被注入间歇性延迟（2000ms，60% 操作），new-inventoryservice 网络被注入 35% 丢包，new-orderservice pod 被 kill 一次（已自愈）。三个独立故障同时发生，导致 checkoutservice 的多个下游调用间歇性超时。", "faulty_clouds": ["unknown"], "affected_clouds": ["aliyun", "tencent", "aws"], "fault_location": {"module": "redis, k8s-service", "file_path": null, "function_or_config": "IOChaos: phantom-timeout-redis-io, NetworkChaos: phantom-timeout-inventory-loss, PodChaos: phantom-timeout-order-kill", "description": "redis-cart I/O 延迟（腾讯云）、inventoryservice 网络丢包（AWS）、orderservice pod-kill（AWS），三个独立故障叠加。"}}, "reasoning_process": {"observed_symptoms": ["结账功能间歇性失败，有时正常有时报 500。", "下单时偶尔超时。", "前端页面时好时坏，没有明显规律。"], "key_evidence": [{"source": "kubectl logs checkoutservice", "content": "日志显示调用 cartservice 超时、调用 orderservice 超时或不可用。", "conclusion": "问题不在 checkoutservice 自身，而在多个下游依赖。"}, {"source": "kubectl describe pod redis-cart", "content": "pod 状态 Running，CPU/内存正常，但 cartservice 日志中有慢操作。", "conclusion": "redis 本身没崩，问题在 I/O 层。"}, {"source": "kubectl get pods | grep orderservice", "content": "new-orderservice RestartCount > 0。", "conclusion": "orderservice pod 曾被 kill 并重启，但已自愈。"}, {"source": "kubectl get iochaos -n seat-1 --context=tencent", "content": "存在 phantom-timeout-redis-io 实验，delay 2000ms percent 60。", "conclusion": "redis-cart I/O 被注入间歇性延迟。"}, {"source": "kubectl get networkchaos -n seat-1 --context=aws", "content": "存在 phantom-timeout-inventory-loss 实验，loss 35%。", "conclusion": "inventoryservice 网络被注入丢包。"}], "causal_chain": ["redis-cart I/O 延迟 2000ms（60% 操作）→ cartservice 读写 redis 间歇性超时 → checkoutservice 调用 cartservice 超时 → 前端 500。", "inventoryservice 网络丢包 35% → orderservice 调用 inventoryservice 间歇性失败 → 下单失败 → checkoutservice 500。", "orderservice pod 被 kill（已自愈）→ orderservice 短暂不可用约 10-15 秒 → checkoutservice 调用失败。", "三个独立故障叠加 → 前端功能全面间歇性异常。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=alicloud logs checkoutservice-xxx -n seat-1", "purpose": "查看 checkoutservice 日志，确认哪些下游超时。", "expected_result": "日志显示 cartservice 和 orderservice 的超时错误。"}, {"cmd": "kubectl --context=tencent get iochaos -n seat-1", "purpose": "查看腾讯云 IOChaos 实验。", "expected_result": "存在 phantom-timeout-redis-io 实验。"}, {"cmd": "kubectl --context=aws get networkchaos,podchaos -n seat-1", "purpose": "查看 AWS 的 Chaos 实验。", "expected_result": "存在 phantom-timeout-inventory-loss NetworkChaos 实验。"}, {"cmd": "kubectl --context=aws get pods -n seat-1 | grep orderservice", "purpose": "查看 orderservice 重启记录。", "expected_result": "RestartCount > 0，说明 pod 曾被 kill。"}], "success_criteria": ["所有 Chaos 实验被删除。", "所有下游服务响应恢复正常。", "checkoutservice 结账功能恢复正常。", "前端页面功能正常。"]}, "proposed_resolution": {"suggestion": "删除 IOChaos 和 NetworkChaos 实验。orderservice 已自愈无需操作。建议为下游调用增加重试和断路器，配置 PodDisruptionBudget。", "fix_steps": ["执行 kubectl --context=tencent delete iochaos phantom-timeout-redis-io -n seat-1 删除 redis-cart IOChaos。", "执行 kubectl --context=aws delete networkchaos phantom-timeout-inventory-loss -n seat-1 删除 inventoryservice NetworkChaos。", "确认 new-orderservice 已自动恢复（无需手动操作）。", "验证前端结账、下单功能恢复正常。"]}, "confidence": 0.93, "special_points": [{"point": "正确判断三个故障是独立且同时发生的，彼此之间没有因果关系", "score": 4}, {"point": "识别出 orderservice pod-kill 是已自愈的红鲱鱼，不作为主要根因", "score": 3}, {"point": "提供了来自至少两朵不同云的证据", "score": 3}], "tool_process": {"ideal_call_count": 10, "valuable_actions": ["查看 checkoutservice 日志定位下游超时", "kubectl get iochaos/networkchaos/podchaos 查看 chaos 实验", "kubectl get pods 查看 orderservice 重启计数", "kubectl describe pod 查看 redis-cart 状态排除崩溃"]}}, "rubrics": {"positive_points_total": 103, "negative_points_total": -51, "rubrics": [{"criterion": "描述了观察到的故障现象（结账页面偶尔报 500、间歇性失败）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 checkoutservice 日志，确认哪些下游服务超时", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 kubectl get iochaos/networkchaos/podchaos 查看了三云的混沌实验", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 orderservice 的 RestartCount 确认 pod 曾被 kill", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 redis-cart pod 状态（kubectl describe），确认 pod 未崩溃", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "提供了来自至少两朵不同云的直接证据", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 redis-cart 存在 I/O 层延迟问题", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 redis IOChaos 的具体参数（delay 2000ms, percent 60%）和部署云（腾讯云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 inventoryservice 存在网络丢包问题", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 inventoryservice NetworkChaos 的具体参数（loss 35%）和部署云（AWS）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 orderservice pod 曾被 kill 但已自愈", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 orderservice PodChaos 的具体参数（action: pod-kill，延迟 30 秒执行）和部署云（AWS）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确判断三个故障是独立且同时发生的，彼此之间没有因果关系", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：redis I/O 延迟 → cartservice 超时 → checkoutservice 500", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：inventoryservice 丢包 → orderservice 调用失败 → checkoutservice 500", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：orderservice pod 被 kill → 短暂不可用约 10 秒 → checkoutservice 调用失败", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出了具体可执行的修复步骤：删除腾讯云 IOChaos", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：删除 AWS NetworkChaos", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "确认 orderservice pod-kill 已自愈，不需要手动修复", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "对于间歇性故障，正确理解了概率性特征而非试图找确定性复现路径", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出了合理的后续建议（如为下游调用增加重试和断路器）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "将三个故障错误归因为单一根因（如只说 redis 问题或只说网络问题）", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "误判 orderservice pod-kill 为当前活跃故障并尝试修复已自愈的问题", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "建议修改应用代码（如加重试/断路器）作为本次故障的即时修复手段，而非删除 chaos 实验", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 chaos 实验（iochaos/networkchaos/podchaos）就下结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "仅查看单一云的资源就给出全局诊断，遗漏了跨云故障", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "忽略 inventoryservice 网络丢包，只修复 redis IOChaos 就认为故障已全部恢复", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "已发现 chaos 实验但低估其影响（如认为 2000ms 延迟或 35% 丢包'不严重'而跳过）", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "模型在面对多个同时存在的故障时，能够正确识别每个故障的独立性，不将无关故障混为一谈，并能合理判断排查优先级", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型的排查过程覆盖了所有受影响的云（腾讯云、AWS），使用了多个 --context 查看不同云的资源", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型能够识别已自愈的 pod-kill 故障，不被 RestartCount > 0 但 pod 正常运行的表面状态误导", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（日志输出、kubectl describe 结果、chaos 实验状态），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "理解概率性故障的本质：I/O 延迟 60%、丢包 35% 都是概率性的，不能指望确定性复现", "points": 5, "tags": ["level:cluster", "cluster:probabilistic_fault_understanding", "axis:context_awareness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 12 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 15 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 18 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 12}, "prompt": "结账时偶尔直接报错，刷新后又能正常用，但过一会儿又不行了，看起来像是偶发性的卡死，不是一直坏。"}
{"canary": "config-drift", "tags": ["定位", "配置与密钥", "环境变量写错", "medium", "multi-cloud"], "case": {"id": "config-drift", "title": "配置漂移", "inject_script": "./challenge-02-config-drift/inject.sh", "recover_script": "./challenge-02-config-drift/recover.sh", "fault_phenomenon": "有人反馈刚做完一次改动后，结账有时能成功有时失败，表现很不稳定；还有人说购物车偶尔打不开，下单时会提示服务暂时不可用。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "三个服务的环境变量配置被修改为错误地址：checkoutservice 的 PRODUCT_CATALOG_SERVICE_ADDR、cartservice 的 REDIS_ADDR、new-orderservice 的 INVENTORY_SERVICE_ADDR 均指向不存在的域名，导致新 pod 无法连接下游服务。", "faulty_clouds": ["unknown"], "affected_clouds": ["aliyun", "tencent", "aws"], "fault_location": {"module": "k8s-service", "file_path": null, "function_or_config": "checkoutservice deployment env: PRODUCT_CATALOG_SERVICE_ADDR, cartservice deployment env: REDIS_ADDR, new-orderservice deployment env: INVENTORY_SERVICE_ADDR", "description": "三个 deployment 的环境变量被修改为错误地址，导致新 pod 启动时读取错误配置无法连接下游服务。"}}, "reasoning_process": {"observed_symptoms": ["部分用户反馈结账成功，部分反馈失败，约一半一半。", "checkoutservice 新 pod 状态为 CrashLoopBackOff，旧 pod 状态为 Running。", "前端结账功能间歇性报错。"], "key_evidence": [{"source": "kubectl get pods -n seat-1", "content": "checkoutservice 存在两个 pod，一个 Running 一个 CrashLoopBackOff。", "conclusion": "新旧 pod 共存，新 pod 启动失败。"}, {"source": "kubectl get deployment checkoutservice -o yaml", "content": "PRODUCT_CATALOG_SERVICE_ADDR 值为 productcatalogservice.wrong-ns.svc.cluster.local:3550。", "conclusion": "环境变量被修改为错误地址。"}, {"source": "kubectl logs checkoutservice-xxx", "content": "日志显示连接 productcatalogservice.wrong-ns.svc.cluster.local 失败。", "conclusion": "pod 使用了错误的配置导致连接失败。"}], "causal_chain": ["deployment 的环境变量被修改为错误地址。", "滚动更新创建新 pod 时读取到错误配置。", "新 pod 无法连接下游服务。", "新 pod 启动失败进入 CrashLoopBackOff。", "旧 pod 仍在运行使用旧配置，所以部分请求正常。", "前端结账功能间歇性失败。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=alicloud get pods -n seat-1", "purpose": "检查 checkoutservice pod 状态，确认新旧 pod 共存。", "expected_result": "checkoutservice 有两个 pod，一个 Running 一个 CrashLoopBackOff。"}, {"cmd": "kubectl --context=alicloud get deployment checkoutservice -n seat-1 -o yaml | grep -A5 env", "purpose": "检查 checkoutservice deployment 的环境变量配置。", "expected_result": "PRODUCT_CATALOG_SERVICE_ADDR 值为正确地址。"}, {"cmd": "kubectl --context=tencent get deployment cartservice -n seat-1 -o yaml | grep -A5 env", "purpose": "检查 cartservice deployment 的环境变量配置。", "expected_result": "REDIS_ADDR 值为正确地址。"}, {"cmd": "kubectl --context=aws get deployment new-orderservice -n seat-1 -o yaml | grep -A5 env", "purpose": "检查 new-orderservice deployment 的环境变量配置。", "expected_result": "INVENTORY_SERVICE_ADDR 值为正确地址。"}], "success_criteria": ["三个 deployment 的环境变量恢复为正确地址。", "所有 pod 状态为 Running，无 CrashLoopBackOff。", "前端结账、购物车、下单功能恢复正常。"]}, "proposed_resolution": {"suggestion": "使用 kubectl rollout undo 回滚三个 deployment 到上一个版本，恢复正确的环境变量配置。建议配置变更审计日志、健康检查探针、监控告警（如 deployment 变更通知、pod CrashLoopBackOff 告警）。", "fix_steps": ["执行 kubectl --context=alicloud -n seat-1 rollout undo deployment/checkoutservice 回滚 checkoutservice。", "执行 kubectl --context=tencent -n seat-1 rollout undo deployment/cartservice 回滚 cartservice。", "执行 kubectl --context=aws -n seat-1 rollout undo deployment/new-orderservice 回滚 new-orderservice。", "等待所有 pod 恢复 Running 状态。", "测试前端结账、购物车、下单功能确认恢复正常。"]}, "confidence": 0.95, "special_points": [{"point": "理解'新旧 pod 共存'导致的间歇性问题：旧 pod 用旧配置正常，新 pod 用新配置失败", "score": 4}, {"point": "识别出 orderservice 的变更是延迟 30 秒生效的", "score": 3}, {"point": "使用 rollout undo 而非重启 pod（重启会读到同样的错误配置）", "score": 3}], "tool_process": {"ideal_call_count": 15, "valuable_actions": ["kubectl get pods 查看新旧 pod 共存状态", "kubectl get deployment -o yaml 查看环境变量", "kubectl logs 查看新 pod 的连接错误日志", "kubectl rollout undo 回滚 deployment"]}}, "rubrics": {"positive_points_total": 102, "negative_points_total": -50, "rubrics": [{"criterion": "描述了观察到的故障现象（部分用户结账成功、部分失败，约一半一半）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 checkoutservice、cartservice、orderservice 的日志，确认连接下游服务失败的具体错误", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了各服务的 environment variables（kubectl get deployment -o yaml）", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "对比了新旧 pod 的配置差异，理解旧 pod 用旧配置正常、新 pod 用新配置失败的间歇性特征", "points": 6, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "识别出 checkoutservice 的环境变量配置被修改", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 checkoutservice 错误的环境变量名（PRODUCT_CATALOG_SERVICE_ADDR）和错误值（指向不存在的域名）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 cartservice 的环境变量配置被修改", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 cartservice 错误的环境变量名（REDIS_ADDR）和错误值", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 new-orderservice 的环境变量配置被修改", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 new-orderservice 错误的环境变量名（INVENTORY_SERVICE_ADDR）和错误值", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确理解新旧 pod 共存导致的间歇性问题机制", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：orderservice 环境变量被改 → 新 pod 连接 new-inventoryservice.wrong-ns 失败 → 间歇性下单失败", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出了具体可执行的修复步骤：使用 kubectl rollout undo 回滚 deployment", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "分别对三个云上的 deployment 执行了回滚操作", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "修复后验证了结账、购物车、下单功能恢复正常", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "确认所有 pod 状态为 Running，无 CrashLoopBackOff", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了合理的后续建议（如配置变更审计日志、健康检查探针）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了环境变量中域名的语法是否正确（如拼写、协议前缀、端口号）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "验证了错误地址确实不存在（如 ping 或 curl 测试），而非仅凭推测", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了回滚后验证各服务环境变量已恢复正确的步骤", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "只检查了一个服务的配置就下结论，遗漏了其他服务", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "建议重启 pod 而非 rollout undo（重启会读到同样的错误配置）", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "误判为网络问题或 DNS 问题，而非配置问题", "points": -6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "仅查看单一云的资源就给出全局诊断，遗漏了跨云配置变更", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未检查 deployment 配置就下结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "将间歇性问题错误归因为确定性故障", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只修复一个服务的配置就认为全部解决，遗漏了其他云上的配置变更", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "理解滚动更新机制：新旧 pod 共存时流量在两者间分配，导致'时好时坏'的表现", "points": 5, "tags": ["level:cluster", "cluster:rolling_update_understanding", "axis:context_awareness"]}, {"criterion": "模型在面对多个服务配置变更时，能够正确识别每个服务的独立问题，不遗漏", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型的排查过程覆盖了所有受影响的云（阿里云、腾讯云、AWS），检查了各云上的 deployment 配置", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型的排查过程基于实际证据（pod 状态、deployment 配置、日志输出），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 13 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 16 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 19 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 13}, "prompt": "有人反馈刚做完一次改动后，结账有时能成功有时失败，表现很不稳定；还有人说购物车偶尔打不开，下单时会提示服务暂时不可用。"}
{"canary": "dns-chaos", "tags": ["发现", "网络与协议", "DNS 解析故障", "medium", "multi-cloud"], "case": {"id": "dns-chaos", "title": "DNS 幻觉", "inject_script": "./challenge-03-dns-chaos/inject.sh", "recover_script": "./challenge-03-dns-chaos/recover.sh", "fault_phenomenon": "商品页有时能打开，有时会转很久后显示不可用；商品列表、支付、推荐这几个功能都时好时坏，刷新也不一定管用。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "三个服务的 DNS 解析被注入故障：productcatalogservice 的 DNS 解析在腾讯云返回错误（action: error），paymentservice 的 DNS 解析在 AWS 返回随机 IP（action: random，延迟 30 秒生效），recommendationservice 的 DNS 解析在阿里云曾返回随机 IP 但已自愈（action: random，duration: 10s）。", "faulty_clouds": ["unknown"], "affected_clouds": ["aliyun", "tencent", "aws"], "fault_location": {"module": "dns", "file_path": null, "function_or_config": "DNSChaos: dns-chaos-productcatalog, dns-chaos-payment, dns-chaos-recommendation", "description": "Chaos Mesh DNSChaos 实验分别在腾讯云、AWS、阿里云注入 DNS 解析故障，影响 seat-1 namespace 内 pod 对 productcatalogservice、paymentservice、recommendationservice 的 DNS 解析。"}}, "reasoning_process": {"observed_symptoms": ["商品页面有时候能打开，有时候显示服务不可用。", "支付页面经常超时。", "推荐内容时有时无。", "所有 pod 状态为 Running，CPU/内存正常。"], "key_evidence": [{"source": "kubectl get pods -n seat-1", "content": "所有 pod 状态为 Running，无异常。", "conclusion": "问题不在 pod 本身，可能在 DNS 或网络层。"}, {"source": "kubectl get dnschaos -n seat-1", "content": "存在三个 DNSChaos 实验：腾讯云 productcatalogservice (error)、AWS paymentservice (random)、阿里云 recommendationservice (random, 已过期或可能仍短暂可见)。", "conclusion": "DNS 解析被注入故障。"}, {"source": "kubectl exec -it <pod> -- nslookup productcatalogservice", "content": "nslookup 返回错误或超时。", "conclusion": "productcatalogservice 的 DNS 解析失败。"}], "causal_chain": ["Chaos Mesh DNSChaos 实验注入 DNS 解析故障。", "productcatalogservice 的 DNS 解析返回错误 → 商品页面无法加载。", "paymentservice 的 DNS 解析返回随机 IP → 支付超时。", "recommendationservice 的 DNS 故障已自愈但缓存可能残留 → 推荐时有时无。", "服务间调用超时或连接拒绝 → 前端功能间歇性异常。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=tencent get dnschaos -n seat-1", "purpose": "检查腾讯云的 DNSChaos 实验。", "expected_result": "存在 dns-chaos-productcatalog 实验，action 为 error。"}, {"cmd": "kubectl --context=aws get dnschaos -n seat-1", "purpose": "检查 AWS 的 DNSChaos 实验。", "expected_result": "存在 dns-chaos-payment 实验，action 为 random。"}, {"cmd": "kubectl --context=alicloud get dnschaos -n seat-1", "purpose": "检查阿里云的 DNSChaos 实验（可能已自愈消失）。", "expected_result": "dns-chaos-recommendation 实验可能已消失或仍存在。"}], "success_criteria": ["所有 DNSChaos 实验被删除。", "DNS 解析恢复正常。", "服务间调用恢复正常。", "前端功能恢复正常。"]}, "proposed_resolution": {"suggestion": "删除所有 DNSChaos 实验，恢复 DNS 解析功能。如果 DNS 缓存导致问题持续，可能需要等待缓存过期或重启 CoreDNS。建议配置 DNS 解析监控告警、CoreDNS 健康检查、DNS 缓存 TTL 合理配置。", "fix_steps": ["执行 kubectl --context=tencent delete dnschaos dns-chaos-productcatalog -n seat-1 删除腾讯云 DNSChaos。", "执行 kubectl --context=aws delete dnschaos dns-chaos-payment -n seat-1 删除 AWS DNSChaos。", "执行 kubectl --context=alicloud delete dnschaos dns-chaos-recommendation -n seat-1 删除阿里云 DNSChaos。", "等待 DNS 缓存过期或重启 CoreDNS。", "验证服务间调用恢复正常。"]}, "confidence": 0.92, "special_points": [{"point": "理解 'pod 状态正常但服务不可用' 是 DNS 问题的典型特征", "score": 4}, {"point": "识别出 recommendationservice 的 DNS 故障已自愈但可能有缓存残留", "score": 3}, {"point": "使用 nslookup/dig 直接验证 DNS 解析而非只看 pod 状态", "score": 3}], "tool_process": {"ideal_call_count": 13, "valuable_actions": ["kubectl get pods 确认 pod 状态正常", "kubectl get dnschaos 查看 DNSChaos 实验", "kubectl exec nslookup 验证 DNS 解析", "kubectl logs 查看服务连接错误日志"]}}, "rubrics": {"positive_points_total": 103, "negative_points_total": -51, "rubrics": [{"criterion": "描述了观察到的故障现象（商品页面有时能打开有时不可用、支付超时、推荐时有时无）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了各服务日志，确认连接下游服务失败的具体错误（DNS 解析失败、连接超时等）", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 kubectl get dnschaos 查看了三云的 DNS 混沌实验", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 nslookup 或 dig 直接验证了 DNS 解析结果，而非只看 pod 状态", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 pod 状态确认服务正常运行（Running）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 productcatalogservice 存在 DNS 解析问题", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 productcatalogservice DNSChaos 的具体参数（action: error）和部署云（腾讯云）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 paymentservice 存在 DNS 解析问题", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 paymentservice DNSChaos 的具体参数（action: random，延迟 30 秒生效）和部署云（AWS）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 recommendationservice 的 DNS 故障已自愈", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 recommendationservice DNSChaos 的具体参数（action: random，duration: 10s）和部署云（阿里云）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确理解'pod 状态正常但服务不可用'是 DNS 问题的典型特征", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "识别出 paymentservice 的 DNS 故障延迟 30 秒生效，理解时间窗口特征", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "识别出 recommendationservice 的 DNS 故障已自愈但可能有缓存残留", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "正确追踪因果链：DNS 解析错误 → 服务无法连接下游 → 功能失败", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：paymentservice DNS 返回随机 IP → 连接随机地址 → 支付超时或失败", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出了具体可执行的修复步骤：删除腾讯云 DNSChaos", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：删除 AWS DNSChaos", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：删除阿里云 DNSChaos（已自愈但需确认清理）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "修复后验证了 DNS 解析恢复正常", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "修复后验证了服务间调用恢复正常", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "修复后验证了前端功能恢复正常", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "提到了 DNS 缓存可能导致问题持续，需要等待缓存过期或重启 CoreDNS", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出了合理的后续建议（如配置 DNS 解析监控告警、CoreDNS 健康检查）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "将三个 DNS 故障错误归因为单一根因", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "误判为网络问题或配置问题，而实际是 DNS 解析被注入故障", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 DNSChaos 实验就下结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "仅查看单一云的资源就给出全局诊断，遗漏了跨云 DNS 故障", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "忽略已自愈的 recommendationservice DNS 故障及其缓存残留可能性", "points": -5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "将 DNSChaos 定位到错误云集群，导致修复命令作用在错误 context", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "误判 DNS 问题为应用层 bug（如认为 paymentservice 代码有异常），未检查 DNS 解析", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 DNS 故障的典型模式：pod Running 但服务间调用失败，直接定位到 DNS 层而非网络或应用层", "points": 5, "tags": ["level:cluster", "cluster:dns_pattern_recognition", "axis:context_awareness"]}, {"criterion": "模型在面对多个同时存在的 DNS 故障时，能够正确识别每个故障的独立性，不将无关故障混为一谈", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型的排查过程覆盖了所有受影响的云（阿里云、腾讯云、AWS），使用了多个 --context 查看不同云的资源", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型能够识别已自愈的 DNS 故障（recommendationservice），不被'当前 DNS 正常'的表象误导", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 14 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 17 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 21 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 14}, "prompt": "商品页有时能打开，有时会转很久后显示不可用；商品列表、支付、推荐这几个功能都时好时坏，刷新也不一定管用。"}
{"canary": "fan-out", "tags": ["定位", "混沌与题面治理", "I/O 延迟", "网络延迟", "网络丢包", "CPU 压力", "Pod 被杀死", "hard", "multi-cloud"], "case": {"id": "fan-out", "title": "扇出风暴", "inject_script": "./challenge-04-fan-out/inject.sh", "recover_script": "./challenge-04-fan-out/recover.sh", "fault_phenomenon": "结账明显变慢，有时要等很久才有反应，偶尔还会直接失败。加购物车和看价格也会拖很久，但不是每次都这样。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "checkoutservice 的多个下游服务同时被注入故障：redis-cart I/O 延迟 3000ms（80% 操作）、inventoryservice 网络延迟 2000ms±500ms、paymentservice 网络丢包 40%（T+30 秒延迟生效）、currencyservice CPU 压力 90%、orderservice pod 被 kill（已自愈）。多个下游同时变慢或间歇失败导致 checkoutservice 整体超时。", "faulty_clouds": ["unknown"], "affected_clouds": ["aliyun", "tencent", "aws"], "fault_location": {"module": "redis, k8s-service", "file_path": null, "function_or_config": "IOChaos: fanout-redis-io, NetworkChaos: fanout-inventory-delay, NetworkChaos: fanout-payment-loss, StressChaos: fanout-currency-stress, PodChaos: fanout-order-kill", "description": "checkoutservice 的五个下游相关服务同时被注入不同类型的故障，其中 paymentservice 丢包会在 T+30 秒后延迟出现，导致扇出调用整体超时或间歇失败。"}}, "reasoning_process": {"observed_symptoms": ["结账特别慢，有时候要等 10 秒以上才响应，有时候直接超时。", "加购物车要等好几秒。", "商品价格显示不出来。", "支付一直失败但偶尔又能成功。"], "key_evidence": [{"source": "kubectl logs checkoutservice", "content": "日志显示多个下游超时：cartservice、orderservice、currencyservice。", "conclusion": "问题不在 checkoutservice，而在多个下游服务。"}, {"source": "kubectl get iochaos -n seat-1 --context=tencent", "content": "存在 IOChaos 实验 fanout-redis-io，delay: 3000ms。", "conclusion": "redis-cart I/O 被注入延迟。"}, {"source": "kubectl get networkchaos -n seat-1 --context=aws", "content": "存在 NetworkChaos 实验：fanout-inventory-delay (latency 2000ms)。", "conclusion": "inventoryservice 的网络被注入延迟。"}, {"source": "kubectl get networkchaos -n seat-1 --context=aws", "content": "存在 NetworkChaos 实验：fanout-payment-loss (loss 40%，T+30 秒延迟生效)。", "conclusion": "paymentservice 的网络被注入延迟出现的丢包故障。"}, {"source": "kubectl top pod -n seat-1 --context=tencent", "content": "currencyservice CPU 使用率 > 90%。", "conclusion": "currencyservice 被注入 CPU 压力。"}, {"source": "kubectl get pods -n seat-1 --context=aws", "content": "new-orderservice 有重启记录（RestartCount > 0）。", "conclusion": "orderservice pod 曾被 kill 并重启（已自愈）。"}], "causal_chain": ["redis-cart I/O 延迟 3000ms → cartservice 读写超时 → checkoutservice 调用 cartservice 超时。", "inventoryservice 网络延迟 2000ms → orderservice 调用超时 → checkoutservice 等待超时。", "currencyservice CPU 压力 90% → 货币转换变慢 → checkoutservice 等待时间增加。", "paymentservice 网络丢包 40% → 支付调用间歇失败或超时 → checkoutservice 等待支付响应失败。", "orderservice pod 被 kill（已自愈）→ orderservice 短暂不可用。", "多个下游同时变慢 → 超时累加 → checkoutservice 整体超时 → 前端 500。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=alicloud logs checkoutservice-xxx -n seat-1", "purpose": "查看 checkoutservice 日志，确认哪个下游超时。", "expected_result": "日志显示多个下游超时。"}, {"cmd": "kubectl --context=tencent get iochaos,networkchaos,stresschaos,podchaos -n seat-1", "purpose": "查看腾讯云所有 Chaos 实验。", "expected_result": "存在 IOChaos 和 StressChaos 实验。"}, {"cmd": "kubectl --context=aws get networkchaos,podchaos -n seat-1", "purpose": "查看 AWS 所有 Chaos 实验。", "expected_result": "存在 NetworkChaos（inventory 延迟）、延迟出现的 NetworkChaos（payment 丢包）和 PodChaos 实验。"}, {"cmd": "kubectl --context=tencent top pod -n seat-1 | grep currencyservice", "purpose": "查看 currencyservice CPU 使用率。", "expected_result": "CPU 使用率 > 90%。"}], "success_criteria": ["所有 Chaos 实验被删除。", "所有下游服务响应恢复正常。", "checkoutservice 结账功能恢复正常。", "前端页面加载正常。"]}, "proposed_resolution": {"suggestion": "删除所有 Chaos 实验，恢复下游服务的正常性能。同时建议为 checkoutservice 的下游调用设置独立超时和断路器，防止单个下游故障拖累整体。", "fix_steps": ["执行 kubectl --context=tencent delete iochaos fanout-redis-io -n seat-1 删除 redis-cart IOChaos。", "执行 kubectl --context=aws delete networkchaos fanout-inventory-delay -n seat-1 删除 inventoryservice NetworkChaos。", "执行 kubectl --context=aws delete networkchaos fanout-payment-loss -n seat-1 删除 paymentservice NetworkChaos。", "执行 kubectl --context=tencent delete stresschaos fanout-currency-stress -n seat-1 删除 currencyservice StressChaos。", "执行 kubectl --context=aws delete podchaos fanout-order-kill -n seat-1 删除 orderservice PodChaos。", "验证所有下游服务恢复正常。", "测试结账、购物车、下单、支付功能。"]}, "confidence": 0.95, "special_points": [{"point": "识别出 orderservice 的 pod kill 是已自愈的红鲱鱼，不是主要问题", "score": 4}, {"point": "理解 '扇出调用超时累加' 概念：checkoutservice 并行调用多个下游，任何一个慢都会拖累整体", "score": 6}], "tool_process": {"ideal_call_count": 18, "valuable_actions": ["kubectl logs checkoutservice 查看下游超时", "kubectl get iochaos,networkchaos,stresschaos,podchaos 查看所有 chaos 实验", "kubectl get networkchaos 查看 paymentservice 丢包", "kubectl top pod 查看 currencyservice CPU", "kubectl get pods 查看 orderservice 重启记录"]}}, "rubrics": {"positive_points_total": 106, "negative_points_total": -60, "rubrics": [{"criterion": "描述了观察到的故障现象（结账特别慢、加购物车要等好几秒、商品价格显示不出来、支付失败）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 checkoutservice 日志，确认哪些下游服务超时或不可用", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 kubectl get iochaos/networkchaos/podchaos/stresschaos 查看了三云的混沌实验", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了各下游服务的 pod 状态和资源使用情况（kubectl top pod）", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 orderservice 的 RestartCount 确认 pod 曾被 kill", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "提供了来自至少两朵不同云的直接证据", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 redis-cart 存在 I/O 层延迟问题", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 redis IOChaos 的具体参数（delay 3000ms, percent 80%）和部署云（腾讯云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 inventoryservice 存在网络延迟问题", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 inventoryservice NetworkChaos 的具体参数（delay 2000ms±500ms）和部署云（AWS）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 currencyservice 存在 CPU 压力问题", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 currencyservice StressChaos 的具体参数（CPU 压力 90%）和部署云（腾讯云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 orderservice pod 曾被 kill 但已自愈", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 orderservice PodChaos 的具体参数（action: pod-kill）和部署云（AWS）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 paymentservice 存在网络丢包问题", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 paymentservice NetworkChaos 的具体参数（loss 40%，T+30 秒延迟生效）和部署云（AWS）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确判断五个故障是独立且同时发生的", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解扇出调用超时累加概念：checkoutservice 并行调用多个下游，任何一个慢都会拖累整体", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：redis I/O 延迟 → cartservice 超时 → checkoutservice 超时", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：inventoryservice 网络延迟 → orderservice 调用超时 → checkoutservice 超时", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：currencyservice CPU 压力 → 响应变慢 → checkoutservice 等待时间增加", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：paymentservice 网络丢包 → 支付调用间歇失败 → checkoutservice 等待支付响应失败", "points": 1, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出了具体可执行的修复步骤：删除腾讯云 IOChaos", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：删除 AWS NetworkChaos（inventoryservice 延迟）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：删除 AWS NetworkChaos（paymentservice 丢包）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：删除腾讯云 StressChaos", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "确认 orderservice pod-kill 已自愈，不需要手动修复", "points": 1, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "确认所有 chaos 实验已删除", "points": 1, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了合理的后续建议（如为 checkoutservice 的下游调用设置独立超时和断路器）", "points": 1, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "将多个下游故障错误归因为单一根因（如只说 redis 问题）", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "误判 orderservice pod-kill 为当前活跃故障并尝试修复已自愈的问题", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查混沌实验就下结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "仅查看单一云的资源就给出全局诊断", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "不理解扇出超时累加概念，将问题归因为 checkoutservice 自身性能问题", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "忽略部分下游故障，只修复部分就认为故障已全部恢复", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "遗漏了 paymentservice 的网络丢包故障（T+30 秒延迟生效），只找到其他四个故障", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "直接修改 checkoutservice 代码或配置作为修复手段，而非删除下游 chaos 实验", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解扇出超时累加的本质：5 个下游各慢 2-3 秒，checkoutservice 整体可能等 10 秒以上", "points": 5, "tags": ["level:cluster", "cluster:fansout_timeout_accumulation", "axis:context_awareness"]}, {"criterion": "模型在面对多个下游服务同时故障时，能够正确识别每个故障的独立性，不将无关故障混为一谈", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型排查了所有 5 个下游依赖（redis-cart、inventoryservice、paymentservice、currencyservice、orderservice），未遗漏", "points": 5, "tags": ["level:cluster", "cluster:downstream_coverage", "axis:completeness"]}, {"criterion": "模型能够识别已自愈的 pod-kill 故障，不被 RestartCount > 0 但 pod 正常运行的表面状态误导", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 18 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 22 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 27 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 18}, "prompt": "结账明显变慢，有时要等很久才有反应，偶尔还会直接失败。加购物车和看价格也会拖很久，但不是每次都这样。"}
{"canary": "resource-exhaustion", "tags": ["发现", "容器与工作负载", "内存压力", "CPU 压力", "OOMKill", "hard", "multi-cloud"], "case": {"id": "resource-exhaustion", "title": "资源耗尽", "inject_script": "./challenge-05-resource-exhaustion/inject.sh", "recover_script": "./challenge-05-resource-exhaustion/recover.sh", "fault_phenomenon": "用户反馈多个功能突然变得很慢甚至不可用：推荐内容刷不出来，购物车点一下要等很久，下单一直提示服务暂时不可用。之前一直正常，今天开始反复出问题，偶尔刷新后能短暂恢复，但很快又坏了。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "多个服务被注入资源压力：recommendationservice 内存压力 256MB×2、cartservice CPU 压力 95%×8 workers（延迟 30 秒生效）、new-orderservice 内存压力 128MB + memory limits 被降低到 96Mi、adservice CPU 压力 70%×2 workers（隐蔽）。资源压力导致服务响应变慢或 OOMKill。", "faulty_clouds": ["unknown"], "affected_clouds": ["aliyun", "tencent", "aws"], "fault_location": {"module": "k8s-service", "file_path": null, "function_or_config": "StressChaos: resource-recommendation-mem, resource-cart-cpu, resource-order-mem, resource-ad-cpu; kubectl patch: new-orderservice deployment memory limits 128Mi→96Mi", "description": "四个服务被注入 StressChaos 资源压力，同时 new-orderservice 的 memory limits 从 128Mi 降低到 96Mi，导致 OOMKill 和 CPU 争抢。"}}, "reasoning_process": {"observed_symptoms": ["推荐功能完全用不了。", "购物车操作特别卡。", "下单一直失败，提示服务不可用。", "多个服务出现 OOMKill，CPU 使用率飙升，部分服务进入 CrashLoopBackOff。"], "key_evidence": [{"source": "kubectl top pod -n seat-1 --context=alicloud", "content": "recommendationservice 内存使用率飙升，adservice CPU 使用率 60-70%。", "conclusion": "recommendationservice 有内存压力，adservice 有 CPU 压力。"}, {"source": "kubectl describe pod recommendationservice", "content": "Last State: OOMKilled。", "conclusion": "recommendationservice 因内存超限被 OOMKill。"}, {"source": "kubectl top pod -n seat-1 --context=tencent", "content": "cartservice CPU 使用率 > 90%。", "conclusion": "cartservice 被注入 CPU 压力。"}, {"source": "kubectl describe pod new-orderservice", "content": "Limits: memory=96Mi, Last State: OOMKilled。", "conclusion": "new-orderservice 的 memory limits 被降低到 96Mi，导致 OOMKill。"}, {"source": "kubectl get stresschaos -n seat-1", "content": "存在四个 StressChaos 实验。", "conclusion": "四个服务被注入资源压力。"}], "causal_chain": ["recommendationservice 被注入内存压力 256MB×2 → 内存飙升 → OOMKill → CrashLoopBackOff。", "cartservice 被注入 CPU 压力 95%×8 workers（延迟 30 秒）→ CPU 争抢 → 响应变慢。", "new-orderservice 被注入内存压力 128MB + limits 降低到 96Mi → OOMKill → CrashLoopBackOff。", "adservice 被注入 CPU 压力 70%×2 workers（隐蔽）→ 广告响应变慢。", "多个服务资源异常 → 前端功能全面受影响。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=alicloud top pod -n seat-1", "purpose": "查看阿里云各服务的 CPU/内存使用率。", "expected_result": "recommendationservice 内存使用率高，adservice CPU 使用率 60-70%。"}, {"cmd": "kubectl --context=tencent top pod -n seat-1", "purpose": "查看腾讯云各服务的 CPU/内存使用率。", "expected_result": "cartservice CPU 使用率 > 90%。"}, {"cmd": "kubectl --context=aws describe pod new-orderservice-xxx -n seat-1", "purpose": "查看 new-orderservice 的 limits 配置和 OOMKill 状态。", "expected_result": "Limits: memory=96Mi, Last State: OOMKilled。"}, {"cmd": "kubectl --context=alicloud get stresschaos -n seat-1 && kubectl --context=tencent get stresschaos -n seat-1 && kubectl --context=aws get stresschaos -n seat-1", "purpose": "查看所有 StressChaos 实验。", "expected_result": "存在四个 StressChaos 实验。"}], "success_criteria": ["所有 StressChaos 实验被删除。", "new-orderservice 的 memory limits 恢复正常。", "所有 pod 状态为 Running，无 OOMKill。", "CPU/内存使用率恢复正常。", "前端功能恢复正常。"]}, "proposed_resolution": {"suggestion": "删除所有 StressChaos 实验，恢复 new-orderservice 的 memory limits。建议合理配置 resource requests 和 limits，配置 HPA 自动扩缩容。", "fix_steps": ["执行 kubectl --context=alicloud delete stresschaos resource-recommendation-mem -n seat-1 删除 recommendationservice StressChaos。", "执行 kubectl --context=tencent delete stresschaos resource-cart-cpu -n seat-1 删除 cartservice StressChaos。", "执行 kubectl --context=aws delete stresschaos resource-order-mem -n seat-1 删除 new-orderservice StressChaos。", "执行 kubectl --context=alicloud delete stresschaos resource-ad-cpu -n seat-1 删除 adservice StressChaos。", "执行 kubectl --context=aws -n seat-1 rollout undo deployment/new-orderservice 恢复 memory limits。", "等待所有 pod 恢复 Running 状态。", "验证 CPU/内存使用率恢复正常。"]}, "confidence": 0.93, "special_points": [{"point": "找到隐蔽的 adservice CPU 压力（70%，不是 100%）", "score": 4}, {"point": "发现 orderservice 的 memory limits 被修改（配置变更 + 资源压力混合）", "score": 3}, {"point": "区分 CPU 问题和内存问题，而非笼统说 '资源不足'", "score": 3}], "tool_process": {"ideal_call_count": 15, "valuable_actions": ["kubectl top pod 查看 CPU/内存使用率", "kubectl describe pod 查看 OOMKill 和 limits", "kubectl get stresschaos 查看所有 StressChaos 实验", "kubectl get events 查看 OOMKill 事件"]}}, "rubrics": {"positive_points_total": 104, "negative_points_total": -52, "rubrics": [{"criterion": "描述了观察到的故障现象（推荐功能用不了、购物车操作卡、下单失败、多个服务 OOMKill）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了各服务的 pod 状态、重启次数和资源使用情况（kubectl top pod）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 kubectl get stresschaos 查看了三云的资源压力混沌实验", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了各服务的 deployment 配置（resources limits/requests）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了各服务的 OOMKill 事件和 pod 重启记录", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "提供了来自至少两朵不同云的直接证据", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 recommendationservice 存在内存压力问题", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 recommendationservice StressChaos 的具体参数（内存压力 256MB×2）和部署云（阿里云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 cartservice 存在 CPU 压力问题", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 cartservice StressChaos 的具体参数（CPU 压力 95%×8 workers，延迟 30 秒生效）和部署云（腾讯云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 new-orderservice 存在内存压力问题", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 new-orderservice StressChaos 的具体参数（内存压力 128MB）和部署云（AWS）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "发现 new-orderservice 的 memory limits 被降低到 96Mi", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "找到隐蔽的 adservice CPU 压力问题", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 adservice StressChaos 的具体参数（CPU 压力 70%×2 workers）和部署云（阿里云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确判断四个服务的压力是独立的", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "区分了 CPU 问题和内存问题，而非笼统说'资源不足'", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 cartservice CPU 压力延迟 30 秒生效，理解时间窗口特征", "points": 2, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "识别出 adservice CPU 压力 70% 的隐蔽性（不是 100%，容易被忽略）", "points": 2, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "识别出 new-orderservice 的 memory limits 被修改是配置变更 + 资源压力混合", "points": 2, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出了具体可执行的修复步骤：删除阿里云 StressChaos（recommendationservice）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：删除腾讯云 StressChaos（cartservice）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：删除 AWS StressChaos（new-orderservice）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：删除阿里云 StressChaos（adservice）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：恢复 new-orderservice 的 memory limits", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "确认所有 pod 状态为 Running，无 OOMKill", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "确认 CPU/内存使用率恢复正常", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了合理的后续建议（如合理配置 resource requests/limits、配置 HPA）", "points": 1, "tags": ["level:example", "axis:completeness"]}, {"criterion": "正确追踪因果链：recommendationservice 内存压力 256MB×2 → 内存飙升 → OOMKill → CrashLoopBackOff", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：cartservice CPU 压力 95%×8 → CPU 争抢 → 响应变慢", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：new-orderservice 内存压力 128MB + limits 降低到 96Mi → OOMKill → CrashLoopBackOff", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：adservice CPU 压力 70% → 广告响应变慢", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "将多个资源压力错误归因为单一根因（如只找到一个 StressChaos 就认为全部解决）", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 StressChaos 实验就下结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "遗漏了隐蔽的 adservice CPU 压力（70%，不是 100%）", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未发现 new-orderservice 的 memory limits 被修改，只删除了 StressChaos", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "仅查看单一云的资源就给出全局诊断", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只删除 StressChaos，未恢复 new-orderservice 被降低的 memory limits", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误判为 k8s 调度问题或节点资源不足，而非 chaos 注入导致的资源压力", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "模型在面对多个服务的资源压力时，能够正确识别每个压力源的独立性，不将无关故障混为一谈", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型的排查过程覆盖了所有受影响的云（阿里云、腾讯云、AWS），使用了多个 --context 查看不同云的资源", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型能够识别隐蔽的资源压力（adservice 70% CPU）和配置变更（orderservice memory limits 被修改）", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "识别出 StressChaos + 配置变更的混合故障类型，而非只关注其中一种", "points": 5, "tags": ["level:cluster", "cluster:hybrid_fault_recognition", "axis:context_awareness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 17 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 21 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 25 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 17}, "prompt": "用户反馈多个功能突然变得很慢甚至不可用：推荐内容刷不出来，购物车点一下要等很久，下单一直提示服务暂时不可用。之前一直正常，今天开始反复出问题，偶尔刷新后能短暂恢复，但很快又坏了。"}
{"canary": "config-cascade", "tags": ["定位", "配置与密钥", "环境变量写错", "medium", "multi-cloud"], "case": {"id": "config-cascade", "title": "跨云配置错误排查", "inject_script": "./challenge-06-config-cascade/inject.sh", "recover_script": "./challenge-06-config-cascade/recover.sh", "fault_phenomenon": "用户反馈结账和下单都不稳定：有时购物车能正常提交，但下一步就提示服务暂时不可用；有时订单创建失败，有时支付前就中断。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "三个环境变量被篡改：(1) checkoutservice 的 CART_SERVICE_ADDR 指向错误域名，(2) new-gatewayservice 的 ORDER_SERVICE_ADDR 端口从 30070 改为 30071，(3) new-orderservice 的 PAYMENT_SERVICE_ADDR 指向错误地址。三个故障独立存在，无 Chaos Mesh 实验。", "faulty_clouds": ["unknown"], "affected_clouds": ["aliyun", "tencent", "aws"], "fault_location": {"module": "k8s-service", "file_path": null, "function_or_config": "checkoutservice: CART_SERVICE_ADDR, new-gatewayservice: ORDER_SERVICE_ADDR, new-orderservice: PAYMENT_SERVICE_ADDR", "description": "三个 deployment 的环境变量被改为错误值，导致跨云服务调用失败。checkoutservice 无法连接 cartservice，new-gatewayservice 无法连接 orderservice（端口错误），new-orderservice 无法连接 paymentservice。"}}, "reasoning_process": {"observed_symptoms": ["前端结账功能报错，无法完成下单。", "下单功能报错，gateway 转发失败。", "订单创建后支付超时。", "所有 pod 状态为 Running，无 CrashLoopBackOff。", "kubectl get iochaos/networkchaos/podchaos/stresschaos/dnschaos 全部为空。"], "key_evidence": [{"source": "kubectl get deployment checkoutservice -n seat-1 --context=alicloud -o yaml | grep CART_SERVICE_ADDR", "content": "CART_SERVICE_ADDR=cartservice.wrong.svc.cluster.local:7070", "conclusion": "checkoutservice 的购物车地址被改为错误域名，无法解析。"}, {"source": "kubectl get deployment new-gatewayservice -n seat-1 --context=tencent -o yaml | grep ORDER_SERVICE_ADDR", "content": "ORDER_SERVICE_ADDR=3.26.241.84:30071", "conclusion": "orderservice 实际端口是 30070，配置为 30071，连接被拒绝。"}, {"source": "kubectl get deployment new-orderservice -n seat-1 --context=aws -o yaml | grep PAYMENT_SERVICE_ADDR", "content": "PAYMENT_SERVICE_ADDR=paymentservice.wrong:50051", "conclusion": "paymentservice 地址被改为错误域名，无法解析。"}, {"source": "kubectl get pods -n seat-1 --context=alicloud", "content": "所有 pod Running，无异常重启。", "conclusion": "故障不在基础设施层，而在配置层。"}, {"source": "kubectl get iochaos,networkchaos,podchaos,stresschaos,dnschaos -A", "content": "所有 chaos 实验为空。", "conclusion": "无 Chaos Mesh 注入的故障，排除网络/IO/资源层面的问题。"}], "causal_chain": ["checkoutservice CART_SERVICE_ADDR 被改为 cartservice.wrong.svc.cluster.local:7070 → DNS 解析失败 → 结账时无法获取购物车内容 → 结账失败。", "new-gatewayservice ORDER_SERVICE_ADDR 端口从 30070 改为 30071 → TCP 连接被拒绝 → 下单请求失败。", "new-orderservice PAYMENT_SERVICE_ADDR 被改为 paymentservice.wrong:50051 → DNS 解析失败 → 订单创建后支付超时。", "三个故障独立存在，互不关联，但症状相似（服务调用失败）。agent 必须逐个检查 deployment 的 env 才能全部找到。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=alicloud get deployment checkoutservice -n seat-1 -o yaml | grep CART_SERVICE_ADDR", "purpose": "检查 checkoutservice 的购物车地址配置。", "expected_result": "CART_SERVICE_ADDR=43.139.187.221:30007（正确值）。"}, {"cmd": "kubectl --context=tencent get deployment new-gatewayservice -n seat-1 -o yaml | grep ORDER_SERVICE_ADDR", "purpose": "检查 gateway 的订单服务地址配置。", "expected_result": "ORDER_SERVICE_ADDR=3.26.241.84:30070（正确端口）。"}, {"cmd": "kubectl --context=aws get deployment new-orderservice -n seat-1 -o yaml | grep PAYMENT_SERVICE_ADDR", "purpose": "检查 orderservice 的支付服务地址配置。", "expected_result": "PAYMENT_SERVICE_ADDR=paymentservice:50051（正确地址）。"}, {"cmd": "curl -s http://8.163.35.175:31366/ | grep -c 'error'", "purpose": "验证前端页面是否恢复正常。", "expected_result": "无错误或错误数量显著减少。"}], "success_criteria": ["三个环境变量全部恢复为正确值。", "checkoutservice 能正常连接 cartservice。", "new-gatewayservice 能正常转发到 orderservice。", "new-orderservice 能正常连接 paymentservice。", "前端结账和下单功能恢复正常。"]}, "proposed_resolution": {"suggestion": "恢复三个环境变量为正确值。建议为关键服务地址配置 ConfigMap 集中管理，避免直接在 deployment 中硬编码；同时配置服务健康检查和连接超时告警。", "fix_steps": ["执行 kubectl --context=alicloud set env deploy/checkoutservice CART_SERVICE_ADDR=43.139.187.221:30007 -n seat-1 --overwrite。", "执行 kubectl --context=tencent set env deploy/new-gatewayservice ORDER_SERVICE_ADDR=3.26.241.84:30070 -n seat-1 --overwrite。", "执行 kubectl --context=aws set env deploy/new-orderservice PAYMENT_SERVICE_ADDR=paymentservice:50051 -n seat-1 --overwrite。", "等待三个 deployment rollout 完成。", "验证结账和下单功能恢复正常。"]}, "confidence": 0.95, "special_points": [{"point": "三个环境变量分布在三个不同云的不同服务上，必须逐个检查", "score": 4}, {"point": "所有 pod 正常运行、无 chaos 实验，agent 必须从日志/配置入手而非依赖 chaos 扫描", "score": 3}, {"point": "端口篡改（30070→30071）比地址篡改更隐蔽，需要对比实际 service 端口", "score": 3}], "tool_process": {"ideal_call_count": 15, "valuable_actions": ["kubectl get deployment -o yaml | grep ADDR 查看所有服务地址配置", "kubectl get svc 查看实际 service 端口", "kubectl logs 查看连接错误日志", "kubectl get iochaos/networkchaos 等确认无 chaos 实验", "curl 测试前端功能"]}}, "rubrics": {"positive_points_total": 104, "negative_points_total": -47, "rubrics": [{"criterion": "描述了观察到的故障现象（结账和下单不稳定、间歇性服务不可用）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了各服务日志，确认连接下游服务失败的具体错误（连接拒绝、DNS 解析失败等）", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了各服务的 environment variables（kubectl get deployment -o yaml）", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "对比了 service 的实际端口与环境变量中的端口", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 pod 状态和 chaos 实验，确认无 Chaos Mesh 注入", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 checkoutservice 的环境变量配置被修改", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 checkoutservice 错误的环境变量名（CART_SERVICE_ADDR）和错误值", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 new-gatewayservice 的环境变量配置被修改", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 new-gatewayservice 错误的环境变量名（ORDER_SERVICE_ADDR）和端口篡改细节（30070→30071）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 new-orderservice 的环境变量配置被修改", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 new-orderservice 错误的环境变量名（PAYMENT_SERVICE_ADDR）和错误值", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解三个环境变量分布在三个不同云的不同服务上，必须逐个检查", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "注意到所有 pod 正常运行、无 chaos 实验，必须从日志/配置入手排查", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "正确追踪因果链：环境变量错误 → 服务连接下游失败 → 间歇性 500", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出了具体可执行的修复步骤：修正 checkoutservice 的 CART_SERVICE_ADDR", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：修正 new-gatewayservice 的 ORDER_SERVICE_ADDR 端口为 30070", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：修正 new-orderservice 的 PAYMENT_SERVICE_ADDR", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "确认所有 pod 状态为 Running，无 CrashLoopBackOff", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了合理的后续建议（如 ConfigMap 集中管理、连接超时告警）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 4, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "识别出端口篡改（30070→30071）比地址篡改更隐蔽，需要对比实际 service 端口才能发现", "points": 2, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "将三个配置错误错误归因为单一根因", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 environment variables 就下结论", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "遗漏了端口篡改（30070→30071），只修复了地址篡改", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误判为网络问题或 DNS 问题，而实际是环境变量配置错误", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "仅查看单一云的服务配置就给出全局诊断", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误以为无 Chaos Mesh 实验就没有注入故障，停止继续排查环境变量配置", "points": -6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "模型在面对多个同时存在的配置错误时，能够正确识别每个错误的独立性，不将无关故障混为一谈", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型的排查过程覆盖了所有受影响的云（阿里云、腾讯云、AWS），检查了各云上的 deployment 配置", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型能够识别端口篡改（30070→30071）这种比地址篡改更隐蔽的配置错误，需要对比实际 service 端口才能发现", "points": 5, "tags": ["level:cluster", "cluster:port_tampering_detection", "axis:context_awareness"]}, {"criterion": "模型不被'无 chaos 实验'的表象误导，能够正确判断配置类故障的存在并继续排查", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（环境变量值、service 端口、日志错误），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 13 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 16 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 19 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 13}, "prompt": "用户反馈结账和下单都不稳定：有时购物车能正常提交，但下一步就提示服务暂时不可用；有时订单创建失败，有时支付前就中断。"}
{"canary": "resource-throttle", "tags": ["发现", "容器与工作负载", "CPU限流", "OOMKill", "Liveness Probe 误杀", "medium", "multi-cloud"], "case": {"id": "resource-throttle", "title": "资源限制误配排查", "inject_script": "./challenge-07-data-illusion/inject.sh", "recover_script": "./challenge-07-data-illusion/recover.sh", "fault_phenomenon": "用户反馈商品列表加载极慢，有时要等十几秒；推荐内容偶尔突然消失，过一会儿又恢复；前端页面偶尔短暂不可用。之前一直好好的，今天突然开始出问题。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "三个资源配置故障：(1) tencent 上 productcatalogservice CPU 被压到 request=5m/limit=10m，导致严重限流；(2) alicloud 上 recommendationservice memory 被压到 request=32Mi/limit=64Mi，导致 OOMKill 风险；(3) alicloud 上 frontend livenessProbe.failureThreshold 被改为 1，导致一次探针失败即重启。无 Chaos Mesh 实验。", "faulty_clouds": ["unknown"], "affected_clouds": ["aliyun", "tencent"], "fault_location": {"module": "k8s-service", "file_path": null, "function_or_config": "productcatalogservice resources CPU, recommendationservice resources memory, frontend livenessProbe.failureThreshold", "description": "三个 deployment 的资源/探针容错被改为过于激进的值，导致慢、抖、重启并存。"}}, "reasoning_process": {"observed_symptoms": ["商品列表加载明显变慢或超时。", "推荐服务偶发不可用并伴随重启。", "前端偶发短暂不可用后恢复。", "chaos 资源列表为空。"], "key_evidence": [{"source": "kubectl get deployment productcatalogservice -n seat-1 --context=tencent -o yaml", "content": "resources.requests.cpu=5m, resources.limits.cpu=10m。", "conclusion": "CPU 配置过低，导致持续限流。"}, {"source": "kubectl get deployment recommendationservice -n seat-1 --context=alicloud -o yaml", "content": "resources.requests.memory=32Mi, resources.limits.memory=64Mi。", "conclusion": "内存配置过低，存在 OOMKill 风险。"}, {"source": "kubectl get deployment frontend -n seat-1 --context=alicloud -o yaml", "content": "livenessProbe.failureThreshold=1。", "conclusion": "探针容错过低，容易触发频繁重启。"}, {"source": "kubectl get iochaos,networkchaos,podchaos,stresschaos,dnschaos -A", "content": "无相关 chaos 实验对象。", "conclusion": "故障主要是配置层，不是 Chaos Mesh 注入。"}], "causal_chain": ["CPU 限额过低导致 productcatalogservice 请求堆积和超时。", "内存限额过低导致 recommendationservice 偶发 OOM 与重启。", "frontend liveness 容错仅 1 次，瞬时抖动也会触发重启。", "三者叠加形成“慢 + 抖 + 偶发不可用”的复合症状。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=tencent get deployment productcatalogservice -n seat-1 -o jsonpath='{.spec.template.spec.containers[0].resources.requests.cpu} {.spec.template.spec.containers[0].resources.limits.cpu}{\"\\n\"}'", "purpose": "确认 productcatalogservice CPU 被压低。", "expected_result": "5m 10m。"}, {"cmd": "kubectl --context=alicloud get deployment recommendationservice -n seat-1 -o jsonpath='{.spec.template.spec.containers[0].resources.requests.memory} {.spec.template.spec.containers[0].resources.limits.memory}{\"\\n\"}'", "purpose": "确认 recommendationservice 内存被压低。", "expected_result": "32Mi 64Mi。"}, {"cmd": "kubectl --context=alicloud get deployment frontend -n seat-1 -o jsonpath='{.spec.template.spec.containers[0].livenessProbe.failureThreshold}{\"\\n\"}'", "purpose": "确认 frontend 探针阈值异常。", "expected_result": "1。"}], "success_criteria": ["productcatalogservice CPU 恢复为 request=50m, limit=100m。", "recommendationservice memory 恢复为 request=64Mi, limit=128Mi。", "frontend livenessProbe.failureThreshold 恢复为 3。", "相关 deployment rollout 成功，业务恢复稳定。"]}, "proposed_resolution": {"suggestion": "分别恢复三项配置，并等待 rollout 完成后复测业务链路。", "fix_steps": ["恢复 productcatalogservice CPU 到 request=50m/limit=100m。", "恢复 recommendationservice memory 到 request=64Mi/limit=128Mi。", "恢复 frontend livenessProbe.failureThreshold 到 3。", "等待 rollout 完成并验证页面加载与推荐能力恢复。"]}, "confidence": 0.94, "special_points": [{"point": "无需 chaos 线索也能通过 deployment 配置定位故障", "score": 4}, {"point": "同时区分慢查询、OOM 风险和探针抖动三类症状", "score": 3}, {"point": "恢复值必须与注入前原始值一致，避免二次故障", "score": 3}], "tool_process": {"ideal_call_count": 12, "valuable_actions": ["kubectl get deployment -o yaml/jsonpath 核对资源与探针字段", "kubectl get pods/events 观察重启与异常事件", "对照注入前原始值进行复核"]}}, "rubrics": {"positive_points_total": 99, "negative_points_total": -47, "rubrics": [{"criterion": "描述了观察到的故障现象（商品列表加载极慢、推荐内容偶尔消失、前端偶尔不可用）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了各服务日志，确认具体的错误类型（OOMKill、超时、探针失败等）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 kubectl top pod 查看了资源使用情况", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了各服务的 deployment 配置（resources limits/requests、探针参数）", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了各服务的 OOMKill 事件和 pod 重启记录", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "确认无 Chaos Mesh 实验，必须从 deployment 配置入手排查", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "识别出 productcatalogservice 的资源配置被修改", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 productcatalogservice 的具体参数（CPU request=5m/limit=10m）和部署云（腾讯云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 recommendationservice 的资源配置被修改", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 recommendationservice 的具体参数（memory request=32Mi/limit=64Mi）和部署云（阿里云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 frontend 的 livenessProbe 配置被修改", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 frontend livenessProbe 的具体参数（failureThreshold=1）和部署云（阿里云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "区分了 CPU 限流、OOM 风险和探针抖动三类不同症状，而非笼统说'服务异常'", "points": 2, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "正确追踪因果链：CPU limit 过低 → 严重限流 → 接口延迟上升", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：memory limit 过低 → OOMKill → pod 重启", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出了具体可执行的修复步骤：恢复 productcatalogservice CPU 到 request=50m/limit=100m", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：恢复 recommendationservice memory 到 request=64Mi/limit=128Mi", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：恢复 frontend livenessProbe.failureThreshold 到 3", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "确认所有 pod 状态为 Running，无 OOMKill", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了合理的后续建议（如合理配置 resource requests/limits、HPA）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 4, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "识别出无 chaos 线索的配置类故障，不被'没有 chaos 实验'误导认为没有故障", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "将三个资源配置故障错误归因为单一根因", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 deployment 配置就下结论，仅从日志推测原因", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "遗漏了某个服务的资源配置问题，只修复了部分", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误判为 chaos 注入的故障，而实际是配置变更", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "仅查看单一云的资源配置就给出全局诊断", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只查看 top pod 或日志，未检查 deployment 的 resources 与 livenessProbe 配置", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "模型在面对多个服务的资源配置故障时，能够正确识别每个故障的独立性，不将无关故障混为一谈", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型的排查过程覆盖了所有受影响的云（阿里云、腾讯云），检查了各云上的 deployment 配置", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型能够识别无 chaos 线索的配置类故障，不被'没有 chaos 实验'误导认为没有故障", "points": 5, "tags": ["level:cluster", "cluster:no_chaos_diagnosis", "axis:context_awareness"]}, {"criterion": "模型不被'无 chaos 实验就等于无故障'的表象误导，能够正确判断配置类故障的存在并继续排查", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（deployment 配置、资源使用数据、日志错误），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 12 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 15 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 18 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 12}, "prompt": "用户反馈商品列表加载极慢，有时要等十几秒；推荐内容偶尔突然消失，过一会儿又恢复；前端页面偶尔短暂不可用。之前一直好好的，今天突然开始出问题。"}
{"canary": "hybrid-fault", "tags": ["定位", "混沌与题面治理", "网络丢包", "环境变量写错", "medium", "multi-cloud"], "case": {"id": "hybrid-fault", "title": "网络抖动与配置误改", "inject_script": "./challenge-08-latency-maze/inject.sh", "recover_script": "./challenge-08-latency-maze/recover.sh", "fault_phenomenon": "用户反馈商品列表偶尔加载失败，重试后又能恢复；运费计算有时会明显变慢，等好久才有结果。不是每次都这样，时好时坏，刷新几次可能就正常了。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "两个独立故障同时存在：(1) NetworkChaos 对 productcatalogservice 注入 15% 网络丢包，导致商品列表偶尔加载失败；(2) shippingservice 的 DISABLE_PROFILER 环境变量被从 1 改为 0，开启性能分析导致 CPU 开销增加、响应变慢。两个故障必须分别修复。", "faulty_clouds": ["unknown"], "affected_clouds": ["aliyun", "tencent"], "fault_location": {"module": "network, k8s-service", "file_path": null, "function_or_config": "NetworkChaos: hybrid-productcatalog-loss, shippingservice: DISABLE_PROFILER=0", "description": "NetworkChaos 对 productcatalogservice 注入 15% 丢包（腾讯云），shippingservice 的 DISABLE_PROFILER 被改为 0（阿里云）。一个网络层故障 + 一个配置层故障，必须分别排查和修复。"}}, "reasoning_process": {"observed_symptoms": ["商品列表偶尔加载失败，重试后恢复。", "运费计算偶尔响应慢。", "所有 pod 状态为 Running。", "前端偶尔报错，但不是每次都失败。"], "key_evidence": [{"source": "kubectl get networkchaos -n seat-1 --context=tencent", "content": "存在 hybrid-productcatalog-loss 实验，action: loss, loss: 15%。", "conclusion": "productcatalogservice 被注入网络丢包，导致偶尔连接失败。"}, {"source": "kubectl get deployment shippingservice -n seat-1 --context=alicloud -o yaml | grep DISABLE_PROFILER", "content": "DISABLE_PROFILER=0。", "conclusion": "性能分析被开启，增加 CPU 开销，导致响应变慢。"}, {"source": "kubectl get pods -n seat-1 --context=alicloud | grep shipping", "content": "shippingservice CPU 使用偏高。", "conclusion": "profiler 开启导致 CPU 开销增加。"}, {"source": "curl -s http://43.139.187.221:30008/products", "content": "偶尔返回错误或超时。", "conclusion": "productcatalogservice 受丢包影响，偶尔不可用。"}], "causal_chain": ["NetworkChaos 注入 15% 丢包 → productcatalogservice 网络偶尔不通 → 商品列表偶尔加载失败。", "DISABLE_PROFILER 被改为 0 → shippingservice 开启性能分析 → CPU 开销增加 → 运费计算响应变慢。", "两个故障独立存在，互不关联。agent 只删 chaos 实验不够，还得找到配置问题；只改配置也不够，还得删 chaos 实验。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=tencent get networkchaos -n seat-1", "purpose": "查看 NetworkChaos 实验。", "expected_result": "存在 hybrid-productcatalog-loss 实验。"}, {"cmd": "kubectl --context=alicloud get deployment shippingservice -n seat-1 -o yaml | grep DISABLE_PROFILER", "purpose": "检查 shippingservice 的 profiler 配置。", "expected_result": "DISABLE_PROFILER=0（被篡改）。"}, {"cmd": "kubectl --context=alicloud top pod -n seat-1 | grep shipping", "purpose": "查看 shippingservice 的 CPU 使用情况。", "expected_result": "CPU 使用偏高。"}, {"cmd": "curl -s http://43.139.187.221:30008/products", "purpose": "测试 productcatalogservice 可用性。", "expected_result": "偶尔返回错误（受丢包影响）。"}], "success_criteria": ["NetworkChaos 实验被删除。", "shippingservice 的 DISABLE_PROFILER 恢复为 1。", "productcatalogservice 响应恢复正常。", "shippingservice CPU 使用恢复正常。"]}, "proposed_resolution": {"suggestion": "删除 NetworkChaos 实验，恢复 shippingservice 的 DISABLE_PROFILER=1。建议监控服务的 profiler 状态，避免生产环境意外开启。", "fix_steps": ["执行 kubectl --context=tencent delete networkchaos hybrid-productcatalog-loss -n seat-1。", "执行 kubectl --context=alicloud set env deploy/shippingservice DISABLE_PROFILER=1 -n seat-1 --overwrite。", "验证 productcatalogservice 响应恢复正常。", "验证 shippingservice CPU 使用恢复正常。"]}, "confidence": 0.92, "special_points": [{"point": "同时识别出 NetworkChaos 和配置变更两个独立故障", "score": 4}, {"point": "不能只删 chaos 就认为修好了，还得检查配置", "score": 3}, {"point": "通过 top pod 发现 shippingservice CPU 偏高，关联到 profiler 配置", "score": 3}], "tool_process": {"ideal_call_count": 15, "valuable_actions": ["kubectl get networkchaos 查看 NetworkChaos 实验", "kubectl get deployment -o yaml | grep DISABLE_PROFILER", "kubectl top pod 查看 CPU 使用", "curl 测试服务可用性", "分别修复两个独立故障"]}}, "rubrics": {"positive_points_total": 97, "negative_points_total": -41, "rubrics": [{"criterion": "描述了观察到的故障现象（商品列表偶尔加载失败、运费计算变慢）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了各服务日志，确认具体的错误类型（超时、丢包、响应慢等）", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 kubectl get networkchaos 查看了网络混沌实验", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 shippingservice 的 deployment 配置（environment variables）", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 kubectl top pod 查看了资源使用情况", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 productcatalogservice 存在网络丢包问题", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 NetworkChaos 的具体参数（loss 15%）和部署云（腾讯云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 shippingservice 的 DISABLE_PROFILER 环境变量被修改", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 DISABLE_PROFILER 的具体变更（从 1 改为 0，开启性能分析）和部署云（阿里云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解两个故障独立存在：一个是 chaos 注入，一个是配置变更，不能只删 chaos 就认为修好了", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "通过 top pod 发现 shippingservice CPU 偏高，关联到 profiler 配置变更", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "正确追踪因果链：NetworkChaos 丢包 → productcatalogservice 偶尔超时 → 商品列表加载失败", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：profiler 开启 → CPU 开销增加 → shippingservice 响应变慢", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出了具体可执行的修复步骤：删除 NetworkChaos 实验", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：恢复 DISABLE_PROFILER=1", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "确认 NetworkChaos 已删除，DISABLE_PROFILER 已恢复", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了合理的后续建议（如监控 profiler 状态、避免生产环境意外开启）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 4, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "将两个故障错误归因为单一根因", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 NetworkChaos 实验就下结论", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误判为纯网络问题或纯配置问题，而实际是两者混合", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "仅查看单一云的资源就给出全局诊断", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只恢复 shippingservice profiler 配置，遗漏 productcatalogservice 的 NetworkChaos 丢包", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "模型在面对同时存在的 chaos 故障和配置变更时，能够正确识别两者的独立性，不将无关故障混为一谈", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型的排查过程覆盖了所有受影响的云（阿里云、腾讯云），使用了多个 --context 查看不同云的资源", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型能够识别'只删 chaos 不够'的情况，不被'chaos 已删除'的表象误导认为全部修好", "points": 5, "tags": ["level:cluster", "cluster:hybrid_fault_recognition", "axis:context_awareness"]}, {"criterion": "模型不被'chaos 已删除'的表象误导，能够识别出还需要修复配置变更才算完整修复", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（NetworkChaos 实验状态、deployment 配置、top pod 数据），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 12 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 15 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 18 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 12}, "prompt": "用户反馈商品列表偶尔加载失败，重试后又能恢复；运费计算有时会明显变慢，等好久才有结果。不是每次都这样，时好时坏，刷新几次可能就正常了。"}
{"canary": "time-window", "tags": ["发现", "混沌与题面治理", "Pod 被杀死", "网络延迟", "时间窗口故障", "hard", "multi-cloud"], "case": {"id": "time-window", "title": "短时故障与延迟注入", "inject_script": "./challenge-09-time-window/inject.sh", "recover_script": "./challenge-09-time-window/recover.sh", "fault_phenomenon": "运维刚说有不少用户投诉服务不可用，但你到现场时服务又看起来正常了。日志里能看到 30 秒前有一波超时错误，现在已经恢复。有人怀疑刚才发生过短暂故障，但现场已经看不出明显异常。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "两个时间窗口故障：(1) PodChaos 立即杀死 new-orderservice pod，并在约 10 秒后由后台任务自动清理；(2) NetworkChaos 延迟注入，在 T+15 秒对 new-inventoryservice 生效。考生需要识别“短暂已消失故障”与“延迟出现故障”。", "faulty_clouds": ["aws"], "affected_clouds": ["aws"], "fault_location": {"module": "k8s-service, network", "file_path": null, "function_or_config": "PodChaos: time-window-order-kill (短时注入后自动清理), NetworkChaos: time-window-inventory-delay (T+15s 生效)", "description": "PodChaos 先触发一次 pod-kill 后被自动删除；NetworkChaos 在 15 秒后才创建并生效。两个故障处于不同时间窗口。"}}, "reasoning_process": {"observed_symptoms": ["短时间内出现过服务不可用告警，随后恢复。", "到场初查时 pod 可能全部 Running。", "orderservice 存在近期重建或重启痕迹。", "稍等一段时间后会出现 inventoryservice 相关 NetworkChaos。"], "key_evidence": [{"source": "kubectl get pods -n seat-1 --context=aws | grep new-orderservice", "content": "new-orderservice pod 在短时间窗口内发生过重建或重启迹象。", "conclusion": "orderservice 曾被外力中断（与 PodChaos 行为一致）。"}, {"source": "kubectl get events -n seat-1 --context=aws --sort-by='.lastTimestamp' | grep -Ei 'Killing|Back-off|Created'", "content": "近期存在与 new-orderservice 相关的 Killing/重建事件。", "conclusion": "支持“短时 PodChaos 已发生并结束”的判断。"}, {"source": "kubectl get podchaos time-window-order-kill -n seat-1 --context=aws", "content": "在注入后短时间可能可见，随后被自动清理并消失。", "conclusion": "该故障窗口短，容易漏检。"}, {"source": "sleep 15 && kubectl get networkchaos time-window-inventory-delay -n seat-1 --context=aws -o yaml", "content": "约 15 秒后出现 delay=4000ms 的 NetworkChaos。", "conclusion": "存在延迟生效故障，需等待窗口才能确认。"}, {"source": "cat /tmp/time-window-inventory.log", "content": "日志可见 'inventoryservice delay 已生效' 等记录。", "conclusion": "可用本地注入日志辅助确认延迟故障已触发。"}], "causal_chain": ["PodChaos 先杀死 new-orderservice pod → 产生短暂可用性抖动和事件痕迹。", "后台任务在约 10 秒后自动删除 PodChaos → 到场时可能看不到该实验对象。", "另一后台任务在 T+15 秒创建 NetworkChaos → new-inventoryservice 出现高延迟。", "两个故障不在同一时刻可见，排查必须结合时间线和间接证据。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=aws get pods -n seat-1 | grep new-orderservice", "purpose": "查看 orderservice 是否有近期重建/重启痕迹。", "expected_result": "可观察到短时间内异常波动痕迹。"}, {"cmd": "kubectl --context=aws get events -n seat-1 --sort-by='.lastTimestamp' | grep -Ei 'Killing|Back-off|Created'", "purpose": "确认短时 PodChaos 的间接证据。", "expected_result": "存在 orderservice 相关 Killing/重建事件。"}, {"cmd": "sleep 15 && kubectl --context=aws get networkchaos time-window-inventory-delay -n seat-1 -o yaml", "purpose": "验证延迟故障是否按窗口出现。", "expected_result": "出现 time-window-inventory-delay，delay 延迟为 4000ms。"}, {"cmd": "cat /tmp/time-window-inventory.log", "purpose": "查看本地注入日志中的延迟触发记录。", "expected_result": "出现 'inventoryservice delay 已生效'。"}], "success_criteria": ["识别到 new-orderservice 曾发生短时 pod-kill 故障。", "识别到 T+15 秒出现的 inventoryservice 延迟故障。", "修复后 PodChaos 与 NetworkChaos 均被清理。"]}, "proposed_resolution": {"suggestion": "清理 time-window-order-kill 与 time-window-inventory-delay（含 finalizer 兜底）。建议保留事件与日志采集，便于排查短窗口故障。", "fix_steps": ["kubectl --context=aws -n seat-1 patch podchaos time-window-order-kill --type='json' -p='[{\"op\":\"remove\",\"path\":\"/metadata/finalizers\"}]' 2>/dev/null; kubectl --context=aws -n seat-1 delete podchaos time-window-order-kill --ignore-not-found。", "kubectl --context=aws -n seat-1 patch networkchaos time-window-inventory-delay --type='json' -p='[{\"op\":\"remove\",\"path\":\"/metadata/finalizers\"}]' 2>/dev/null; kubectl --context=aws -n seat-1 delete networkchaos time-window-inventory-delay --ignore-not-found。", "确认 aws 集群中 podchaos/networkchaos 两个对象都不存在。", "确认 new-orderservice 与 new-inventoryservice 恢复稳定。"]}, "confidence": 0.93, "special_points": [{"point": "同一题包含“短时自动消失故障 + 延迟出现故障”两个窗口", "score": 4}, {"point": "不能只看当前对象列表，必须结合事件和时间线", "score": 3}, {"point": "可用本地注入日志辅助确认延迟触发是否发生", "score": 3}], "tool_process": {"ideal_call_count": 18, "valuable_actions": ["kubectl get pods 查看 orderservice 变化痕迹", "kubectl get events 还原时间线", "等待 15 秒后检查 networkchaos", "查看 /tmp/time-window-inventory.log 辅助判断"]}}, "rubrics": {"positive_points_total": 103, "negative_points_total": -48, "rubrics": [{"criterion": "描述了观察到的故障现象（短时间内出现过服务不可用告警，到场时已恢复）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了各服务日志，确认具体的错误类型", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 kubectl get podchaos/networkchaos 查看了混沌实验（当前可能已消失）", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 kubectl get events，从事件时间线推断已消失的 chaos 实验", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了本地注入日志（/tmp/time-window-inventory.log），确认延迟触发是否发生", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 orderservice 的 pod 重建时间线，确认与 PodChaos 吻合", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 PodChaos 杀死了 new-orderservice pod", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 PodChaos 的具体行为（pod-kill，约 10 秒后自动清理）和部署云（AWS）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 NetworkChaos 延迟注入影响了 new-inventoryservice", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 NetworkChaos 的具体参数（delay 4000ms，T+15 秒生效）和部署云（AWS）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解'短时自动消失故障 + 延迟出现故障'两个时间窗口的并存", "points": 5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "不能只看当前对象列表，必须结合事件和时间线推断已消失的故障", "points": 5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "正确追踪因果链：PodChaos 杀死 pod → 短暂不可用 → pod 自动重建恢复", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：NetworkChaos 延迟注入 → T+15 秒后开始影响 inventoryservice → 超时", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出了具体可执行的修复步骤：删除 NetworkChaos 实验 time-window-inventory-delay", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "确认 PodChaos time-window-order-kill 已自动清理，无需手动删除", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "确认 podchaos 和 networkchaos 均已被清理", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了合理的后续建议（如保留事件与日志采集，便于排查短窗口故障）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 4, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "可用本地注入日志（/tmp/time-window-inventory.log）辅助确认延迟触发是否发生", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "只看到当前存在的 NetworkChaos 就认为找到了全部根因，遗漏了已消失的 PodChaos", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误认为 orderservice 曾经重启是应用自身问题，而非 PodChaos 导致", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "不理解延迟注入概念，认为 NetworkChaos 当前没生效就不是问题", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 events 就下结论，遗漏了已消失的 chaos 实验", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "将两个时间窗口故障错误归因为单一根因", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "没有等待 T+15 秒延迟注入窗口或检查 /tmp 日志，过早判定 NetworkChaos 不存在", "points": -4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "模型在面对'短时自动消失'和'延迟出现'两个时间窗口故障时，能够正确识别两者的独立性", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型能够识别已自动清理的 PodChaos 故障，不被'当前没有 chaos 实验'的表象误导", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（events 时间线、日志、注入日志），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "模型能够理解延迟注入、短时自动消失等时间相关特征，正确判断故障的时间窗口", "points": 5, "tags": ["level:cluster", "cluster:time_sensitivity", "axis:context_awareness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 12 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 15 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 18 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 12}, "prompt": "运维刚说有不少用户投诉服务不可用，但你到现场时服务又看起来正常了。日志里能看到 30 秒前有一波超时错误，现在已经恢复。有人怀疑刚才发生过短暂故障，但现场已经看不出明显异常。"}
{"canary": "ultimate-judgment", "tags": ["定位", "混沌与题面治理", "I/O 延迟", "网络丢包", "环境变量写错", "烟雾弹", "hard", "multi-cloud"], "case": {"id": "ultimate-judgment", "title": "多线索综合故障排查", "inject_script": "./challenge-10-ultimate-judgment/inject.sh", "recover_script": "./challenge-10-ultimate-judgment/recover.sh", "fault_phenomenon": "前端页面时不时打不开商品列表，结账时也偶尔会提示服务暂时不可用。现场看起来服务都还在，但日志里有零星超时。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "两个真实 chaos 故障 + 一个配置故障：(1) 腾讯云对 redis-cart 注入 IOChaos latency（3000ms，80%），(2) AWS 对 paymentservice 注入 NetworkChaos 丢包 8%，(3) 阿里云 checkoutservice 的 CURRENCY_SERVICE_ADDR 被改为错误地址。前端若干跨云地址是烟雾弹，表面可疑但实际正确。", "faulty_clouds": ["unknown"], "affected_clouds": ["aliyun", "tencent", "aws"], "fault_location": {"module": "io, network, k8s-service", "file_path": null, "function_or_config": "IOChaos: ultimate-redis-io, NetworkChaos: ultimate-payment-loss, checkoutservice: CURRENCY_SERVICE_ADDR=currency.wrong.svc.cluster.local:7000", "description": "redis-cart I/O 延迟会拖慢购物车链路，paymentservice 丢包导致支付链路偶发超时，checkoutservice 货币服务地址错误导致结账失败。三类故障并存。"}}, "reasoning_process": {"observed_symptoms": ["商品/购物车相关链路偶发超时，结账不稳定。", "支付阶段偶尔失败或超时。", "pod 多数保持 Running，但业务偶发失败。", "frontend 配置里有跨云公网地址，看起来可疑但并非根因。"], "key_evidence": [{"source": "kubectl get iochaos ultimate-redis-io -n seat-1 --context=tencent -o yaml", "content": "action=latency, delay=3000ms, percent=80, selector app=redis-cart。", "conclusion": "redis-cart 被注入 I/O 延迟，是真实故障。"}, {"source": "kubectl get networkchaos ultimate-payment-loss -n seat-1 --context=aws -o yaml", "content": "action=loss, loss=8, selector app=paymentservice。", "conclusion": "paymentservice 存在 8% 丢包，导致支付链路偶发超时。"}, {"source": "kubectl get deployment checkoutservice -n seat-1 --context=alicloud -o yaml | grep CURRENCY_SERVICE_ADDR", "content": "CURRENCY_SERVICE_ADDR=currency.wrong.svc.cluster.local:7000。", "conclusion": "checkoutservice 配置被改错，直接影响结账流程。"}, {"source": "kubectl get deployment frontend -n seat-1 --context=alicloud -o yaml | grep ADDR", "content": "frontend 的跨云地址仍是预期值（烟雾弹）。", "conclusion": "frontend 地址看似可疑但不是根因。"}], "causal_chain": ["IOChaos 注入 redis-cart I/O 延迟 → 购物车链路变慢或超时。", "NetworkChaos 对 paymentservice 丢包 8% → 支付请求偶发失败。", "checkoutservice CURRENCY_SERVICE_ADDR 错误 → 货币转换调用失败，结账失败率上升。", "三类故障同时存在，必须分别识别与修复，不能只修 chaos 或只修配置。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=tencent get iochaos ultimate-redis-io -n seat-1 -o yaml", "purpose": "确认 IOChaos 注入对象与参数。", "expected_result": "action=latency, delay=3000ms, percent=80。"}, {"cmd": "kubectl --context=aws get networkchaos ultimate-payment-loss -n seat-1 -o yaml", "purpose": "确认 NetworkChaos 注入对象与参数。", "expected_result": "action=loss, loss=8。"}, {"cmd": "kubectl --context=alicloud get deployment checkoutservice -n seat-1 -o yaml | grep CURRENCY_SERVICE_ADDR", "purpose": "确认 checkoutservice 配置是否被篡改。", "expected_result": "出现错误地址 currency.wrong.svc.cluster.local:7000。"}, {"cmd": "kubectl --context=alicloud get deployment frontend -n seat-1 -o yaml | grep ADDR", "purpose": "识别烟雾弹配置。", "expected_result": "frontend 跨云地址为正常值，不应误判。"}], "success_criteria": ["识别并删除 ultimate-redis-io。", "识别并删除 ultimate-payment-loss。", "恢复 checkoutservice 的 CURRENCY_SERVICE_ADDR=43.139.187.221:30009。", "业务端商品/结账链路恢复稳定。"]}, "proposed_resolution": {"suggestion": "统一清理 chaos 并恢复 checkoutservice 环境变量（含 finalizer 兜底）。", "fix_steps": ["kubectl --context=tencent -n seat-1 patch iochaos ultimate-redis-io --type='json' -p='[{\"op\":\"remove\",\"path\":\"/metadata/finalizers\"}]' 2>/dev/null; kubectl --context=tencent -n seat-1 delete iochaos ultimate-redis-io --ignore-not-found。", "kubectl --context=aws -n seat-1 patch networkchaos ultimate-payment-loss --type='json' -p='[{\"op\":\"remove\",\"path\":\"/metadata/finalizers\"}]' 2>/dev/null; kubectl --context=aws -n seat-1 delete networkchaos ultimate-payment-loss --ignore-not-found。", "kubectl --context=alicloud -n seat-1 set env deploy/checkoutservice CURRENCY_SERVICE_ADDR=43.139.187.221:30009，恢复正确的货币服务地址。", "确认 tencent 无 iochaos ultimate-redis-io（并兼容清理旧对象 ultimate-productcatalog-io）。", "确认 aws 无 networkchaos ultimate-payment-loss。", "确认 checkoutservice 的 CURRENCY_SERVICE_ADDR 已恢复为 43.139.187.221:30009。"]}, "confidence": 0.94, "special_points": [{"point": "三类故障并存（IO + Network + Config），必须并行排查", "score": 4}, {"point": "frontend 跨云地址是烟雾弹，不能误判", "score": 3}, {"point": "清理 chaos 对象时需处理 finalizer 与历史遗留对象", "score": 3}], "tool_process": {"ideal_call_count": 18, "valuable_actions": ["kubectl get iochaos/networkchaos 精确核对故障对象", "kubectl get deployment -o yaml 核对关键 env", "修复后再次核对对象残留", "区分真实故障与烟雾弹配置"]}}, "rubrics": {"positive_points_total": 99, "negative_points_total": -41, "rubrics": [{"criterion": "描述了观察到的故障现象（商品列表打不开、结账偶尔不可用、日志里有零星超时）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了各服务日志，确认具体的错误类型", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 kubectl get iochaos/networkchaos 查看了三云的混沌实验", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了各服务的 environment variables，发现配置错误", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 frontend 的跨云地址配置，判断其为烟雾弹而非真实故障", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "识别出 redis-cart 存在 I/O 层延迟问题", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 IOChaos 的具体参数（delay 3000ms, percent 80%）和部署云（腾讯云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 paymentservice 存在网络丢包问题", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 NetworkChaos 的具体参数（loss 8%）和部署云（AWS）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 checkoutservice 的 CURRENCY_SERVICE_ADDR 被改为错误地址", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 CURRENCY_SERVICE_ADDR 的错误值和部署云（阿里云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：IOChaos → redis-cart 慢 → cartservice 超时 → checkoutservice 500", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：NetworkChaos → paymentservice 偶尔超时 → 支付失败", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：CURRENCY_SERVICE_ADDR 错误 → 货币转换服务连接失败 → 结账失败", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出了具体可执行的修复步骤：删除 IOChaos ultimate-redis-io", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：删除 NetworkChaos ultimate-payment-loss", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复步骤：恢复 checkoutservice 的 CURRENCY_SERVICE_ADDR", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "确认所有 chaos 实验已删除，环境变量已恢复", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 4, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "理解三类故障并存（IO + Network + Config），必须并行排查，不能只修一种就认为全部修好", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "理解清理 chaos 对象时需要处理 finalizer 与历史遗留对象（如 ultimate-productcatalog-io）", "points": 2, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "将三类故障错误归因为单一根因", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "误判前端跨云地址为故障，修改了正确的配置", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 chaos 实验就下结论", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "仅查看单一云的资源就给出全局诊断", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只检查 frontend 的跨云地址烟雾弹，未继续检查 checkoutservice 的真实配置错误", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "模型在面对三类不同故障（IO + Network + Config）并存时，能够正确识别每个故障的独立性", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型的排查过程覆盖了所有受影响的云（阿里云、腾讯云、AWS），使用了多个 --context", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型能够识别前端跨云地址是烟雾弹，不被'地址看起来可疑'的表象误导", "points": 5, "tags": ["level:cluster", "cluster:smoke_screen_resistance", "axis:context_awareness"]}, {"criterion": "模型不被烟雾弹配置或已删除的 chaos 实验误导，能够区分真实故障和干扰信息", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（chaos 实验状态、环境变量值、日志错误），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 15 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 18 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 22 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 15}, "prompt": "前端页面时不时打不开商品列表，结账时也偶尔会提示服务暂时不可用。现场看起来服务都还在，但日志里有零星超时。"}
{"canary": "memory-limit-oomkill", "tags": ["发现", "容器与工作负载", "OOMKill", "网络丢包", "Pod 被杀死", "hard", "multi-cloud"], "case": {"id": "memory-limit-oomkill", "title": "内存上限误配导致 OOMKill", "inject_script": "./challenge-11-memory-limit-oomkill/inject.sh", "recover_script": "./challenge-11-memory-limit-oomkill/recover.sh", "fault_phenomenon": "用户反馈推荐区内容时有时无，商品页偶尔报 500，结账也有几次挂了。不是每次都失败，大概 10 次里有 3-4 次出问题，刷新一下可能就好了，但过一阵又犯。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "三个独立故障叠加：(1) 阿里云 seat-1 命名空间中的 recommendationservice 被错误下调 memory limit（从128Mi降到64Mi），在真实请求压力下触发 OOMKill；(2) 腾讯云对 productcatalogservice 注入 NetworkChaos 20% 丢包，导致商品列表偶尔加载失败；(3) 阿里云对 checkoutservice 注入 PodChaos pod-kill（30 秒后自动消失），导致 checkoutservice 短暂中断。", "faulty_clouds": ["unknown"], "affected_clouds": ["aliyun", "tencent"], "fault_location": {"module": "k8s-workload-resource, network, k8s-service", "file_path": null, "function_or_config": "deployment/recommendationservice resources.limits.memory=64Mi; NetworkChaos: challenge-11-productcatalog-loss (loss 20%); PodChaos: challenge-11-checkout-kill (pod-kill, 30s duration)", "description": "内存 limit 被压低导致 OOMKill（阿里云），NetworkChaos 注入丢包导致商品页间歇不可用（腾讯云），PodChaos 杀死 checkoutservice pod 后自动消失（阿里云）。三个故障独立存在，必须分别修复。"}}, "reasoning_process": {"observed_symptoms": ["推荐内容时有时无，商品页偶发 500。", "结账也有几次挂了，大概 10 次里有 3-4 次出问题。", "recommendationservice Pod 经常重启，但并非持续 CrashLoop。", "checkoutservice 有重启痕迹，但看不到对应的 chaos 对象。"], "key_evidence": [{"source": "kubectl --context=alicloud -n seat-1 get pods -l app=recommendationservice", "content": "RESTARTS 持续增加。", "conclusion": "服务存在进程级不稳定。"}, {"source": "kubectl --context=alicloud -n seat-1 describe pod <recommendation-pod>", "content": "Last State/Reason 出现 OOMKilled。", "conclusion": "根因为内存不足触发 OOM。"}, {"source": "kubectl --context=alicloud -n seat-1 get deploy recommendationservice -o yaml", "content": "resources.limits.memory 被下调到 64Mi（原值 128Mi）。", "conclusion": "确认内存配置误改位置。"}, {"source": "kubectl --context=tencent -n seat-1 get networkchaos", "content": "存在 challenge-11-productcatalog-loss 实验，action: loss, loss: 20%。", "conclusion": "productcatalogservice 被注入网络丢包，商品页间歇不可用不是内存问题导致的。"}, {"source": "kubectl --context=alicloud -n seat-1 get events --sort-by='.lastTimestamp' | grep checkoutservice", "content": "近期存在 checkoutservice 的 Killing/Pulled/Created 事件。", "conclusion": "checkoutservice pod 曾被外力杀死，但 PodChaos 对象已自动消失，只能从 events 推断。"}], "causal_chain": ["内存 limit/request 下调过低 -> 容器在峰值请求下超内存 -> OOMKilled -> Pod 重启与短时不可用 -> 推荐内容间歇丢失。", "NetworkChaos 注入 20% 丢包 -> productcatalogservice 网络偶尔不通 -> 商品列表偶尔加载失败 -> 前端商品页偶发 500。", "PodChaos 杀死 checkoutservice pod（阿里云） -> checkoutservice 短时不可用 -> 结账偶发失败 -> 30 秒后 PodChaos 自动消失，无法直接看到对象。", "三个故障独立存在，只修内存不能解决商品页和结账问题，只删 NetworkChaos 不能解决推荐和结账问题。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=alicloud -n seat-1 get pods -l app=recommendationservice", "purpose": "观察重启计数与状态。", "expected_result": "存在重启上涨。"}, {"cmd": "kubectl --context=alicloud -n seat-1 describe pod <pod-name>", "purpose": "确认是否 OOMKilled。", "expected_result": "Events/Last State 出现 OOMKilled。"}, {"cmd": "kubectl --context=alicloud -n seat-1 get deploy recommendationservice -o yaml | grep -A6 -n \"resources\"", "purpose": "确认内存资源配置异常。", "expected_result": "limits.memory 为 64Mi（低于 requests 或与之持平）。"}, {"cmd": "kubectl --context=tencent -n seat-1 get networkchaos", "purpose": "查看腾讯云 NetworkChaos 实验。", "expected_result": "存在 challenge-11-productcatalog-loss 实验。"}, {"cmd": "kubectl --context=alicloud -n seat-1 get events --sort-by='.lastTimestamp' | grep -Ei 'checkoutservice.*(Killing|Back-off|Created)'", "purpose": "从 events 推断 checkoutservice 是否被杀死过。", "expected_result": "存在近期 Killing/重建事件，但无对应 PodChaos 对象。"}], "success_criteria": ["recommendationservice 回滚到注入前内存配置。", "NetworkChaos 实验被删除。", "Pod 重启趋势停止继续快速增长。", "商品页和推荐链路均恢复稳定。"]}, "proposed_resolution": {"suggestion": "三个故障必须分别修复：(1) 恢复 recommendationservice 内存配置；(2) 删除 NetworkChaos 实验；(3) 确认 PodChaos 已自动消失（如仍存在则删除）。只修一个不能完全恢复。", "fix_steps": ["kubectl --context=alicloud -n seat-1 set resources deploy/recommendationservice --limits=memory=128Mi --requests=memory=64Mi，等待 rollout 完成。", "kubectl --context=tencent -n seat-1 delete networkchaos challenge-11-productcatalog-loss。", "kubectl --context=alicloud -n seat-1 get podchaos -n seat-1 确认 challenge-11-checkout-kill 已自动消失，如仍存在则 delete。", "确认 recommendationservice 重启趋势停止，Pod 状态稳定。", "验证推荐内容、商品页面和结账功能均恢复正常。"]}, "confidence": 0.93, "special_points": [{"point": "三个独立故障叠加，必须分别识别和修复", "score": 4}, {"point": "PodChaos 30 秒后自动消失，到场时已看不到对象，必须从 events 推断", "score": 4}, {"point": "不能只看到 OOMKill 就认为找到了全部根因，商品页 500 和结账失败有不同来源", "score": 4}, {"point": "NetworkChaos 影响的是 productcatalogservice（商品页，腾讯云），PodChaos 影响的是 checkoutservice（结账，阿里云），症状来源不同", "score": 3}], "tool_process": {"ideal_call_count": 14, "valuable_actions": ["对比 Pod RESTARTS 趋势", "describe pod 核对 OOMKilled 证据", "检查 deployment resources 字段", "kubectl get networkchaos 发现丢包实验", "kubectl get events 还原 checkoutservice 被杀时间线", "恢复后再次核对稳定性"]}}, "rubrics": {"positive_points_total": 103, "negative_points_total": -51, "rubrics": [{"criterion": "描述了观察到的故障现象（推荐内容时有时无、商品页偶发500、结账偶发失败）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了阿里云 recommendationservice 的 pod 重启次数，发现持续增加", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 describe pod 确认 recommendationservice 最近一次退出原因为 OOMKilled", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "检查了 recommendationservice 的 deployment 资源配置，发现 memory limit 被下调到 64Mi", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "检查了腾讯云的 NetworkChaos 实验，发现 productcatalogservice 被注入 20% 丢包", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 kubectl get events 还原 checkoutservice 被杀时间线（PodChaos 对象已自动消失）", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "正确识别出三个独立故障叠加，而非将所有症状归为单一根因", "points": 7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "准确给出 recommendationservice memory limit 的具体值（64Mi）和所在云（阿里云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 NetworkChaos 的具体参数（loss 20%）和目标服务（productcatalogservice，腾讯云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 checkoutservice 故障是由 PodChaos pod-kill 导致的短暂中断，且 PodChaos 30秒后自动消失", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "正确追踪因果链：内存 limit 下调 → OOMKill → Pod 重启 → 推荐内容间歇丢失", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：NetworkChaos 20% 丢包 → productcatalogservice 网络偶尔不通 → 商品页偶发500", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：PodChaos 杀死 checkoutservice pod → 结账偶发失败 → 30秒后自动恢复", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出了修复步骤：恢复 recommendationservice 内存配置", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复步骤：删除 NetworkChaos challenge-11-productcatalog-loss", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "确认 PodChaos 已自动消失（如仍存在则删除）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "验证修复后 recommendationservice 重启趋势停止", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "验证修复后商品页面恢复正常", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "验证修复后结账功能恢复正常", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "在排查过程中使用了合理的命令和工具（kubectl describe、get chaos、get events 等），而非瞎猜", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "模型在面对多个同时存在的故障时，能够正确识别每个故障的独立性，不将无关故障混为一谈", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型的排查过程覆盖了所有受影响的云（阿里云、腾讯云），检查了各云上的 chaos 实验和 deployment 配置", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型能够识别 PodChaos 30秒自动消失的特征，不被'看不到 chaos 对象'的表象误导，从 events 推断出真实原因", "points": 5, "tags": ["level:cluster", "cluster:ephemeral_chaos_detection", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（pod 状态、describe 输出、chaos 对象、events），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "模型能够识别 recommendationservice 当前 Running 且无 OOM 事件的表象，不因'暂时没崩'就忽略 memory limit 被压低的隐患，理解配置不改就随时可能再次 OOM", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "将三个故障错误归因为单一根因（如只说内存问题或只说网络问题）", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 chaos 实验（networkchaos/podchaos）就下结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误判 NetworkChaos 导致的丢包为网络基础设施问题，而非混沌工程注入", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "忽略 PodChaos 30秒自动消失的特征，试图查找不存在的 PodChaos 对象作为证据", "points": -5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "仅查看单一云的资源就给出全局诊断，遗漏了跨云故障", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误判 OOMKill 原因为应用内存泄漏而非内存配置误改", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "建议修改应用代码（如增加内存池、优化缓存）作为即时修复手段，而非恢复配置", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 16 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 20 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 24 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 16}, "prompt": "用户反馈推荐区内容时有时无，商品页偶尔报 500，结账也有几次挂了。不是每次都失败，大概 10 次里有 3-4 次出问题，刷新一下可能就好了，但过一阵又犯。"}
{"canary": "cpu-throttling", "tags": ["发现", "容器与工作负载", "CPU限流", "环境变量写错", "medium", "multi-cloud"], "case": {"id": "cpu-throttling", "title": "CPU 上限过低导致限流", "inject_script": "./challenge-12-cpu-throttling/inject.sh", "recover_script": "./challenge-12-cpu-throttling/recover.sh", "fault_phenomenon": "用户反馈货币换算和结账响应明显变慢，有时候运费也算不出来。偶发请求超时后重试又成功了，感觉像变慢了而不是完全不能用。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "两个独立故障叠加：(1) 腾讯云 seat-1 命名空间中的 currencyservice 被误设为过低的 CPU limit/request（25m），导致在业务调用时发生严重节流，接口延迟显著上升；(2) 阿里云 shippingservice 的 DISABLE_PROFILER 环境变量被从 1 改为 0，开启性能分析导致 CPU 开销增加、运费计算响应变慢。", "faulty_clouds": ["unknown"], "affected_clouds": ["tencent", "aliyun"], "fault_location": {"module": "k8s-workload-resource, k8s-service", "file_path": null, "function_or_config": "deployment/currencyservice resources.limits.cpu=25m; deployment/shippingservice env DISABLE_PROFILER=0", "description": "CPU 配置过低触发 CFS 节流（腾讯云），profiler 开启导致运费计算变慢（阿里云）。两个故障独立存在，必须分别修复。"}}, "reasoning_process": {"observed_symptoms": ["货币换算和结账响应变慢。", "运费计算也变慢了。", "Pod 仍是 Running，重启不明显。", "重试后偶尔成功，更像性能瓶颈而不是完全故障。"], "key_evidence": [{"source": "kubectl --context=tencent -n seat-1 get deploy currencyservice -o yaml", "content": "resources.limits.cpu 与 requests.cpu 被压到 25m。", "conclusion": "CPU 配置异常，导致货币转换服务被节流。"}, {"source": "kubectl --context=alicloud -n seat-1 get deploy shippingservice -o yaml | grep DISABLE_PROFILER", "content": "DISABLE_PROFILER=0。", "conclusion": "性能分析被开启，运费计算 CPU 开销增加。"}, {"source": "kubectl --context=tencent -n seat-1 top pod -l app=currencyservice", "content": "CPU 使用不一定很高，但延迟/吞吐明显变差。", "conclusion": "符合节流而非 OOM。"}, {"source": "kubectl --context=alicloud -n seat-1 top pod -n seat-1 | grep shipping", "content": "shippingservice CPU 使用偏高。", "conclusion": "profiler 开启导致 CPU 开销增加。"}], "causal_chain": ["CPU limit 过低 -> CFS 节流 -> 货币转换请求处理变慢 -> 结账响应变慢。", "DISABLE_PROFILER 被改为 0 -> shippingservice 开启性能分析 -> CPU 开销增加 -> 运费计算响应变慢。", "两个故障独立存在，只修 CPU 不能解决运费慢，只改 profiler 不能解决货币转换慢。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=tencent -n seat-1 get deploy currencyservice -o yaml | grep -A6 -n \"resources\"", "purpose": "确认 CPU 配置被压低。", "expected_result": "cpu=25m。"}, {"cmd": "kubectl --context=tencent -n seat-1 get pods -l app=currencyservice -o wide", "purpose": "确认 Pod 仍在运行但性能不佳。", "expected_result": "Running，无明显重启风暴。"}, {"cmd": "kubectl --context=alicloud -n seat-1 get deploy shippingservice -o yaml | grep DISABLE_PROFILER", "purpose": "确认 shippingservice profiler 配置。", "expected_result": "DISABLE_PROFILER=0（被篡改）。"}, {"cmd": "kubectl --context=alicloud -n seat-1 top pod | grep shipping", "purpose": "查看 shippingservice CPU 使用情况。", "expected_result": "CPU 使用偏高。"}], "success_criteria": ["currencyservice 恢复到注入前资源配置。", "shippingservice 的 DISABLE_PROFILER 恢复为 1。", "接口延迟回落，超时减少。", "运费计算和货币转换均恢复正常。"]}, "proposed_resolution": {"suggestion": "两个故障必须分别修复：(1) 恢复 currencyservice 的 CPU 资源配置；(2) 恢复 shippingservice 的 DISABLE_PROFILER=1。", "fix_steps": ["kubectl --context=tencent -n seat-1 set resources deploy/currencyservice --limits=cpu=100m --requests=cpu=50m，等待 rollout 完成。", "kubectl --context=alicloud -n seat-1 set env deploy/shippingservice DISABLE_PROFILER=1，等待 rollout 完成。", "确认 currencyservice 和 shippingservice Pod 状态正常。", "复核货币转换和运费计算的响应延迟恢复正常。"]}, "confidence": 0.93, "special_points": [{"point": "两个独立故障叠加，必须分别识别和修复", "score": 4}, {"point": "CPU 限流症状和 profiler 导致的性能退化症状容易混淆，需要分别定位到不同服务", "score": 4}, {"point": "核心证据在 deployment 的 resources 字段和 env 字段，不在应用日志", "score": 3}], "tool_process": {"ideal_call_count": 12, "valuable_actions": ["核对 currencyservice resources 字段", "核对 shippingservice env 字段", "观察 top 与延迟表现", "恢复后复核吞吐/超时变化"]}}, "rubrics": {"positive_points_total": 106, "negative_points_total": -47, "rubrics": [{"criterion": "描述了观察到的故障现象（货币换算和结账响应变慢、运费计算变慢、Pod 无大面积重启）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 currencyservice 的 pod 状态，发现性能不佳但无明显重启风暴", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 currencyservice 的 deployment resources 字段，发现 CPU limit 被压低到 25m", "points": 6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "通过 kubectl top pod 观察 currencyservice CPU 使用情况", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 shippingservice 的 deployment env 字段，发现 DISABLE_PROFILER 被改为 0", "points": 6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "通过 kubectl top pod 观察 shippingservice CPU 使用偏高", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "正确识别出两个独立故障叠加，而非将所有症状归为单一根因", "points": 8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "准确给出 currencyservice CPU limit/request 的具体值（25m）和所在云（腾讯云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 shippingservice DISABLE_PROFILER 的具体值（0）和所在云（阿里云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：CPU limit 过低 → CFS 节流 → 货币转换请求处理变慢 → 结账响应变慢", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：DISABLE_PROFILER=0 → 性能分析开启 → CPU 开销增加 → 运费计算响应变慢", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解两个故障独立存在，只修 CPU limit 不能解决运费慢，只改 profiler 不能解决货币转换慢", "points": 5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出了修复步骤：恢复 currencyservice 的资源配置", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复步骤：恢复 shippingservice 的 DISABLE_PROFILER=1", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "在排查过程中使用了合理的命令和工具（kubectl get deploy、top pod、get env 等），而非瞎猜", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "模型在面对多个同时存在的故障时，能够正确识别每个故障的独立性，不将无关故障混为一谈", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型的排查过程覆盖了所有受影响的云（腾讯云、阿里云），检查了各云上的 deployment 配置", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型能够区分 CPU throttling（资源限制导致的节流）和 profiler 开启（额外 CPU 开销）两种不同的性能退化模式", "points": 5, "tags": ["level:cluster", "cluster:resource_throttling_diagnosis", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（resources 字段、env 字段、top pod 输出），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "模型能够识别 CPU throttling 的隐蔽性，不被'Pod Running 且 CPU 不高'的表象误导", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "将两个故障错误归因为单一根因（如只说 CPU 问题或只说配置问题）", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 deployment resources/env 字段就下结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误判为网络问题或 DNS 问题，而实际是资源配置和环境变量问题", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "仅查看单一云的服务配置就给出全局诊断", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误判 CPU throttling 症状为 OOM 或应用内存泄漏", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "忽略 profiler 开启（DISABLE_PROFILER=0）对运费计算性能的影响", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "建议修改应用代码（如优化查询、加重试）作为即时修复手段，而非恢复资源配置", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 12 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 15 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 18 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 12}, "prompt": "用户反馈货币换算和结账响应明显变慢，有时候运费也算不出来。偶发请求超时后重试又成功了，感觉像变慢了而不是完全不能用。"}
{"canary": "request-too-high-pending", "tags": ["发现", "容器与工作负载", "Pod Pending", "网络丢包", "环境变量写错", "hard", "multi-cloud"], "case": {"id": "request-too-high-pending", "title": "资源请求过高导致 Pod Pending", "inject_script": "./challenge-13-request-too-high-pending/inject.sh", "recover_script": "./challenge-13-request-too-high-pending/recover.sh", "fault_phenomenon": "运维反馈刚做完一次发布后，inventory 相关链路偶发失败。Deployment 副本数看着正常，但新 Pod 一直起不来。旧 Pod 还能顶一阵，所以不是全挂，就是时好时坏。运费计算也偶尔出问题，不是每次都失败。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "两个独立故障叠加：(1) AWS seat-1 的 new-inventoryservice 被设置了超出节点能力的 requests/limits（cpu=8, memory=16Gi），新副本无法调度，出现 Pod Pending，造成容量不足和间歇故障；(2) 腾讯云对 shippingservice 注入 NetworkChaos 20% 丢包，导致运费计算偶尔失败。阿里云 checkoutservice 的 CART_SERVICE_ADDR 端口从 30007 改为 30009（烟雾弹），地址看起来正确但端口指向了错误的服务，但当前故障主因是 Pending 和 NetworkChaos。", "faulty_clouds": ["unknown"], "affected_clouds": ["aws", "tencent", "aliyun"], "fault_location": {"module": "k8s-scheduling, network, k8s-service", "file_path": null, "function_or_config": "deployment/new-inventoryservice resources.requests.cpu=8, memory=16Gi; NetworkChaos: challenge-13-shipping-loss (loss 20%); deployment/checkoutservice CART_SERVICE_ADDR=43.139.187.221:30009 (端口篡改 30007→30009)", "description": "资源请求超过节点可分配能力，调度器无法放置新 Pod，触发 Pending（AWS）。NetworkChaos 注入 shippingservice 20% 丢包，运费计算偶尔失败（腾讯云）。checkoutservice 的 CART_SERVICE_ADDR 端口从 30007（cartservice）改为 30009（currencyservice），地址看起来完全正确，但端口指向了错误的服务，需要对比 svc 端口才能发现（阿里云，烟雾弹）。三个配置分布在三朵云上，必须分别检查。"}}, "reasoning_process": {"observed_symptoms": ["Deployment 副本期望值正常，但可用副本不足。", "部分请求失败，故障呈间歇性。", "新 Pod 长时间 Pending。", "运费计算偶尔出问题。", "有人发现 checkoutservice 地址配置看着奇怪。"], "key_evidence": [{"source": "kubectl --context=aws -n seat-1 get pods -l app=new-inventoryservice", "content": "出现 Pending 状态的新 Pod。", "conclusion": "发布后新副本不可用，这是当前故障的直接原因。"}, {"source": "kubectl --context=aws -n seat-1 describe pod <pending-pod>", "content": "Events 出现 Insufficient cpu/memory。", "conclusion": "直接指向调度资源不足。"}, {"source": "kubectl --context=aws -n seat-1 get deploy new-inventoryservice -o yaml", "content": "requests/limits 被异常抬高（cpu=8, memory=16Gi）。", "conclusion": "配置误改是根因。"}, {"source": "kubectl --context=tencent -n seat-1 get networkchaos", "content": "存在 challenge-13-shipping-loss 实验，action: loss, loss: 20%。", "conclusion": "shippingservice 被注入网络丢包，运费计算偶尔失败不是 Pending 导致的。"}, {"source": "kubectl --context=alicloud -n seat-1 get deploy checkoutservice -o yaml | grep CART_SERVICE_ADDR", "content": "CART_SERVICE_ADDR=43.139.187.221:30009，地址正确但端口从 30007 改为 30009。", "conclusion": "端口篡改导致结账失败（30009 是 currencyservice 不是 cartservice），但这是烟雾弹——当前故障主因是 Pending 和 NetworkChaos。需要对比实际 service 端口才能发现差异。"}], "causal_chain": ["requests 过高 -> 新 Pod 调度失败 Pending -> 可用副本减少 -> 峰值流量下部分请求失败。", "NetworkChaos 注入 20% 丢包 -> shippingservice 网络偶尔不通 -> 运费计算偶尔失败。", "checkoutservice 端口篡改（30007→30009）是独立的配置问题，地址看起来完全正确，需要对比 svc 端口才能发现。不应与 Pending 或丢包混淆。", "两个真实故障独立存在，只修资源请求不能解决运费慢，只删 NetworkChaos 不能解决 Pending。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=aws -n seat-1 get pods -l app=new-inventoryservice", "purpose": "确认是否存在 Pending。", "expected_result": "至少一个 Pending Pod。"}, {"cmd": "kubectl --context=aws -n seat-1 describe pod <pending-pod>", "purpose": "确认调度失败原因。", "expected_result": "Insufficient cpu/memory。"}, {"cmd": "kubectl --context=aws -n seat-1 get deploy new-inventoryservice -o yaml | grep -A8 -n \"resources\"", "purpose": "确认资源请求配置异常。", "expected_result": "requests 明显过高（cpu=8, memory=16Gi）。"}, {"cmd": "kubectl --context=tencent -n seat-1 get networkchaos", "purpose": "查看腾讯云 NetworkChaos 实验。", "expected_result": "存在 challenge-13-shipping-loss 实验。"}, {"cmd": "kubectl --context=alicloud -n seat-1 get deploy checkoutservice -o yaml | grep CART_SERVICE_ADDR", "purpose": "确认 checkoutservice 地址配置。", "expected_result": "CART_SERVICE_ADDR=43.139.187.221:30009（端口篡改）。"}], "success_criteria": ["new-inventoryservice 回滚到注入前资源配置。", "Pending Pod 消失并恢复 Running。", "NetworkChaos 实验被删除。", "可用副本恢复，链路稳定。", "运费计算恢复正常。"]}, "proposed_resolution": {"suggestion": "两个真实故障必须分别修复：(1) 回滚 new-inventoryservice 的资源配置；(2) 删除 NetworkChaos 实验。checkoutservice 的地址配置虽然错误，但当前优先修复主因。只修一个不能完全恢复。", "fix_steps": ["kubectl --context=aws -n seat-1 set resources deploy/new-inventoryservice --requests=cpu=50m,memory=64Mi --limits=cpu=100m,memory=128Mi，等待 rollout 完成，确认 Pending Pod 消失。", "kubectl --context=tencent -n seat-1 delete networkchaos challenge-13-shipping-loss。", "kubectl --context=alicloud -n seat-1 set env deploy/checkoutservice CART_SERVICE_ADDR=43.139.187.221:30007，修正烟雾弹端口篡改。", "复核可用副本和业务链路。", "验证运费计算恢复正常。"]}, "confidence": 0.95, "special_points": [{"point": "核心证据在 describe events（Insufficient cpu/networkchaos），不在应用日志", "score": 4}, {"point": "是调度层故障 + 网络层故障，不是应用代码故障", "score": 3}, {"point": "两个独立故障叠加，必须分别识别和修复", "score": 4}, {"point": "checkoutservice 端口从 30007 改为 30009，地址看起来完全正确（都是腾讯云 NodePort），需要对比 svc 实际端口才能发现差异——是烟雾弹", "score": 4}, {"point": "核心证据分布在三朵云上（aws: resources, tencent: networkchaos, alicloud: env），必须跨云排查", "score": 3}], "tool_process": {"ideal_call_count": 14, "valuable_actions": ["get pods 发现 Pending", "describe pod 抓取 Insufficient 证据", "检查 deployment resources", "kubectl get networkchaos 发现实验", "区分真实故障与烟雾弹", "恢复后复核副本"]}}, "rubrics": {"positive_points_total": 100, "negative_points_total": -60, "rubrics": [{"criterion": "描述了观察到的故障现象（inventory 链路偶发失败、新 Pod Pending、运费计算偶尔出问题、checkoutservice 地址配置看着奇怪）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 AWS new-inventoryservice 的 pods，发现存在 Pending 状态的新 Pod", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 describe pod 发现 Events 中出现 Insufficient cpu/memory", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "检查了 new-inventoryservice 的 deployment resources，发现 requests 被异常抬高（cpu=8, memory=16Gi）", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "检查了腾讯云的 NetworkChaos 实验，发现 shippingservice 被注入 20% 丢包", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "准确给出 NetworkChaos 的具体参数（loss 20%）和目标服务（shippingservice，腾讯云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "检查了阿里云 checkoutservice 的 env 字段，发现 CART_SERVICE_ADDR 端口异常", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "对比了实际 service 端口，确认 CART_SERVICE_ADDR 端口从 30007 被改为 30009", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "正确识别出两个真实故障叠加（资源请求过高 + NetworkChaos 丢包），而非将所有症状归为单一根因", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "正确识别出 checkoutservice 端口篡改是烟雾弹，地址看起来完全正确但端口指向了错误的服务", "points": 5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "准确给出 new-inventoryservice requests 的具体值（cpu=8, memory=16Gi）和所在云（AWS）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 CART_SERVICE_ADDR 端口篡改细节（30007→30009）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：requests 过高 → 新 Pod 调度失败 Pending → 可用副本减少 → 峰值流量下部分请求失败", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：NetworkChaos 20% 丢包 → shippingservice 网络偶尔不通 → 运费计算偶尔失败", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解三个配置分布在三朵云上（aws: resources, tencent: networkchaos, alicloud: env），必须跨云排查", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出了修复步骤：回滚 new-inventoryservice 的资源配置", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复步骤：删除 NetworkChaos challenge-13-shipping-loss", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复步骤：修正 checkoutservice 的 CART_SERVICE_ADDR 端口为正确值（30007）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "在排查过程中使用了合理的命令和工具（kubectl get pods、describe pod、get networkchaos 等），而非瞎猜", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "模型在面对多个同时存在的故障时，能够正确识别每个故障的独立性，不将无关故障混为一谈", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型的排查过程覆盖了所有受影响的云（AWS、腾讯云、阿里云），检查了各云上的 chaos 实验和 deployment 配置", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型能够识别端口篡改（30007→30009）这种地址看起来完全正确的隐蔽配置错误，需要对比实际 service 端口才能发现", "points": 5, "tags": ["level:cluster", "cluster:smoke_bomb_detection", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（pod 状态、describe events、resources 字段、networkchaos 对象、env 字段），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "模型能够识别 inventoryservice 旧 Pod 仍正常服务的表象，不因'业务还能用'就忽略新 Pod Pending 的事实，理解可用副本正在衰减", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "将两个真实故障错误归因为单一根因（如只说资源问题或只说网络问题）", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 NetworkChaos 实验就下结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未检查 deployment resources 字段就下结论", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误判为网络基础设施问题或应用代码 bug，而实际是配置错误和混沌工程注入", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "仅查看单一云的服务配置就给出全局诊断", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "将烟雾弹（端口篡改）当作当前故障的主因，忽略了真正的 Pending 和 NetworkChaos 问题", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "忽略 checkoutservice 端口篡改，只修复 Pending 和 NetworkChaos 就认为全部解决", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "忽略 NetworkChaos 丢包，只修复资源请求就认为全部解决", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "建议扩容节点而非修复配置作为即时修复手段", "points": -2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 15 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 18 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 22 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 15}, "prompt": "运维反馈刚做完一次发布后，inventory 相关链路偶发失败。Deployment 副本数看着正常，但新 Pod 一直起不来。旧 Pod 还能顶一阵，所以不是全挂，就是时好时坏。运费计算也偶尔出问题，不是每次都失败。"}
{"canary": "imagepullbackoff-registry", "tags": ["发现", "容器与工作负载", "ImagePullBackOff", "网络丢包", "环境变量写错", "hard", "multi-cloud"], "case": {"id": "imagepullbackoff-registry", "title": "镜像拉取失败导致 ImagePullBackOff", "inject_script": "./challenge-14-imagepullbackoff-registry/inject.sh", "recover_script": "./challenge-14-imagepullbackoff-registry/recover.sh", "fault_phenomenon": "运维反馈某个新版本服务始终起不来，Deployment 一直卡在创建副本阶段。其它服务看起来正常，就这一个服务的 Pod 一直在拉镜像失败。导致链路中某一环节完全不可用，前端页面打开后部分功能直接报错。下单偶尔失败，不是每次都挂，10 次里有 2-3 次会超时。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "两个独立故障叠加：(1) 腾讯云 seat-1 的 new-gatewayservice 被改成了不存在的镜像 tag（v2.1.3-canary），tag 格式看起来像真实版本号，但实际不存在于仓库中，导致镜像拉取失败并进入 ImagePullBackOff；(2) AWS 对 orderservice 注入 NetworkChaos 25% 丢包，导致下单偶尔失败。阿里云 checkoutservice 的 PRODUCT_SERVICE_ADDR 端口被改为 30007（烟雾弹），地址看起来正确但端口指向了 cartservice 而非 productcatalogservice，不是当前故障的主因。", "faulty_clouds": ["unknown"], "affected_clouds": ["tencent", "aws", "aliyun"], "fault_location": {"module": "k8s-workload-image, network, k8s-service", "file_path": null, "function_or_config": "deployment/new-gatewayservice image=docker.io/library/new-gatewayservice:v2.1.3-canary; NetworkChaos: challenge-14-order-loss (loss 25%); deployment/checkoutservice PRODUCT_SERVICE_ADDR=43.139.187.221:30007", "description": "镜像 tag 写错（v2.1.3-canary），格式看起来像真实版本但实际不存在，导致镜像仓库返回 manifest unknown，Pod 无法启动（腾讯云）。NetworkChaos 注入 orderservice 25% 丢包，导致下单偶尔失败（AWS）。checkoutservice 的 PRODUCT_SERVICE_ADDR 端口被设为 30007（cartservice 的 NodePort），地址看起来完全正确，但端口指向了错误的服务，需要对比 svc 端口才能发现（阿里云，烟雾弹）。三个配置分布在三朵云上，必须分别检查。"}}, "reasoning_process": {"observed_symptoms": ["目标服务始终起不来，Pod 卡在拉镜像阶段。", "下单偶尔失败，10 次里有 2-3 次超时。", "其他服务看起来基本正常。", "有人发现 checkoutservice 地址配置看着奇怪。"], "key_evidence": [{"source": "kubectl --context=tencent -n seat-1 get pods -l app=new-gatewayservice", "content": "Pod 状态显示 ImagePullBackOff 或 ErrImagePull。", "conclusion": "镜像拉取失败。"}, {"source": "kubectl --context=tencent -n seat-1 describe pod <pod-name>", "content": "Events 出现 manifest unknown / not found。", "conclusion": "镜像 tag 不存在。"}, {"source": "kubectl --context=tencent -n seat-1 get deploy new-gatewayservice -o yaml", "content": "image 字段为 v2.1.3-canary，格式看起来像真实版本号。", "conclusion": "根因为镜像配置错误。"}, {"source": "kubectl --context=aws -n seat-1 get networkchaos", "content": "存在 challenge-14-order-loss 实验，action: loss, loss: 25%。", "conclusion": "orderservice 被注入网络丢包，下单偶尔失败不是镜像问题导致的。"}, {"source": "kubectl --context=alicloud -n seat-1 get deploy checkoutservice -o yaml | grep PRODUCT_SERVICE_ADDR", "content": "PRODUCT_SERVICE_ADDR=43.139.187.221:30007，地址正确但端口指向了 cartservice 而非 productcatalogservice。", "conclusion": "端口篡改是烟雾弹（30007 是 cartservice 不是 productcatalogservice），当前故障主因是镜像拉取失败和 NetworkChaos。需要对比实际 service 端口才能发现差异。"}], "causal_chain": ["镜像 tag 错误 -> 仓库找不到镜像 -> ErrImagePull/ImagePullBackOff -> 服务无法启动。", "NetworkChaos 注入 25% 丢包 -> orderservice 网络偶尔不通 -> 下单偶尔失败。", "checkoutservice 端口篡改是独立的配置问题，地址看起来完全正确，需要对比 svc 端口才能发现。不应与 ImagePullBackOff 或丢包混淆。", "两个真实故障独立存在，只修镜像不能解决下单丢包，只删 NetworkChaos 不能解决镜像问题。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=tencent -n seat-1 get pods -l app=new-gatewayservice", "purpose": "确认是否处于 ImagePullBackOff。", "expected_result": "ImagePullBackOff/ErrImagePull。"}, {"cmd": "kubectl --context=tencent -n seat-1 describe pod <pod-name>", "purpose": "确认镜像拉取失败原因。", "expected_result": "manifest unknown / not found。"}, {"cmd": "kubectl --context=aws -n seat-1 get networkchaos", "purpose": "查看 AWS NetworkChaos 实验。", "expected_result": "存在 challenge-14-order-loss 实验。"}, {"cmd": "kubectl --context=alicloud -n seat-1 get deploy checkoutservice -o yaml | grep PRODUCT_SERVICE_ADDR", "purpose": "确认 checkoutservice 地址配置。", "expected_result": "PRODUCT_SERVICE_ADDR=43.139.187.221:30007（端口篡改）。"}], "success_criteria": ["镜像恢复为有效 tag。", "Pod 成功拉起并 Running。", "NetworkChaos 实验被删除。", "服务恢复可访问。", "下单功能恢复正常。"]}, "proposed_resolution": {"suggestion": "两个真实故障必须分别修复：(1) 恢复 new-gatewayservice 镜像 tag；(2) 删除 NetworkChaos 实验。checkoutservice 的地址配置虽然错误，但当前优先修复主因。只修一个不能完全恢复。", "fix_steps": ["kubectl --context=tencent -n seat-1 set image deploy/new-gatewayservice server=docker.io/library/new-gatewayservice:v1.0-amd64，等待 rollout 完成，确认 Pod 拉取成功。", "kubectl --context=aws -n seat-1 delete networkchaos challenge-14-order-loss。", "kubectl --context=alicloud -n seat-1 set env deploy/checkoutservice PRODUCT_SERVICE_ADDR=43.139.187.221:30008，修正烟雾弹端口篡改。", "复核 Pod 是否 Running。", "验证下单功能恢复正常。"]}, "confidence": 0.95, "special_points": [{"point": "典型的拉镜像失败，优先看 events 而不是应用日志", "score": 4}, {"point": "镜像 tag v2.1.3-canary 格式看起来像真实版本，不能仅凭肉眼判断对错，必须通过 describe pod events 确认 manifest unknown", "score": 4}, {"point": "两个独立故障叠加，必须分别识别和修复", "score": 4}, {"point": "checkoutservice 端口篡改是烟雾弹，地址看起来完全正确，需要对比 svc 实际端口才能发现差异", "score": 4}, {"point": "核心证据分布在三朵云上（tencent: image, aws: networkchaos, alicloud: env），必须跨云排查", "score": 3}], "tool_process": {"ideal_call_count": 12, "valuable_actions": ["get pods 看拉取状态", "describe pod 看 events", "检查 deployment image", "kubectl get networkchaos 发现实验", "区分真实故障与烟雾弹", "恢复后核对 rollout"]}}, "rubrics": {"positive_points_total": 101, "negative_points_total": -59, "rubrics": [{"criterion": "描述了观察到的故障现象（某服务起不来、Pod 卡在拉镜像、下单偶尔失败、checkoutservice 地址配置看着奇怪）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了其他服务状态，确认故障涉及多个环节", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了腾讯云 new-gatewayservice 的 pods，发现状态为 ImagePullBackOff 或 ErrImagePull", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "通过 describe pod 发现 Events 中出现 manifest unknown / not found", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "检查了 new-gatewayservice 的 deployment image 字段，发现 tag 为 v2.1.3-canary", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "检查了 AWS 的 NetworkChaos 实验，发现 orderservice 被注入 25% 丢包", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "准确给出 NetworkChaos 的具体参数（loss 25%）和目标服务（orderservice，AWS）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "检查了阿里云 checkoutservice 的 env 字段，发现 PRODUCT_SERVICE_ADDR 端口指向了 cartservice（30007）而非 productcatalogservice", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "对比了实际 service 端口，确认 PRODUCT_SERVICE_ADDR 端口 30007 指向了 cartservice 而非 productcatalogservice", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "正确识别出两个真实故障叠加（镜像 tag 错误 + NetworkChaos 丢包），而非将所有症状归为单一根因", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "正确识别出 checkoutservice 端口篡改是烟雾弹，地址看起来完全正确但端口指向了错误的服务", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "准确给出 image tag 的具体值（v2.1.3-canary）和所在云（腾讯云）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 PRODUCT_SERVICE_ADDR 的错误端口值", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：镜像 tag 错误 → 仓库找不到镜像 → ErrImagePull/ImagePullBackOff → 服务无法启动", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：NetworkChaos 25% 丢包 → orderservice 网络偶尔不通 → 下单偶尔失败", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解三个配置分布在三朵云上（tencent: image, aws: networkchaos, alicloud: env），必须跨云排查", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出了修复步骤：将镜像改回正确 tag", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复步骤：删除 NetworkChaos challenge-14-order-loss", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复步骤：修正 checkoutservice 的 PRODUCT_SERVICE_ADDR 端口", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "修复时明确改回正确 tag，避免 rollout 卡死", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "验证修复后 checkoutservice PRODUCT_SERVICE_ADDR 恢复正确", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "在排查过程中使用了合理的命令和工具（kubectl get pods、describe pod、get networkchaos 等），而非瞎猜", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "模型在面对多个同时存在的故障时，能够正确识别每个故障的独立性，不将无关故障混为一谈", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型的排查过程覆盖了所有受影响的云（腾讯云、AWS、阿里云），检查了各云上的 chaos 实验和 deployment 配置", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型能够识别镜像 tag 格式正确（v2.1.3-canary）但实际不存在的误导性配置，不被 tag 外观和烟雾弹迷惑", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（pod 状态、describe events、image 字段、networkchaos 对象、env 字段），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "模型通过 describe pod Events 获取具体报错信息（manifest unknown / unauthorized），而非仅凭 ImagePullBackOff 状态就下结论", "points": 5, "tags": ["level:cluster", "cluster:image_tag_forensics", "axis:accuracy"]}, {"criterion": "将两个真实故障错误归因为单一根因（如只说镜像问题或只说网络问题）", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "误判为网络基础设施问题或镜像仓库故障，而实际是配置错误和混沌工程注入", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未通过 describe pod 确认 manifest unknown 就下结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "仅凭肉眼判断 tag v2.1.3-canary 格式正确就认为镜像没问题", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "将烟雾弹（端口篡改）当作当前故障的主因，忽略了真正的镜像 tag 和 NetworkChaos 问题", "points": -6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "误判为应用代码 bug 或配置文件错误，而实际是镜像 tag 不存在和 NetworkChaos 注入", "points": -4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "尝试重启 pod 而不修复 image tag，导致反复进入 ImagePullBackOff", "points": -3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "忽略 checkoutservice 端口篡改，只修复镜像 tag 和 NetworkChaos 就认为全部解决", "points": -4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "忽略 NetworkChaos 丢包，只修复镜像 tag 就认为全部解决", "points": -4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "建议修改 Dockerfile 重新构建镜像作为即时修复手段，而非改回正确 tag", "points": -3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 12 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 15 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 18 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 12}, "prompt": "运维反馈某个新版本服务始终起不来，Deployment 一直卡在创建副本阶段。其它服务看起来正常，就这一个服务的 Pod 一直在拉镜像失败。导致链路中某一环节完全不可用，前端页面打开后部分功能直接报错。下单偶尔失败，不是每次都挂，10 次里有 2-3 次会超时。"}
{"canary": "probe-too-aggressive", "tags": ["发现", "容器与工作负载", "Readiness Probe 摘除", "I/O 延迟", "medium", "multi-cloud"], "case": {"id": "probe-too-aggressive", "title": "探针配置过激导致服务抖动", "inject_script": "./challenge-15-probe-too-aggressive/inject.sh", "recover_script": "./challenge-15-probe-too-aggressive/recover.sh", "fault_phenomenon": "用户反馈前端偶尔短暂不可用，刷新一下又好了。购物车操作也变慢了，加个商品要等好几秒，偶尔还超时。不是一直坏，就是时好时坏，没有明显规律。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "两个独立故障叠加：(1) 阿里云 seat-1 的 frontend readiness probe 参数被设置得过于激进（initialDelay=0, timeout=1, failureThreshold=1, period=2），导致健康检查误判，Pod 被频繁从 Service 后端摘除；(2) 腾讯云对 redis-cart 注入 IOChaos 2500ms 延迟（70% 操作），导致购物车操作变慢或超时。", "faulty_clouds": ["unknown"], "affected_clouds": ["aliyun", "tencent"], "fault_location": {"module": "k8s-probe, redis", "file_path": null, "function_or_config": "deployment/frontend readinessProbe (initialDelay=0, timeout=1, failureThreshold=1, period=2); IOChaos: challenge-15-cart-io (delay 2500ms, 70%)", "description": "探针过激造成 readiness 抖动导致前端短时不可达（阿里云），I/O 延迟导致购物车操作变慢（腾讯云）。两个故障独立存在，必须分别修复。"}}, "reasoning_process": {"observed_symptoms": ["前端偶尔短暂不可用，刷新一下又好了。", "购物车操作变慢，加个商品要等好几秒，偶尔还超时。", "Pod 多数为 Running，但请求偶发失败。", "表现为短时波动而非持续宕机。"], "key_evidence": [{"source": "kubectl --context=alicloud -n seat-1 describe pod <frontend-pod>", "content": "Readiness probe failed 事件频繁出现。", "conclusion": "健康检查参数不合理，导致 Pod 间歇摘除。"}, {"source": "kubectl --context=alicloud -n seat-1 get deploy frontend -o yaml", "content": "readinessProbe 参数过于激进（initialDelay=0, timeout=1, failureThreshold=1）。", "conclusion": "配置误改为前端抖动的根因。"}, {"source": "kubectl --context=tencent -n seat-1 get iochaos", "content": "存在 challenge-15-cart-io 实验，action=latency, delay=2500ms, percent=70。", "conclusion": "redis-cart 被注入 I/O 延迟，购物车操作变慢不是探针问题导致的。"}, {"source": "kubectl --context=tencent -n seat-1 get endpoints frontend -w", "content": "后端 endpoint 间歇减少/恢复。", "conclusion": "readiness 抖动导致摘除。"}], "causal_chain": ["readiness 参数过激 -> 探针误判失败 -> Pod 从 endpoints 摘除 -> 前端偶发不可用 -> 刷新后又恢复。", "IOChaos 注入 redis-cart I/O 延迟 -> 读写 redis 变慢 -> 购物车操作变慢或超时。", "两个故障独立存在，只修探针不能解决购物车慢，只删 IOChaos 不能解决前端抖动。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=alicloud -n seat-1 describe pod <frontend-pod>", "purpose": "检查 readiness 失败事件。", "expected_result": "Readiness probe failed 频繁。"}, {"cmd": "kubectl --context=alicloud -n seat-1 get deploy frontend -o yaml | grep -A15 -n \"readinessProbe\"", "purpose": "确认探针参数。", "expected_result": "timeout/failureThreshold/initialDelay 明显激进。"}, {"cmd": "kubectl --context=tencent -n seat-1 get iochaos", "purpose": "查看腾讯云 IOChaos 实验。", "expected_result": "存在 challenge-15-cart-io 实验。"}, {"cmd": "kubectl --context=alicloud -n seat-1 get endpoints frontend -w", "purpose": "观察 endpoints 抖动。", "expected_result": "后端 endpoint 间歇减少/恢复。"}], "success_criteria": ["frontend 探针配置恢复。", "IOChaos 实验被删除。", "endpoints 不再频繁抖动。", "前端访问和购物车操作均恢复稳定。"]}, "proposed_resolution": {"suggestion": "两个故障必须分别修复：(1) 回滚 frontend 的探针配置；(2) 删除 IOChaos 实验。只修一个不能完全恢复。", "fix_steps": ["kubectl --context=alicloud -n seat-1 patch deploy frontend --type='json' -p='[{\"op\":\"replace\",\"path\":\"/spec/template/spec/containers/0/readinessProbe\",\"value\":{\"httpGet\":{\"path\":\"/_healthz\",\"port\":8080},\"initialDelaySeconds\":10,\"periodSeconds\":10,\"timeoutSeconds\":5,\"failureThreshold\":3,\"successThreshold\":1}}]'，等待 rollout 完成。", "kubectl --context=tencent -n seat-1 delete iochaos challenge-15-cart-io。", "确认 frontend Pod readiness 事件不再频繁失败。", "验证前端访问和购物车操作均恢复正常。"]}, "confidence": 0.93, "special_points": [{"point": "两个独立故障叠加，必须分别识别和修复", "score": 4}, {"point": "Running 不代表可接流量，需看 readiness 与 endpoints", "score": 4}, {"point": "前端抖动和购物车慢是两个不同来源的故障，不能混为一谈", "score": 3}, {"point": "IOChaos 影响的是 redis-cart（购物车），readiness 影响的是 frontend（前端），症状来源不同", "score": 3}], "tool_process": {"ideal_call_count": 12, "valuable_actions": ["describe pod 看 probe fail", "检查 deployment probe 参数", "kubectl get iochaos 发现 I/O 延迟实验", "观察 endpoints 抖动", "恢复后复核稳定性"]}}, "rubrics": {"positive_points_total": 101, "negative_points_total": -51, "rubrics": [{"criterion": "描述了观察到的故障现象（前端偶尔短暂不可用、刷新后恢复、购物车操作变慢偶尔超时）", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了阿里云 frontend 的 pod 状态，发现 Running 但 readiness 事件频繁失败", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 describe pod 发现 Readiness probe failed 事件频繁出现", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "检查了 frontend 的 deployment readinessProbe 参数，发现配置过于激进（initialDelay=0, timeout=1, failureThreshold=1）", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "检查了腾讯云 IOChaos 实验，发现 redis-cart 被注入 I/O 延迟", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 kubectl get endpoints frontend -w 观察到后端 endpoint 间歇减少/恢复", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "正确识别出两个独立故障叠加，而非将所有症状归为单一根因", "points": 7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "准确给出 readinessProbe 的具体参数（initialDelay=0, timeout=1, failureThreshold=1, period=2）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 IOChaos 的具体参数（delay 2500ms, percent 70）和目标（redis-cart，腾讯云）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：readiness 参数过激 → 探针误判失败 → Pod 从 endpoints 摘除 → 前端偶发不可用 → 刷新后恢复", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：IOChaos 注入 redis-cart I/O 延迟 → 读写 redis 变慢 → 购物车操作变慢或超时", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解两个故障独立存在，只修探针不能解决购物车慢，只删 IOChaos 不能解决前端抖动", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出了修复步骤：回滚 frontend 的探针配置", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复步骤：删除 IOChaos challenge-15-cart-io", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "在排查过程中使用了合理的命令和工具（kubectl describe、get iochaos、get endpoints 等），而非瞎猜", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "模型在面对多个同时存在的故障时，能够正确识别每个故障的独立性，不将无关故障混为一谈", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型的排查过程覆盖了所有受影响的云（阿里云、腾讯云），检查了各云上的 chaos 实验和 deployment 配置", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型能够区分 Pod Running 和 Pod 可接流量（readiness 通过）的区别，不被'Running'状态误导", "points": 5, "tags": ["level:cluster", "cluster:readiness_vs_running_distinction", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（pod 状态、describe events、readinessProbe 参数、iochaos 对象、endpoints），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "模型能够识别 IOChaos 70% 概率生效的特征，不因'有时候购物车又正常了'就认为问题已消失或不是 Redis 层的问题", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "将两个故障错误归因为单一根因（如只说探针问题或只说 Redis 问题）", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 readinessProbe 参数和 IOChaos 实验就下结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误判为 Redis 服务本身故障（如 OOM、崩溃），而非 IOChaos 注入的 I/O 延迟", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "仅查看单一云的资源就给出全局诊断", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误判为网络问题或 DNS 问题，而实际是探针配置和 I/O 延迟问题", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "忽略 readiness probe 配置问题，只关注 IOChaos 就认为全部解决", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "建议修改应用代码（如增加重试、优化 Redis 连接池）作为即时修复手段，而非恢复配置和删除 chaos", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 12 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 15 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 18 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 12}, "prompt": "用户反馈前端偶尔短暂不可用，刷新一下又好了。购物车操作也变慢了，加个商品要等好几秒，偶尔还超时。不是一直坏，就是时好时坏，没有明显规律。"}
{"canary": "rum-route-split", "tags": ["定位", "可观测系统自身", "RUM数据异常", "指标缺失", "medium", "multi-cloud"], "case": {"id": "rum-route-split", "title": "RUM路由分裂：首页正常、商品页观测失真", "inject_script": "./challenge-16-rum-route-split/inject.sh", "recover_script": "./challenge-16-rum-route-split/recover.sh", "fault_phenomenon": "用户说商城可用，但运营发现 RUM 中首页指标基本正常，商品详情相关指标与测速几乎消失", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。该题为 RUM-only 题。建议用 RUM 控制台/API 做页面维度对比，再结合 31380 页面注入脚本与配置排查。"}, "ideal_answer": {"fault_info": {"root_cause": "rum-nginx-config 的 ConfigMap 中，RUM 初始化脚本按 pathname 分支配置：当路径包含 /product/ 时，sample 从 1 降为 0.01，且 reportApiSpeed 和 reportAssetSpeed 被关闭。导致商品页 RUM 数据极度稀疏、测速指标消失。", "faulty_clouds": ["aliyun"], "affected_clouds": ["aliyun", "tencent"], "fault_location": {"module": "k8s-configmap", "file_path": "rum-nginx-config/default.conf", "function_or_config": "sub_filter 注入的 Aegis 初始化脚本中 isProductPath 分支", "description": "ConfigMap 中 nginx sub_filter 注入的 RUM SDK 初始化代码，对 /product/ 路由做了降级：sample=0.01、关闭 reportApiSpeed 和 reportAssetSpeed。"}}, "reasoning_process": {"observed_symptoms": ["RUM 控制台中首页指标基本正常，商品详情页指标与测速几乎消失。", "页面源码中可见 Aegis SDK（aegis.min.js）已加载。", "业务功能正常，用户可正常访问商品页。", "腾讯云 new-gatewayservice 有 GATEWAY_VERBOSE_LOG=on 环境变量（烟雾弹）。"], "key_evidence": [{"source": "RUM 控制台页面维度对比", "content": "首页 / PV 正常，商品页 /product/* PV 极低或为零", "conclusion": "RUM 数据按路由分裂，商品页上报被压制。"}, {"source": "curl http://8.163.35.175:31366/product/1 | grep aegis", "content": "页面 HTML 中 Aegis SDK 存在，但初始化参数中 sample=0.01、reportApiSpeed=false", "conclusion": "SDK 存在但配置被降级，非埋点缺失。"}, {"source": "kubectl --context=aliyun get configmap rum-nginx-config -n seat-1 -o yaml", "content": "default.conf 中 sub_filter 注入的 JS 包含 isProductPath 分支逻辑", "conclusion": "根因在 ConfigMap 中的路由分支降级逻辑。"}, {"source": "kubectl --context=tencent get deployment new-gatewayservice -n seat-1 -o yaml | grep GATEWAY_VERBOSE_LOG", "content": "GATEWAY_VERBOSE_LOG=on", "conclusion": "这是烟雾弹，仅影响 gateway 日志详细度，与 RUM 无关。"}], "causal_chain": ["rum-nginx-config ConfigMap 中注入的 RUM 初始化脚本按 pathname 分支", "当路径包含 /product/ 时，sample 从 1 降为 0.01，reportApiSpeed 和 reportAssetSpeed 设为 false", "商品页 RUM 上报采样率降至 1%，测速数据完全不上报", "RUM 控制台表现为首页正常、商品页数据极度稀疏", "腾讯云 GATEWAY_VERBOSE_LOG=on 是烟雾弹，与 RUM 数据分裂无关"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=aliyun get configmap rum-nginx-config -n seat-1 -o yaml", "purpose": "检查 RUM 注入配置中是否存在路由分支降级逻辑。", "expected_result": "ConfigMap 中不应有 isProductPath 分支，sample 应统一为 1。"}, {"cmd": "curl -s http://8.163.35.175:31366/product/1 | grep -o 'sample:[^,]*'", "purpose": "验证商品页 RUM sample 值。", "expected_result": "sample: 1（修复后应为统一值）。"}, {"cmd": "curl -s http://8.163.35.175:31366/ | grep -o 'sample:[^,]*'", "purpose": "对比首页 RUM sample 值。", "expected_result": "sample: 1，与商品页一致。"}, {"cmd": "RUM 控制台页面维度对比", "purpose": "验证修复后首页和商品页 RUM 数据趋势恢复一致。", "expected_result": "商品页 PV 和测速数据恢复正常密度。"}], "success_criteria": ["rum-nginx-config ConfigMap 中移除了 isProductPath 分支逻辑。", "所有页面 RUM sample 统一为 1，reportApiSpeed 和 reportAssetSpeed 统一开启。", "RUM 控制台中首页和商品页数据密度恢复正常。", "腾讯云烟雾弹（GATEWAY_VERBOSE_LOG）被正确识别，未被误修复。"]}, "proposed_resolution": {"suggestion": "修改 rum-nginx-config ConfigMap，移除 isProductPath 路由分支逻辑，恢复统一的 RUM 配置（sample=1、reportApiSpeed=true、reportAssetSpeed=true）。重启 frontend-rum-proxy 使配置生效。", "fix_steps": ["执行 kubectl --context=aliyun edit configmap rum-nginx-config -n seat-1，移除 isProductPath 分支逻辑，恢复 sample=1、reportApiSpeed=true、reportAssetSpeed=true。", "执行 kubectl --context=aliyun rollout restart deploy/frontend-rum-proxy -n seat-1。", "等待 rollout 完成。", "验证商品页 HTML 中 RUM 参数已恢复统一。", "在 RUM 控制台确认首页和商品页数据趋势恢复正常。"]}, "confidence": 0.95, "special_points": [{"point": "页面源码有 SDK ≠ RUM 正常，必须检查初始化参数", "score": 4}, {"point": "RUM 数据必须按页面维度对比，不能只看全局均值", "score": 4}, {"point": "腾讯云 GATEWAY_VERBOSE_LOG=on 是烟雾弹，与 RUM 无关", "score": 3}, {"point": "sample=0.01 意味着 99% 数据被丢弃，但不是完全消失", "score": 3}], "tool_process": {"ideal_call_count": 14, "valuable_actions": ["RUM 控制台/API 按页面维度对比数据密度", "curl 检查页面 HTML 中 RUM 初始化参数", "kubectl get configmap rum-nginx-config 查看注入脚本", "kubectl rollout restart 恢复服务", "修复后验证 RUM 数据恢复"]}}, "rubrics": {"positive_points_total": 101, "negative_points_total": -52, "rubrics": [{"criterion": "描述了现象：业务可用但 RUM 在页面维度出现分裂（首页与商品页表现不一致）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "明确指出‘页面有 SDK’与‘RUM 上报正常’不是同一件事", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "通过 RUM 控制台或 API 对比出首页和商品页数据密度差异（必须使用 RUM）", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "收集了页面路径维度证据（至少区分 / 与 /product/）", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 31380 返回 HTML，确认 SDK 与初始化代码存在", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 rum-nginx-config 的注入脚本内容", "points": 8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出存在按 pathname 分支的 RUM 配置逻辑", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 /product/ 路由命中降级逻辑", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确指出 sample 在商品页被下调为 0.01", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确指出商品页 reportApiSpeed 被关闭", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确指出商品页 reportAssetSpeed 被关闭", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别 tencent new-gatewayservice 的配置变更为烟雾弹而非主根因", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "能区分‘业务可用性’和‘可观测性完整性’两条线", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出完整因果链：路由分支降级→商品页上报稀疏→RUM 页面维度分裂", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "提出修复动作：移除路由分支降级逻辑或恢复统一采样与测速开关", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出离谱操作（如重启无关服务、修改不存在的ConfigMap、跨命名空间变更）", "points": -6, "tags": ["level:example", "axis:instruction_following"]}, {"criterion": "响应结构清晰：现象→证据→根因→修复→验证", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "模型在多线索下能识别主因与烟雾弹，避免混因", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型排查覆盖跨云信息（阿里云主因 + 腾讯云烟雾弹），避免单云结论", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型结论基于 RUM 现象 + 配置证据闭环，而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "模型能识别‘SDK存在但观测异常’这类误导信号", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型能识别‘页面维度分裂’这一题目独有模式，而非只看全局均值", "points": 5, "tags": ["level:cluster", "cluster:route_level_observability_diagnosis", "axis:context_awareness"]}, {"criterion": "仅凭‘页面有 aegis.min.js’就断言 RUM 正常", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未使用 RUM 页面维度证据就直接下结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "将烟雾弹（new-gatewayservice）误判为主根因", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "忽略 sample 与测速开关参数，只做模糊归因", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "修复后未做 RUM 恢复验证就宣布完成", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 14 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 17 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 21 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "未识别出 RUM 初始化脚本在 nginx sub_filter 中注入而非前端代码", "points": -1, "tags": ["level:example", "axis:accuracy"]}], "ideal_call_count": 14}, "prompt": "用户说商城可用，但运营发现 RUM 中首页指标基本正常，商品详情相关指标与测速几乎消失"}
{"canary": "rum-bot-filter-trap", "tags": ["定位", "可观测系统自身", "RUM数据异常", "指标缺失", "medium", "multi-cloud"], "case": {"id": "rum-bot-filter-trap", "title": "RUM User-Agent 条件分流陷阱", "inject_script": "./challenge-17-trace-gap-hidden-latency/inject.sh", "recover_script": "./challenge-17-trace-gap-hidden-latency/recover.sh", "fault_phenomenon": "监控同学说：浏览器访问看起来正常，但用脚本压测或巡检时，RUM 数据几乎没有。页面源码依然有 SDK。请解释为何「不同访问来源」在 RUM 中表现不同，并定位根因。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。该题为 RUM-only 题。必须利用 RUM 现象差异（浏览器 vs 非浏览器来源）定位问题。"}, "ideal_answer": {"fault_info": {"root_cause": "rum-nginx-config 的 ConfigMap 中，RUM 初始化脚本按 User-Agent 分支配置：当 UA 不包含 'mozilla'（即非浏览器访问）时，sample 从 1 降为 0.01，且 reportApiSpeed 和 reportAssetSpeed 被关闭。导致脚本/压测工具的 RUM 数据极度稀疏。", "faulty_clouds": ["aliyun"], "affected_clouds": ["aliyun", "tencent"], "fault_location": {"module": "k8s-configmap", "file_path": "rum-nginx-config/default.conf", "function_or_config": "sub_filter 注入的 Aegis 初始化脚本中 UA 条件分支", "description": "ConfigMap 中 nginx sub_filter 注入的 RUM SDK 初始化代码，对非 mozilla UA 做了降级：sample=0.01、关闭 reportApiSpeed 和 reportAssetSpeed。"}}, "reasoning_process": {"observed_symptoms": ["浏览器访问时 RUM 数据正常，脚本/压测工具访问时 RUM 数据几乎没有。", "页面源码中可见 Aegis SDK 已加载。", "业务功能正常，不同访问方式都能正常获取页面。", "腾讯云 new-gatewayservice 有 FEATURE_FLAG_GATEWAY_CANARY=on 环境变量（烟雾弹）。"], "key_evidence": [{"source": "RUM 控制台来源维度对比", "content": "浏览器来源 PV 正常，非浏览器来源 PV 极低", "conclusion": "RUM 数据按 UA 来源分裂，非浏览器上报被压制。"}, {"source": "curl http://8.163.35.175:31366/ | grep aegis", "content": "curl 返回的 HTML 中 Aegis 初始化参数 sample=0.01（因为 curl UA 不含 mozilla）", "conclusion": "非浏览器 UA 命中降级逻辑。"}, {"source": "kubectl --context=aliyun get configmap rum-nginx-config -n seat-1 -o yaml", "content": "default.conf 中 sub_filter 注入的 JS 包含 ua.indexOf('mozilla') === -1 分支逻辑", "conclusion": "根因在 ConfigMap 中的 UA 条件降级逻辑。"}, {"source": "kubectl --context=tencent get deployment new-gatewayservice -n seat-1 -o yaml | grep FEATURE_FLAG_GATEWAY_CANARY", "content": "FEATURE_FLAG_GATEWAY_CANARY=on", "conclusion": "这是烟雾弹，仅影响 gateway 金丝雀发布，与 RUM 无关。"}], "causal_chain": ["rum-nginx-config ConfigMap 中注入的 RUM 初始化脚本按 UA 分支", "当 navigator.userAgent 不包含 'mozilla' 时，sample 从 1 降为 0.01，reportApiSpeed 和 reportAssetSpeed 设为 false", "脚本/压测工具 UA 通常不含 'mozilla'，命中降级逻辑", "非浏览器来源 RUM 上报采样率降至 1%，测速数据完全不上报", "RUM 控制台表现为浏览器正常、脚本/压测数据极度稀疏", "腾讯云 FEATURE_FLAG_GATEWAY_CANARY=on 是烟雾弹，与 RUM 无关"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=aliyun get configmap rum-nginx-config -n seat-1 -o yaml", "purpose": "检查 RUM 注入配置中是否存在 UA 条件降级逻辑。", "expected_result": "ConfigMap 中不应有 ua.indexOf('mozilla') 分支，sample 应统一为 1。"}, {"cmd": "curl -s http://8.163.35.175:31366/ | grep -o 'sample:[^,]*'", "purpose": "验证非浏览器访问时 RUM sample 值。", "expected_result": "sample: 1（修复后应为统一值，不受 UA 影响）。"}, {"cmd": "RUM 控制台来源维度对比", "purpose": "验证修复后浏览器和非浏览器来源 RUM 数据趋势恢复一致。", "expected_result": "非浏览器来源 PV 和测速数据恢复正常密度。"}], "success_criteria": ["rum-nginx-config ConfigMap 中移除了 UA 条件分支逻辑。", "所有访问来源 RUM sample 统一为 1，reportApiSpeed 和 reportAssetSpeed 统一开启。", "RUM 控制台中浏览器和非浏览器来源数据密度恢复正常。", "腾讯云烟雾弹（FEATURE_FLAG_GATEWAY_CANARY）被正确识别，未被误修复。"]}, "proposed_resolution": {"suggestion": "修改 rum-nginx-config ConfigMap，移除 UA 条件分支逻辑，恢复统一的 RUM 配置（sample=1、reportApiSpeed=true、reportAssetSpeed=true）。重启 frontend-rum-proxy 使配置生效。", "fix_steps": ["执行 kubectl --context=aliyun edit configmap rum-nginx-config -n seat-1，移除 ua.indexOf('mozilla') 分支逻辑，恢复 sample=1、reportApiSpeed=true、reportAssetSpeed=true。", "执行 kubectl --context=aliyun rollout restart deploy/frontend-rum-proxy -n seat-1。", "等待 rollout 完成。", "验证 curl 返回的 HTML 中 RUM 参数已恢复统一。", "在 RUM 控制台确认不同来源数据趋势恢复正常。"]}, "confidence": 0.95, "special_points": [{"point": "浏览器 vs 非浏览器的 UA 差异是关键区分条件", "score": 4}, {"point": "curl 等工具默认 UA 不含 'mozilla'，会命中降级逻辑", "score": 4}, {"point": "腾讯云 FEATURE_FLAG_GATEWAY_CANARY=on 是烟雾弹，与 RUM 无关", "score": 3}, {"point": "必须用 RUM 控制台按来源维度对比，不能只看全局数据", "score": 3}], "tool_process": {"ideal_call_count": 14, "valuable_actions": ["RUM 控制台/API 按来源维度对比数据密度", "curl 检查返回 HTML 中 RUM 初始化参数", "kubectl get configmap rum-nginx-config 查看注入脚本", "kubectl rollout restart 恢复服务", "修复后验证 RUM 数据恢复"]}}, "rubrics": {"positive_points_total": 101, "negative_points_total": -53, "rubrics": [{"criterion": "描述了现象：浏览器访问与脚本/巡检访问在 RUM 中表现不一致", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 RUM 数据证明来源维度存在明显分裂（必须使用 RUM）", "points": 6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "明确指出‘业务可访问’不代表‘所有来源的 RUM 都健康’", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "检查了页面注入代码并确认 SDK 初始化存在", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 rum-nginx-config 中与 UA 相关的条件逻辑", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 non-mozilla 来源触发降级分支", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确指出非 mozilla 来源 sample 降至 0.01", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确指出非 mozilla 来源 reportApiSpeed 被关闭", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确指出非 mozilla 来源 reportAssetSpeed 被关闭", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别 tencent new-gatewayservice 配置变更为烟雾弹", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "能够解释为什么‘只看浏览器’可能误判为系统正常", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "能够解释为什么‘只看脚本来源’会放大故障感知", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出完整因果链：UA条件命中→采样/测速降级→来源维度RUM分裂", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出修复动作：移除 UA 条件降级分支", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出修复动作：恢复统一 sample=1 与测速开关", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出离谱操作（如重启无关服务、修改不存在的ConfigMap、跨命名空间变更）", "points": -6, "tags": ["level:example", "axis:instruction_following"]}, {"criterion": "响应结构清晰：现象→证据→根因→修复→验证", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "模型在多信号下能区分主因与烟雾弹", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型跨云检查证据而非单云拍脑袋", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型结论由 RUM 现象与配置证据共同支撑", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "模型不被‘SDK存在’与‘业务可用’表象误导", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型能识别该题独有的来源维度分裂模式（UA条件观测偏置）", "points": 5, "tags": ["level:cluster", "cluster:source_bias_observability_diagnosis", "axis:context_awareness"]}, {"criterion": "不看 RUM 来源差异证据就直接下结论", "points": -10, "tags": ["level:example", "axis:completeness"]}, {"criterion": "将烟雾弹配置误判为核心根因", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "遗漏 sample 或测速开关等关键参数细节", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "把问题误判为后端服务故障而非 RUM 采集策略问题", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "修复后未验证不同来源恢复一致", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 14 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 17 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 21 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 14}, "prompt": "监控同学说：浏览器访问看起来正常，但用脚本压测或巡检时，RUM 数据几乎没有。页面源码依然有 SDK。请解释为何「不同访问来源」在 RUM 中表现不同，并定位根因。"}
{"canary": "rum-multi-factor-split", "tags": ["定位", "可观测系统自身", "RUM数据异常", "指标缺失", "hard", "multi-cloud"], "case": {"id": "rum-multi-factor-split", "title": "RUM多因子分裂：路由与来源双重条件", "inject_script": "./challenge-18-trace-context-lost/inject.sh", "recover_script": "./challenge-18-trace-context-lost/recover.sh", "fault_phenomenon": "运营发现 RUM 指标出现复杂分裂：部分页面有数据、部分页面稀疏；浏览器访问和脚本访问表现也不同。页面可访问、SDK存在。请定位双条件导致的 RUM 观测分裂并完成修复。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。该题为高难 RUM-only 题。必须基于 RUM 证据识别「路由+来源」双条件影响，再结合注入参数闭环定位。"}, "ideal_answer": {"fault_info": {"root_cause": "rum-nginx-config 的 ConfigMap 中，RUM 初始化脚本同时按 pathname 和 User-Agent 双条件分支配置：(1) /product/ 路由 sample 从 1 降为 0.05；(2) 非 mozilla UA 关闭 reportApiSpeed 和 reportAssetSpeed。双条件叠加导致 RUM 数据出现复杂分裂。", "faulty_clouds": ["aliyun"], "affected_clouds": ["aliyun", "tencent"], "fault_location": {"module": "k8s-configmap", "file_path": "rum-nginx-config/default.conf", "function_or_config": "sub_filter 注入的 Aegis 初始化脚本中 isProductPath 和 UA 双条件分支", "description": "ConfigMap 中 nginx sub_filter 注入的 RUM SDK 初始化代码，同时存在两个独立的降级条件：路由条件（/product/ 降采样）和 UA 条件（非浏览器关闭测速）。"}}, "reasoning_process": {"observed_symptoms": ["RUM 控制台中部分页面有数据、部分页面稀疏。", "浏览器访问和脚本访问表现不同。", "页面源码中可见 Aegis SDK 已加载。", "业务功能正常，所有页面可正常访问。", "腾讯云 new-gatewayservice 有 GATEWAY_DIAG_MODE=verbose 环境变量（烟雾弹）。"], "key_evidence": [{"source": "RUM 控制台页面+来源交叉维度对比", "content": "首页浏览器 PV 正常，首页脚本 PV 正常；商品页浏览器 PV 稀疏，商品页脚本 PV 极稀疏", "conclusion": "路由条件影响所有来源（sample=0.05），UA 条件额外影响非浏览器测速。"}, {"source": "curl http://8.163.35.175:31366/product/1 | grep aegis", "content": "sample=0.05, reportApiSpeed=false, reportAssetSpeed=false", "conclusion": "商品页+非浏览器同时命中两个降级条件。"}, {"source": "浏览器访问 http://8.163.35.175:31366/product/1 查看源码", "content": "sample=0.05, reportApiSpeed=true, reportAssetSpeed=true", "conclusion": "商品页浏览器只命中路由条件（sample 降级），测速正常。"}, {"source": "kubectl --context=aliyun get configmap rum-nginx-config -n seat-1 -o yaml", "content": "default.conf 中同时存在 isProductPath 和 ua.indexOf('mozilla') 两个独立分支", "conclusion": "根因是双条件叠加，非单一条件。"}, {"source": "kubectl --context=tencent get deployment new-gatewayservice -n seat-1 -o yaml | grep GATEWAY_DIAG_MODE", "content": "GATEWAY_DIAG_MODE=verbose", "conclusion": "这是烟雾弹，仅影响 gateway 诊断日志，与 RUM 无关。"}], "causal_chain": ["rum-nginx-config ConfigMap 中注入的 RUM 初始化脚本存在两个独立分支", "条件1：isProductPath → sample 从 1 降为 0.05（影响所有来源的商品页）", "条件2：ua.indexOf('mozilla') === -1 → reportApiSpeed=false, reportAssetSpeed=false（影响非浏览器的所有页面）", "双条件叠加：商品页+非浏览器 → sample=0.05 + 测速关闭 → 数据极度稀疏", "商品页+浏览器 → sample=0.05 + 测速正常 → 数据稀疏但有测速", "首页+非浏览器 → sample=1 + 测速关闭 → 数据正常但无测速", "首页+浏览器 → 无降级 → 完全正常", "RUM 控制台表现为复杂的多维分裂", "腾讯云 GATEWAY_DIAG_MODE=verbose 是烟雾弹，与 RUM 无关"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=aliyun get configmap rum-nginx-config -n seat-1 -o yaml", "purpose": "检查 RUM 注入配置中是否存在双条件降级逻辑。", "expected_result": "ConfigMap 中不应有 isProductPath 和 ua 分支，sample 应统一为 1。"}, {"cmd": "curl -s http://8.163.35.175:31366/product/1 | grep -o 'sample:[^,]*'", "purpose": "验证商品页+非浏览器 RUM sample 值。", "expected_result": "sample: 1（修复后应为统一值）。"}, {"cmd": "curl -s http://8.163.35.175:31366/ | grep -o 'sample:[^,]*'", "purpose": "对比首页 RUM sample 值。", "expected_result": "sample: 1，与商品页一致。"}, {"cmd": "RUM 控制台交叉维度对比", "purpose": "验证修复后所有页面+来源组合的 RUM 数据恢复正常。", "expected_result": "所有维度数据密度恢复正常，无分裂现象。"}], "success_criteria": ["rum-nginx-config ConfigMap 中移除了 isProductPath 和 UA 条件分支逻辑。", "所有页面和来源 RUM sample 统一为 1，reportApiSpeed 和 reportAssetSpeed 统一开启。", "RUM 控制台中所有维度数据恢复正常，无分裂现象。", "腾讯云烟雾弹（GATEWAY_DIAG_MODE）被正确识别，未被误修复。"]}, "proposed_resolution": {"suggestion": "修改 rum-nginx-config ConfigMap，移除 isProductPath 和 UA 双条件分支逻辑，恢复统一的 RUM 配置（sample=1、reportApiSpeed=true、reportAssetSpeed=true）。重启 frontend-rum-proxy 使配置生效。", "fix_steps": ["执行 kubectl --context=aliyun edit configmap rum-nginx-config -n seat-1，移除 isProductPath 和 ua.indexOf('mozilla') 分支逻辑，恢复 sample=1、reportApiSpeed=true、reportAssetSpeed=true。", "执行 kubectl --context=aliyun rollout restart deploy/frontend-rum-proxy -n seat-1。", "等待 rollout 完成。", "验证各页面 HTML 中 RUM 参数已恢复统一。", "在 RUM 控制台确认所有维度数据恢复正常。"]}, "confidence": 0.95, "special_points": [{"point": "双条件叠加比单一条件更难诊断，必须交叉维度分析", "score": 5}, {"point": "路由条件影响 sample（数据量），UA 条件影响测速（数据质量），两者独立", "score": 4}, {"point": "必须同时检查页面维度和来源维度才能发现双条件", "score": 4}, {"point": "腾讯云 GATEWAY_DIAG_MODE=verbose 是烟雾弹，与 RUM 无关", "score": 3}], "tool_process": {"ideal_call_count": 16, "valuable_actions": ["RUM 控制台/API 按页面+来源交叉维度对比", "curl 检查不同页面 HTML 中 RUM 初始化参数", "浏览器开发者工具对比浏览器访问的 RUM 参数", "kubectl get configmap rum-nginx-config 查看注入脚本", "kubectl rollout restart 恢复服务", "修复后验证 RUM 数据恢复"]}}, "rubrics": {"positive_points_total": 102, "negative_points_total": -55, "rubrics": [{"criterion": "描述了复杂现象：RUM 同时出现页面维度和来源维度的分裂", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 RUM 证据确认商品页比首页更稀疏（必须使用 RUM）", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "通过 RUM 证据确认非浏览器来源测速数据缺失更明显（必须使用 RUM）", "points": 6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "指出单看某一维度（仅页面或仅来源）会误判", "points": 5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "检查了 rum-nginx-config 并确认存在双条件逻辑", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出商品页条件触发 sample=0.05", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出非 mozilla 来源关闭 reportApiSpeed", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出非 mozilla 来源关闭 reportAssetSpeed", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "明确指出这不是单根因故障，而是两个条件共同导致的观测分裂", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "检查了页面源码并确认 SDK 仍存在", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "解释‘SDK存在但观测异常’为何在本题成立", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "识别 tencent new-gatewayservice 变量为烟雾弹", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出双条件因果链：页面条件+来源条件→RUM多维分裂", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出修复动作：移除商品页 sample 降级", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出修复动作：移除来源条件测速关闭", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出离谱操作（如重启无关服务、修改不存在的ConfigMap、跨命名空间变更）", "points": -7, "tags": ["level:example", "axis:instruction_following"]}, {"criterion": "响应结构清晰：现象→证据→根因→修复→验证", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "模型能在多故障信号下正确拆分根因而非混因", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型在跨云背景下仍能聚焦阿里云主因并正确识别腾讯云烟雾弹", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型结论基于 RUM 多维证据与配置证据闭环", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "模型不被‘业务可用/SDK存在’表象误导", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型能识别该题独有的‘多维观测分裂’模式并完成整合诊断", "points": 5, "tags": ["level:cluster", "cluster:multi_dimensional_observability_synthesis", "axis:context_awareness"]}, {"criterion": "未使用 RUM 多维证据，仅凭配置猜测", "points": -10, "tags": ["level:example", "axis:completeness"]}, {"criterion": "将双条件问题误判为单条件根因", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "遗漏商品页 sample 参数细节或来源测速开关细节", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "被烟雾弹带偏，偏离主诊断路径", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "修复后仅验证单一维度就宣布完成", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 16 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 20 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 24 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 16}, "prompt": "运营发现 RUM 指标出现复杂分裂：部分页面有数据、部分页面稀疏；浏览器访问和脚本访问表现也不同。页面可访问、SDK存在。请定位双条件导致的 RUM 观测分裂并完成修复。"}
{"canary": "cross-cloud-latency-overlap", "tags": ["度量", "跨云一致性与容量稳定性", "跨云网络延迟", "网络延迟", "应用延迟", "hard", "multi-cloud"], "case": {"id": "cross-cloud-latency-overlap", "title": "跨云网络延迟与应用延迟叠加", "inject_script": "./challenge-19-cross-cloud-latency-overlap/inject.sh", "recover_script": "./challenge-19-cross-cloud-latency-overlap/recover.sh", "fault_phenomenon": "用户反馈整个购物流程都不太顺畅：浏览商品时偶尔卡顿，结账时经常要等好几秒才能响应。商品查询和结账操作都有问题，但 Pod 都是 Running。运维若去看 APM，会看到 checkoutservice-aliyun-direct 的调用链延迟偏高，但仅靠这条线索仍看不出完整根因。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。系统已接入 RUM 与 APM，可对照查看用户侧页面表现与服务端调用链。"}, "ideal_answer": {"fault_info": {"root_cause": "两个独立故障分布在不同云：阿里云 checkoutservice 被注入 NetworkChaos 网络延迟 1500ms，导致结账响应变慢；腾讯云 productcatalogservice 被设置为 EXTRA_LATENCY=1800ms，导致商品查询偶尔卡顿。两种不同机制叠加后，整个购物流程都不顺畅。", "faulty_clouds": ["aliyun", "tencent"], "affected_clouds": ["aliyun", "tencent"], "fault_location": {"module": "network, application runtime", "file_path": null, "function_or_config": "NetworkChaos: cross-cloud-checkout-network-delay / productcatalogservice EXTRA_LATENCY", "description": "checkoutservice 被注入 1500ms 网络延迟，productcatalogservice 被设置为 EXTRA_LATENCY=1800ms。"}}, "reasoning_process": {"observed_symptoms": ["浏览商品时偶尔卡顿，结账时经常要等好几秒才能响应。", "商品查询和结账两个操作都不顺畅，但症状不完全一样：前者更偏偶发卡顿，后者更偏稳定变慢。", "Pod 都是 Running。"], "key_evidence": [{"source": "业务访问结果", "content": "商品查询和结账操作都有问题，但两个操作的症状不完全一样。", "conclusion": "更像是不同服务上叠加了两种独立的延迟型故障，需要分别核对相关对象和配置。"}, {"source": "kubectl --context=aliyun get networkchaos cross-cloud-checkout-network-delay -n seat-1 -o yaml", "content": "action=delay, latency=1500ms。", "conclusion": "checkoutservice 被注入网络延迟，这是结账变慢的原因。"}, {"source": "kubectl --context=tencent get deploy productcatalogservice -n seat-1 -o jsonpath='{.spec.template.spec.containers[0].env}'", "content": "EXTRA_LATENCY=1800ms。", "conclusion": "productcatalogservice 每次请求都会额外延迟 1800ms，这是商品查询卡顿的原因。"}], "causal_chain": ["checkoutservice NetworkChaos 1500ms → 网络延迟 → 结账响应变慢 → 用户等待时间增加。", "productcatalogservice EXTRA_LATENCY=1800ms → 商品查询额外延迟 → 商品页偶发加载慢。", "两个故障分布在不同云，需要跨云排查才能定位完整根因。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=aliyun get networkchaos cross-cloud-checkout-network-delay -n seat-1 -o yaml", "purpose": "确认 checkoutservice 的网络延迟对象与参数。", "expected_result": "action=delay, latency=1500ms, correlation=70, jitter=300ms。"}, {"cmd": "kubectl --context=tencent get deploy productcatalogservice -n seat-1 -o jsonpath='{.spec.template.spec.containers[0].env}'", "purpose": "确认 productcatalogservice 的额外延迟配置。", "expected_result": "EXTRA_LATENCY=1800ms。"}, {"cmd": "kubectl --context=tencent rollout status deploy/productcatalogservice -n seat-1 --timeout=120s", "purpose": "确认额外延迟配置已实际下发到 Pod。", "expected_result": "rollout successfully completed。"}, {"cmd": "kubectl --context=aliyun get pods -n seat-1 && kubectl --context=tencent get pods -n seat-1", "purpose": "确认不是 Pod 宕机。", "expected_result": "相关 Pod 仍为 Running。"}], "success_criteria": ["NetworkChaos 已被删除。", "productcatalogservice 的 EXTRA_LATENCY 已移除。", "商品查询和结账操作恢复正常。", "Pod 保持 Running 且不再出现随机超时。"]}, "proposed_resolution": {"suggestion": "分别删除阿里云 checkoutservice 的 NetworkChaos 和移除腾讯云 productcatalogservice 的 EXTRA_LATENCY，注意两个故障分布在不同云，需要跨云操作。", "fix_steps": ["执行 kubectl --context=aliyun delete networkchaos cross-cloud-checkout-network-delay -n seat-1。", "执行 kubectl --context=tencent set env deploy/productcatalogservice EXTRA_LATENCY- -n seat-1。", "执行 kubectl --context=tencent rollout status deploy/productcatalogservice -n seat-1 --timeout=120s。", "复查阿里云的 Chaos 对象、腾讯云 productcatalogservice 配置以及商品/结账访问结果。"]}, "confidence": 0.91, "special_points": [{"point": "网络延迟和应用层额外延迟是两种不同机制，分布在不同云，需要分别识别", "score": 5}, {"point": "跨云故障需要跨云排查，不能只看单一云", "score": 4}, {"point": "Pod Running 不能排除网络延迟和应用层延迟", "score": 3}], "tool_process": {"ideal_call_count": 13, "valuable_actions": ["对比商品浏览和结账两个链路的业务表现", "查 aliyun 的 NetworkChaos 列表和详情", "查 tencent productcatalogservice 的 Deployment 环境变量", "确认 checkoutservice 网络延迟参数和 productcatalogservice EXTRA_LATENCY", "确认 Pod 仍为 Running", "分别清理 Chaos 对象和延迟配置"]}}, "rubrics": {"positive_points_total": 98, "negative_points_total": -46, "rubrics": [{"criterion": "描述了现象：结账链路和商品查询均出现慢请求，且跨云分布", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "收集了阿里云 checkoutservice 的 chaos 证据", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "收集了腾讯云 productcatalogservice 的配置证据", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了两云服务状态与对象是否同时存在故障", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别 checkoutservice 存在网络时延故障", "points": 6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 NetworkChaos=1500ms（阿里云 checkoutservice）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别 productcatalogservice 存在额外时延故障", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 EXTRA_LATENCY=1800ms（腾讯云 productcatalogservice）", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "能区分两个故障位于不同云且相互独立", "points": 5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "故障链路完整：checkout 网络时延影响结账；catalog 额外时延影响商品查询", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出修复动作：删除阿里云 checkout NetworkChaos", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出修复动作：移除腾讯云 EXTRA_LATENCY", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "正常性约束：未跨云误删无关 chaos 对象", "points": 3, "tags": ["level:example", "axis:instruction_following"]}, {"criterion": "正常性约束：未误改无关 deployment 配置", "points": 3, "tags": ["level:example", "axis:instruction_following"]}, {"criterion": "响应结构清晰：现象→证据→根因→修复→验证", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "只检查单一云就给出全局结论", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "将双故障错误归并为单根因", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只修复 NetworkChaos，遗漏 EXTRA_LATENCY", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只移除 EXTRA_LATENCY，遗漏 NetworkChaos", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未给出恢复验证命令与预期结果就宣称修复完成", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出离谱操作（如跨命名空间批量删除资源）", "points": -6, "tags": ["level:example", "axis:instruction_following"]}, {"criterion": "模型能在跨云双故障并存时逐一隔离根因", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型排查覆盖阿里与腾讯并能定位对应对象", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型以对象状态、配置与访问表现做证据闭环", "points": 6, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "模型不被单点线索误导，保持双线并行验证", "points": 6, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 13 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 16 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 19 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 13}, "prompt": "用户反馈整个购物流程都不太顺畅：浏览商品时偶尔卡顿，结账时经常要等好几秒才能响应。商品查询和结账操作都有问题，但 Pod 都是 Running。运维若去看 APM，会看到 checkoutservice-aliyun-direct 的调用链延迟偏高，但仅靠这条线索仍看不出完整根因。"}
{"canary": "cascading-timeout-smoke", "tags": ["度量", "跨云一致性与容量稳定性", "网络延迟", "超时重试", "烟雾弹", "hard", "multi-cloud"], "case": {"id": "cascading-timeout-smoke", "title": "级联超时与重试放大", "inject_script": "./challenge-20-cascading-timeout-smoke/inject.sh", "recover_script": "./challenge-20-cascading-timeout-smoke/recover.sh", "fault_phenomenon": "顾客侧反馈最近在商城上下单时体验很差：前面浏览商品、加入购物车都正常，但一到最后结账提交订单，页面就会卡很久，随后要么直接超时，要么返回 500。不是每次都这样，但走到下单这一步就比较容易撞上。有时候刷新重试反而更慢，像是越重试越卡。\n\n运维侧检查发现前端本身没有明显异常，但 checkoutservice-aliyun-direct 和下游 paymentservice-aws-direct 的超时告警在持续触发。在 APM 里看链路时，checkoutservice-aliyun-direct 调用 paymentservice-aws-direct 的响应时间波动很大，经常超时，且超时后的重试进一步加剧了下游压力。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。系统已接入 RUM 与 APM，可对照查看用户侧页面表现与服务端调用链。"}, "ideal_answer": {"fault_info": {"root_cause": "AWS 的 paymentservice 先被注入 2000ms 网络延迟，形成结账链路上的第一个主要慢点；阿里云 checkoutservice 又被注入 500ms 网络延迟，进一步放大超时窗口。同时 checkoutservice 的 CHECKOUT_TIMEOUT_MS 被设为 2000ms，与 payment 的延迟几乎持平，导致超时后触发重试，重试请求进一步打满 payment，形成级联恶化。三层叠加：根因慢点 + 次生放大 + 超时重试。", "faulty_clouds": ["aliyun", "aws"], "affected_clouds": ["aliyun", "aws"], "fault_location": {"module": "paymentservice, checkoutservice, network, config", "file_path": null, "function_or_config": "paymentservice NetworkChaos / checkoutservice NetworkChaos / checkoutservice CHECKOUT_TIMEOUT_MS", "description": "paymentservice 2000ms 延迟制造首个慢点，checkoutservice 500ms 延迟制造次生放大，CHECKOUT_TIMEOUT_MS=2000 制造超时重试死循环。"}}, "reasoning_process": {"observed_symptoms": ["阿里云前端的结账阶段明显变慢，偶发超时或 500。", "APM 中 checkoutservice-aliyun-direct 与 paymentservice-aws-direct 都会冒烟，但需要区分谁先慢。", "刷新重试有时更慢，像是越重试越卡。"], "key_evidence": [{"source": "APM trace", "content": "paymentservice-aws-direct 的 span 先出现明显长尾，随后 checkoutservice-aliyun-direct 自身耗时也被拉高。", "conclusion": "payment 是第一个主要慢点，checkout 是放大项。"}, {"source": "kubectl --context=aws get networkchaos cascading-payment-delay -o yaml", "content": "paymentservice 存在 2000ms 延迟注入。", "conclusion": "AWS paymentservice 是首个明确故障点。"}, {"source": "kubectl --context=aliyun get networkchaos cascading-checkout-delay -o yaml", "content": "checkoutservice 存在 500ms 延迟注入。", "conclusion": "checkoutservice 额外承担了次生放大。"}, {"source": "kubectl --context=aliyun get deploy checkoutservice -o jsonpath", "content": "CHECKOUT_TIMEOUT_MS=2000。", "conclusion": "checkoutservice 超时阈值与 payment 延迟持平，会触发超时重试放大。"}], "causal_chain": ["paymentservice NetworkChaos 2000ms → 支付请求先慢下来。", "checkoutservice 自身额外 500ms 网络延迟 → 总延迟接近 2500ms。", "CHECKOUT_TIMEOUT_MS=2000 → 触发超时 → 重试请求打满 payment → 级联恶化。", "前端看到的是结账全面冒烟，但根因是 payment 的首个慢点加上超时重试放大。"]}, "verification_method": {"verification_commands": [{"cmd": "查看 APM 中 paymentservice-aws-direct 与 checkoutservice-aliyun-direct span 的先后和耗时变化", "purpose": "确认谁是第一个主要慢点。", "expected_result": "paymentservice-aws-direct 先慢，checkoutservice-aliyun-direct 后被放大。"}, {"cmd": "kubectl --context=aws get networkchaos cascading-payment-delay -n seat-1 -o yaml", "purpose": "确认 paymentservice 的延迟注入对象和参数。", "expected_result": "delay=2000ms, correlation=70, jitter=500ms。"}, {"cmd": "kubectl --context=aliyun get networkchaos cascading-checkout-delay -n seat-1 -o yaml", "purpose": "确认 checkoutservice 的放大项注入对象和参数。", "expected_result": "delay=500ms, correlation=50, jitter=150ms。"}, {"cmd": "kubectl --context=aliyun get deploy checkoutservice -n seat-1 -o jsonpath='{.spec.template.spec.containers[0].env}'", "purpose": "检查 checkoutservice 的超时配置。", "expected_result": "CHECKOUT_TIMEOUT_MS=2000。"}], "success_criteria": ["AWS paymentservice 的 NetworkChaos 已删除。", "阿里云 checkoutservice 的 NetworkChaos 已删除。", "checkoutservice 的 CHECKOUT_TIMEOUT_MS 已移除或恢复合理值。", "结账链路恢复稳定，APM 中首个慢点和次生放大都消失。"]}, "proposed_resolution": {"suggestion": "先消除 paymentservice 的首个慢点，再清理 checkoutservice 的放大项和超时配置，别把级联烟雾看成多个并列根因。", "fix_steps": ["执行 kubectl --context=aws delete networkchaos cascading-payment-delay -n seat-1。", "执行 kubectl --context=aliyun delete networkchaos cascading-checkout-delay -n seat-1。", "执行 kubectl --context=aliyun set env deploy/checkoutservice CHECKOUT_TIMEOUT_MS- -n seat-1。", "复查 APM 链路和结账访问结果。"]}, "confidence": 0.92, "special_points": [{"point": "要用 span 先后顺序找第一个慢点", "score": 5}, {"point": "checkout 冒烟是次生放大，不应与 payment 并列为第一根因", "score": 4}, {"point": "超时配置与下游延迟持平会制造重试死循环，这是级联恶化的关键机制", "score": 4}, {"point": "三处故障都要清，但排查顺序不能乱", "score": 3}], "tool_process": {"ideal_call_count": 14, "valuable_actions": ["看 APM 中 paymentservice-aws-direct 与 checkoutservice-aliyun-direct 的先后关系", "查 AWS paymentservice 的 NetworkChaos", "查 aliyun checkoutservice 的 NetworkChaos", "查 checkoutservice 的超时配置", "分别清理首个慢点、放大项和超时配置"]}}, "rubrics": {"positive_points_total": 96, "negative_points_total": -48, "rubrics": [{"criterion": "描述了现象：结账链路出现超时/重试放大，用户侧明显卡顿或失败", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "收集了 AWS paymentservice 的 NetworkChaos 证据", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "收集了阿里云 checkoutservice 的 NetworkChaos 与超时配置证据", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "收集了调用链证据，确认先慢点在 payment，再放大到 checkout", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别 paymentservice 存在网络时延故障", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 payment NetworkChaos=2000ms（AWS）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别 checkoutservice 存在网络时延故障", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 checkout NetworkChaos=500ms（阿里云）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别 checkoutservice 超时配置过紧", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 CHECKOUT_TIMEOUT_MS=2000（阿里云）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "能区分主因慢点与次生放大链路，不把所有症状归给单点", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "故障链路完整：payment慢→checkout等待/超时→重试放大→级联恶化", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出修复动作：删除 AWS payment NetworkChaos", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出修复动作：删除阿里 checkout NetworkChaos", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出修复动作：移除 CHECKOUT_TIMEOUT_MS 异常配置", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "正常性约束：未误将次生症状当主因进行离谱修改", "points": 3, "tags": ["level:example", "axis:instruction_following"]}, {"criterion": "正常性约束：未对无关云或无关服务执行破坏性动作", "points": 3, "tags": ["level:example", "axis:instruction_following"]}, {"criterion": "响应结构清晰：现象→证据→根因→修复→验证", "points": 4, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "将三故障错误归并为单一根因", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只清理一种 chaos 或只改超时配置，未完成全量恢复", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未核验主因与次因先后关系就下结论", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "遗漏跨云检查导致恢复不完整", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未给出恢复验证命令与预期结果就宣称修复完成", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出离谱操作（如大范围重启/删除无关资源）", "points": -6, "tags": ["level:example", "axis:instruction_following"]}, {"criterion": "模型能在多故障级联中识别主因与次因并分层处置", "points": 5, "tags": ["level:cluster", "cluster:cascading_root_cause_separation", "axis:accuracy"]}, {"criterion": "模型排查覆盖 AWS 与阿里并准确定位对应故障对象", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型基于链路、配置、chaos 对象进行证据闭环", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "模型不被次生告警误导，能回溯首个慢点", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 16 次）", "points": 3, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 20 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 24 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 16}, "prompt": "顾客侧反馈最近在商城上下单时体验很差：前面浏览商品、加入购物车都正常，但一到最后结账提交订单，页面就会卡很久，随后要么直接超时，要么返回 500。不是每次都这样，但走到下单这一步就比较容易撞上。有时候刷新重试反而更慢，像是越重试越卡。\n\n运维侧检查发现前端本身没有明显异常，但 checkoutservice-aliyun-direct 和下游 paymentservice-aws-direct 的超时告警在持续触发。在 APM 里看链路时，checkoutservice-aliyun-direct 调用 paymentservice-aws-direct 的响应时间波动很大，经常超时，且超时后的重试进一步加剧了下游压力。"}
{"canary": "mysql-slow-query", "tags": ["度量", "中间件与依赖", "慢查询", "连接池耗尽", "网络延迟", "medium", "multi-cloud"], "case": {"id": "mysql-slow-query", "title": "数据库链路延迟导致结账超时", "inject_script": "./challenge-21-mysql-slow-query/inject.sh", "recover_script": "./challenge-21-mysql-slow-query/recover.sh", "fault_phenomenon": "从今天上午开始，结账偶尔会卡很久才返回结果，有时直接报500错误。商品浏览和购物车操作都正常，只有结账时出问题。不是每次都失败，大概30%的请求会超时。刷新页面没用，问题一直在。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "两个故障叠加：(1) 阿里云 checkoutservice 到 mysql-orders 被注入 NetworkChaos 延迟（latency=120ms, jitter=20ms, correlation=50）；(2) 阿里云 checkoutservice 的 DB_MAX_CONNECTIONS 被设为 5。数据库链路变慢叠加连接池收紧，导致并发结账请求排队超时。", "faulty_clouds": ["aliyun"], "affected_clouds": ["aliyun"], "fault_location": {"module": "checkoutservice, mysql-orders, network", "file_path": null, "function_or_config": "NetworkChaos:mysql-orders-slow-net + DB_MAX_CONNECTIONS=5", "description": "阿里云 checkoutservice 到 mysql-orders 注入网络延迟，同时 checkoutservice 连接池被收紧到 5。"}}, "reasoning_process": {"observed_symptoms": ["结账请求变慢且出现超时失败（概率性）。", "商品浏览与购物车操作基本正常。", "pod 大多为 Running。"], "key_evidence": [{"source": "kubectl get networkchaos -n seat-1 --context=aliyun", "content": "存在 mysql-orders-slow-net，action=delay, latency=120ms", "conclusion": "checkoutservice->mysql-orders 链路被注入延迟"}, {"source": "kubectl get deploy checkoutservice -o yaml --context=aliyun", "content": "DB_MAX_CONNECTIONS=5", "conclusion": "连接池限制过小"}, {"source": "kubectl logs deploy/checkoutservice --context=aliyun", "content": "出现 connection pool exhausted / timeout", "conclusion": "并发下请求排队超时"}], "causal_chain": ["mysql 链路延迟上升 → 单次结账占用连接更久", "DB_MAX_CONNECTIONS=5 → 并发下连接池耗尽", "两者叠加 → 结账出现明显超时失败"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=aliyun get networkchaos mysql-orders-slow-net -n seat-1", "purpose": "确认延迟实验存在", "expected_result": "action=delay 且 latency=120ms"}, {"cmd": "kubectl --context=aliyun get deploy checkoutservice -n seat-1 -o jsonpath='{.spec.template.spec.containers[0].env}'", "purpose": "确认连接池限制", "expected_result": "DB_MAX_CONNECTIONS=5"}, {"cmd": "kubectl --context=aliyun logs deploy/checkoutservice -n seat-1 --tail=50", "purpose": "确认超时/连接池错误", "expected_result": "出现连接池相关超时日志"}], "success_criteria": ["NetworkChaos mysql-orders-slow-net 被删除", "DB_MAX_CONNECTIONS 限制被移除", "结账延迟与超时恢复正常"]}, "proposed_resolution": {"suggestion": "删除 mysql 链路延迟实验并移除 checkoutservice 的 DB_MAX_CONNECTIONS 限制。", "fix_steps": ["kubectl --context=aliyun delete networkchaos mysql-orders-slow-net -n seat-1", "kubectl --context=aliyun set env deploy/checkoutservice DB_MAX_CONNECTIONS- -n seat-1", "等待 rollout 完成并验证结账恢复"]}, "confidence": 0.93, "tool_process": {"ideal_call_count": 12, "valuable_actions": ["get networkchaos 查看延迟注入", "get deploy checkoutservice 检查 DB_MAX_CONNECTIONS", "logs checkoutservice 观察连接池超时", "delete networkchaos + set env 回收配置"]}}, "rubrics": {"positive_points_total": 96, "negative_points_total": -45, "rubrics": [{"criterion": "描述了结账慢/超时、但浏览与购物车基本正常的现象", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 alicloud 的 checkoutservice 日志", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 alicloud 的 NetworkChaos 实验", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 checkoutservice 环境变量", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "确认 mysql-orders-slow-net 为 delay 且 latency=120ms", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "确认 DB_MAX_CONNECTIONS=5（alicloud）", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别根因1：checkoutservice→mysql-orders 链路延迟注入", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别根因2：checkoutservice 连接池限制过小", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "能说明两故障叠加导致并发排队超时（故障链路）", "points": 6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出删除 mysql-orders-slow-net 的修复命令", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出移除 DB_MAX_CONNECTIONS 的修复命令", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应结构清晰：现象→证据→根因→修复→验证", "points": 10, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "过程正常性：不把 pod Running 当作“无故障”结论", "points": 5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "模型能正确处理双故障叠加而非只停在单点", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型排查覆盖跨对象证据（chaos + deployment/env + logs）", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:completeness"]}, {"criterion": "模型能在 alicloud 正确定位到 mysql 链路层故障", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:accuracy"]}, {"criterion": "误判为 Redis 或前端问题", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 NetworkChaos 就下根因结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "遗漏 DB_MAX_CONNECTIONS 配置故障", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只建议重启服务，不回收 chaos/配置", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "定位到错误云（非 alicloud）", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "把概率性超时说成 100% 确定性失败", "points": -4, "tags": ["level:example", "axis:normality"]}, {"criterion": "模型能够识别‘链路变慢+连接池收紧’属于中间件容量放大型复合故障", "points": 5, "tags": ["level:cluster", "cluster:middleware_capacity_coupling", "axis:context_awareness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 12 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 15 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 18 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 12}, "prompt": "从今天上午开始，结账偶尔会卡很久才返回结果，有时直接报500错误。商品浏览和购物车操作都正常，只有结账时出问题。不是每次都失败，大概30%的请求会超时。刷新页面没用，问题一直在。"}
{"canary": "redis-cache-stampede", "tags": ["度量", "中间件与依赖", "缓存雪崩", "Redis延迟", "缓存回源", "medium", "multi-cloud"], "case": {"id": "redis-cache-stampede", "title": "Redis缓存层性能退化导致购物车变慢", "inject_script": "./challenge-22-redis-cache-stampede/inject.sh", "recover_script": "./challenge-22-redis-cache-stampede/recover.sh", "fault_phenomenon": "今天开始，购物车操作变得特别慢：加商品、查看购物车、结账都要等好几秒。商品浏览偶尔也会变慢（价格显示不出来），但大部分时间正常。刷新没用，问题一直在。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "两个故障叠加：(1) 腾讯云 cartservice 到 redis-cart 被注入 NetworkChaos 延迟（latency=150ms, jitter=30ms, correlation=50）；(2) cartservice 的 CACHE_TTL 被设为 1 秒。缓存链路变慢叠加 TTL 过短，导致频繁回源和购物车操作持续变慢。", "faulty_clouds": ["tencent"], "affected_clouds": ["tencent", "aliyun"], "fault_location": {"module": "cartservice, redis-cart, network", "file_path": null, "function_or_config": "NetworkChaos:redis-cart-latency + CACHE_TTL=1s", "description": "腾讯云 cartservice->redis-cart 注入延迟，同时缓存 TTL 被压缩至 1 秒。"}}, "reasoning_process": {"observed_symptoms": ["购物车操作明显变慢，结账受影响。", "商品浏览大多正常但偶发变慢。", "pod 状态多数 Running。"], "key_evidence": [{"source": "kubectl get networkchaos -n seat-1 --context=tencent", "content": "存在 redis-cart-latency，action=delay, latency=150ms", "conclusion": "redis 通路存在人为延迟"}, {"source": "kubectl get deploy cartservice -o yaml --context=tencent", "content": "CACHE_TTL=1s", "conclusion": "缓存过期过快"}, {"source": "kubectl logs deploy/cartservice --context=tencent", "content": "大量 cache miss / fetching from backend", "conclusion": "命中率下降并持续回源"}], "causal_chain": ["redis 通路延迟升高 → 读写缓存变慢", "CACHE_TTL=1s → 缓存频繁失效", "两者叠加 → 回源压力放大，购物车持续变慢"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=tencent get networkchaos redis-cart-latency -n seat-1", "purpose": "确认延迟实验存在", "expected_result": "action=delay 且 latency=150ms"}, {"cmd": "kubectl --context=tencent get deploy cartservice -n seat-1 -o jsonpath='{.spec.template.spec.containers[0].env}'", "purpose": "确认 TTL 被改短", "expected_result": "CACHE_TTL=1s"}, {"cmd": "kubectl --context=tencent logs deploy/cartservice -n seat-1 --tail=50", "purpose": "确认缓存 miss 回源", "expected_result": "出现 cache miss/fetching from backend"}], "success_criteria": ["redis-cart-latency 被删除", "CACHE_TTL 限制被移除", "购物车响应恢复正常"]}, "proposed_resolution": {"suggestion": "删除 redis 延迟实验并恢复默认缓存 TTL 策略。", "fix_steps": ["kubectl --context=tencent delete networkchaos redis-cart-latency -n seat-1", "kubectl --context=tencent set env deploy/cartservice CACHE_TTL- -n seat-1", "等待 rollout 并验证恢复"]}, "confidence": 0.94, "tool_process": {"ideal_call_count": 12, "valuable_actions": ["get networkchaos 查看 redis 延迟注入", "get deploy cartservice 检查 CACHE_TTL", "logs cartservice 观察 cache miss", "delete chaos + set env 回收"]}}, "rubrics": {"positive_points_total": 96, "negative_points_total": -45, "rubrics": [{"criterion": "描述了购物车慢、结账受影响但非全站瘫痪", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 tencent 的 cartservice 日志", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 tencent 的 NetworkChaos 实验", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 cartservice 环境变量", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "确认 redis-cart-latency 为 delay 且 latency=150ms", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "确认 CACHE_TTL=1s（tencent）", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别根因1：cartservice→redis-cart 链路延迟", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别根因2：CACHE_TTL 过短导致频繁回源", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "能说明两故障叠加放大缓存 miss 与延迟（故障链路）", "points": 6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出删除 redis-cart-latency 的修复命令", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出移除 CACHE_TTL 的修复命令", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应结构清晰：现象→证据→根因→修复→验证", "points": 10, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "过程正常性：不把单次变慢当全链路崩溃", "points": 5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "模型能区分缓存层/链路层叠加而非单一应用 bug", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型在腾讯云正确收敛到 redis 通路问题", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:accuracy"]}, {"criterion": "模型基于 chaos+env+log 证据闭环定位", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:completeness"]}, {"criterion": "误判为 DB/前端故障", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未查 NetworkChaos 就给结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "遗漏 CACHE_TTL=1s", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只重启服务，不删除 chaos/不回收配置", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "定位到错误云（非 tencent）", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "把“持续变慢”描述成“完全不可用”", "points": -4, "tags": ["level:example", "axis:normality"]}, {"criterion": "模型能够识别‘缓存通路延迟+TTL过短’会共同放大回源压力的中间件复合故障", "points": 5, "tags": ["level:cluster", "cluster:cache_pressure_amplification", "axis:context_awareness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 12 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 15 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 18 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 12}, "prompt": "今天开始，购物车操作变得特别慢：加商品、查看购物车、结账都要等好几秒。商品浏览偶尔也会变慢（价格显示不出来），但大部分时间正常。刷新没用，问题一直在。"}
{"canary": "cart-network-loss", "tags": ["发现", "网络与协议", "网络丢包", "连接超时", "medium", "multi-cloud"], "case": {"id": "cart-network-loss", "title": "购物车网络丢包", "inject_script": "./challenge-23-cart-network-loss/inject.sh", "recover_script": "./challenge-23-cart-network-loss/recover.sh", "fault_phenomenon": "购物车功能变得很不稳定：有时候加购物车成功，有时候失败；购物车页面有时候是空的，刷新一下又有了；结账时偶尔提示购物车内容异常。不是每次都坏，时好时坏，概率性的。商品浏览和其他功能正常。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "两个故障叠加：(1) 腾讯云 cartservice 被注入 20% 网络丢包；(2) cartservice 的 REDIS_CONNECT_TIMEOUT 被设为 200ms。丢包导致连接重试，但 200ms 超时截断了重试窗口，失败率从 ~20% 上升到 ~40%。", "faulty_clouds": ["tencent"], "affected_clouds": ["tencent", "aliyun"], "fault_location": {"module": "cartservice, network", "file_path": null, "function_or_config": "NetworkChaos: cart-network-loss", "description": "Chaos Mesh NetworkChaos 对 cartservice 注入 20% 网络丢包，影响 cartservice 与 redis-cart 之间的通信。同时 cartservice 的 REDIS_CONNECT_TIMEOUT 被收紧到 200ms，截断了丢包后的重试窗口。"}}, "reasoning_process": {"observed_symptoms": ["购物车操作频繁失败，失败率明显高于 20%。", "购物车页面偶尔为空，刷新后恢复。", "结账经常提示购物车异常。", "商品浏览正常。"], "key_evidence": [{"source": "kubectl logs cartservice", "content": "日志显示大量连接 redis-cart 超时（connection timed out after 200ms）。", "conclusion": "cartservice 与 redis-cart 之间存在网络问题，且超时设置过紧。"}, {"source": "kubectl get networkchaos -n seat-1 --context=tencent", "content": "存在 cart-network-loss 实验，action=loss, loss=20%。", "conclusion": "cartservice 网络被注入 20% 丢包。"}, {"source": "kubectl get pods -n seat-1 --context=tencent", "content": "cartservice 和 redis-cart 都 Running。", "conclusion": "pod 本身没崩，问题在网络层。"}, {"source": "kubectl get deploy cartservice -n seat-1 --context=tencent -o yaml", "content": "env 中 REDIS_CONNECT_TIMEOUT=200ms。", "conclusion": "连接超时被收紧，丢包后没有足够时间重试。"}], "causal_chain": ["NetworkChaos 注入 20% 丢包 → 部分连接失败需要重试。REDIS_CONNECT_TIMEOUT=200ms → 重试窗口被截断 → 失败率从 ~20% 上升到 ~40%。两个故障叠加放大了症状。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=tencent get networkchaos -n seat-1", "purpose": "查看 NetworkChaos 实验。", "expected_result": "存在 cart-network-loss 实验，loss=20%。"}, {"cmd": "kubectl --context=tencent logs cartservice-xxx -n seat-1 --tail=30", "purpose": "查看 cartservice 日志中的网络错误。", "expected_result": "出现连接超时或重置的错误日志。"}, {"cmd": "kubectl --context=tencent get pods -n seat-1 | grep -E 'cartservice|redis-cart'", "purpose": "确认两个 pod 都正常运行。", "expected_result": "两个 pod 都 Running。"}, {"cmd": "kubectl --context=tencent get deploy cartservice -n seat-1 -o jsonpath='{.spec.template.spec.containers[0].env}'", "purpose": "查看 cartservice 的环境变量配置。", "expected_result": "发现 REDIS_CONNECT_TIMEOUT=200ms。"}], "success_criteria": ["NetworkChaos 实验被删除。", "cartservice 的 REDIS_CONNECT_TIMEOUT 配置被恢复。", "购物车操作恢复稳定。", "cartservice 日志不再出现网络错误。"]}, "proposed_resolution": {"suggestion": "删除 NetworkChaos 实验并恢复 cartservice 的 REDIS_CONNECT_TIMEOUT 配置。两个故障必须同时修复。建议监控服务间网络质量和连接超时配置。", "fix_steps": ["执行 kubectl --context=tencent delete networkchaos cart-network-loss -n seat-1。", "执行 kubectl --context=tencent set env deploy/cartservice REDIS_CONNECT_TIMEOUT- -n seat-1 恢复超时配置。", "等待 cartservice rollout 完成。", "验证购物车操作恢复稳定。"]}, "confidence": 0.94, "special_points": [{"point": "是网络层丢包，不是 Redis 本身故障", "score": 4}, {"point": "20% 丢包导致概率性失败，时好时坏是典型特征", "score": 4}, {"point": "需要从日志中的网络错误关联到 NetworkChaos 实验", "score": 3}, {"point": "REDIS_CONNECT_TIMEOUT=200ms 截断了丢包后的重试窗口，放大了失败率", "score": 4}, {"point": "两个故障都在腾讯云但影响不同层（网络层 + 应用配置层）", "score": 3}], "tool_process": {"ideal_call_count": 13, "valuable_actions": ["kubectl get networkchaos 查看 NetworkChaos 实验", "kubectl logs cartservice 查看网络错误日志", "kubectl get pods 确认 pod 状态", "kubectl get deploy cartservice -o yaml 检查环境变量配置", "kubectl delete networkchaos 删除实验", "kubectl set env deploy/cartservice REDIS_CONNECT_TIMEOUT- 恢复超时配置并验证"]}}, "rubrics": {"positive_points_total": 103, "negative_points_total": -53, "rubrics": [{"criterion": "描述了观察到的故障现象：购物车时好时坏，概率性失败，商品浏览正常", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "验证了故障主要集中在购物车路径，商品浏览和其他功能不受影响", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 cartservice 日志，确认间歇性网络连接错误", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 kubectl get networkchaos 查看了三云的 NetworkChaos 实验", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 cartservice 和 redis-cart pod 状态，确认两个 pod 都正常运行", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "提供了来自至少两朵不同云的直接证据（如腾讯云 NetworkChaos + 阿里云日志）", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 cartservice 的环境变量配置，发现 REDIS_CONNECT_TIMEOUT 被设为不合理的 200ms", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 cartservice 存在网络丢包问题", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 NetworkChaos 的具体参数（loss 20%）和部署云（腾讯云）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 cartservice 的 REDIS_CONNECT_TIMEOUT 错误值（200ms）和部署云（腾讯云）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出故障发生在腾讯云 cartservice 的网络层，而不是 Redis 本身故障", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：NetworkChaos 注入丢包 → cartservice 与 redis-cart 通信间歇失败 → 购物车时好时坏", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解 20% 丢包导致概率性失败，'时好时坏'是网络丢包的典型特征", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "理解需要从日志中的网络错误关联到 NetworkChaos 实验，而不是只看 pod 状态", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "理解 Pod 处于 Running 但网络层被注入丢包时，业务仍会受影响", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "理解 20% 丢包 + 200ms 连接超时的叠加效应：丢包导致重试，但 200ms 超时截断了重试窗口，失败率显著上升", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出了具体可执行的修复步骤：删除腾讯云 NetworkChaos 实验", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体可执行的修复命令：kubectl --context=tencent delete networkchaos cart-network-loss -n seat-1", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复步骤：恢复 cartservice 的 REDIS_CONNECT_TIMEOUT 配置", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复步骤：等待网络通信恢复正常", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了合理的后续建议（如监控服务间网络质量、配置网络丢包告警）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "模型能够从概率性购物车失败收敛到网络丢包单一根因", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型的排查过程覆盖了腾讯云的 cartservice 和阿里云的 checkoutservice", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型的排查过程基于实际证据（NetworkChaos 实验、日志、pod 状态），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "模型能够识别概率性故障的本质：网络丢包导致间歇性失败，不能指望确定性复现", "points": 5, "tags": ["level:cluster", "cluster:probabilistic_fault_understanding", "axis:context_awareness"]}, {"criterion": "把问题归因成 redis-cart 宕机，实际是网络丢包", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "误判为应用 Bug 或配置错误，未检查 NetworkChaos", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "没有检查 NetworkChaos 实验就下结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只查看阿里云入口，遗漏了腾讯云 cartservice 的 NetworkChaos 证据", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只建议重启 cartservice pod，而没有删除 NetworkChaos 实验", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "定位到错误云集群（如把 NetworkChaos 定位到阿里云而非腾讯云）", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只修复了 NetworkChaos 但遗漏了 cartservice 的 REDIS_CONNECT_TIMEOUT 配置错误", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 13 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 16 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 19 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 13}, "prompt": "购物车功能变得很不稳定：有时候加购物车成功，有时候失败；购物车页面有时候是空的，刷新一下又有了；结账时偶尔提示购物车内容异常。不是每次都坏，时好时坏，概率性的。商品浏览和其他功能正常。"}
{"canary": "config-drift-cascade", "tags": ["定位", "配置与密钥", "环境变量写错", "级联故障", "hard", "multi-cloud"], "case": {"id": "config-drift-cascade", "title": "配置漂移导致订单链路级联失败", "inject_script": "./challenge-24-config-drift-cascade/inject.sh", "recover_script": "./challenge-24-config-drift-cascade/recover.sh", "fault_phenomenon": "今天开始，结账完全失败。选好商品点结账，页面一直转圈最后报错说下单失败。但奇怪的是，浏览商品没问题，加购物车也正常，购物车里的东西也能看到。就是最后一步结账不行。试了好多次都不行，不是偶发的。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "两处环境变量配置漂移导致订单链路级联失败：(1) 阿里云 checkoutservice 的 CART_SERVICE_ADDR 被改为 43.139.187.221:30099（正确值应为 30007）；(2) AWS new-orderservice 的 CART_SERVICE_ADDR 被改为 43.139.187.221:30099（正确值应为 30007）。两处错误配置指向同一个不存在的端口，导致 checkoutservice 和 new-orderservice 均无法连接 cartservice，结账链路完全中断。", "faulty_clouds": ["unknown"], "affected_clouds": ["aliyun", "aws", "tencent"], "fault_location": {"module": "checkoutservice, new-orderservice", "file_path": null, "function_or_config": "CART_SERVICE_ADDR=43.139.187.221:30099", "description": "阿里云 checkoutservice 和 AWS new-orderservice 的 CART_SERVICE_ADDR 环境变量被修改为错误端口 30099。cartservice 实际运行在腾讯云 43.139.187.221:30007，端口 30099 无任何服务监听，导致 TCP 连接被拒绝。"}}, "reasoning_process": {"observed_symptoms": ["结账操作100%失败，页面报错下单失败。", "商品浏览正常。", "购物车加购、查看均正常。", "所有 pod 状态为 Running，无异常重启。"], "key_evidence": [{"source": "kubectl logs checkoutservice (阿里云)", "content": "日志显示 dial tcp 43.139.187.221:30099: connect: connection refused。", "conclusion": "checkoutservice 无法连接 cartservice，端口 30099 无服务监听。"}, {"source": "kubectl logs new-orderservice (AWS)", "content": "日志显示 dial tcp 43.139.187.221:30099: connect: connection refused。", "conclusion": "new-orderservice 同样无法连接 cartservice，两处错误指向同一端口。"}, {"source": "kubectl get svc cartservice (腾讯云)", "content": "cartservice NodePort 端口为 30007。", "conclusion": "cartservice 实际端口是 30007，不是 30099。"}, {"source": "kubectl get deploy checkoutservice -o yaml (阿里云)", "content": "env 中 CART_SERVICE_ADDR=43.139.187.221:30099。", "conclusion": "checkoutservice 的 cartservice 地址被改为错误端口。"}, {"source": "kubectl get deploy new-orderservice -o yaml (AWS)", "content": "env 中 CART_SERVICE_ADDR=43.139.187.221:30099。", "conclusion": "new-orderservice 的 cartservice 地址也被改为错误端口，两处配置漂移。"}], "causal_chain": ["checkoutservice(阿里云) CART_SERVICE_ADDR 被改为 30099 + new-orderservice(AWS) CART_SERVICE_ADDR 被改为 30099 → 两个服务均无法连接腾讯云 cartservice(实际端口 30007) → 结账和下单两条链路同时失败 → 用户看到结账100%失败。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=aliyun logs deploy/checkoutservice -n seat-1 --tail=30", "purpose": "查看 checkoutservice 日志中的连接错误。", "expected_result": "出现 dial tcp 43.139.187.221:30099: connect: connection refused。"}, {"cmd": "kubectl --context=aws logs deploy/new-orderservice -n seat-1 --tail=30", "purpose": "查看 new-orderservice 日志中的连接错误。", "expected_result": "出现 dial tcp 43.139.187.221:30099: connect: connection refused。"}, {"cmd": "kubectl --context=tencent get svc cartservice -n seat-1", "purpose": "确认 cartservice 实际端口。", "expected_result": "NodePort 端口为 30007，不是 30099。"}, {"cmd": "kubectl --context=aliyun get deploy checkoutservice -n seat-1 -o jsonpath='{.spec.template.spec.containers[0].env}'", "purpose": "检查 checkoutservice 的环境变量配置。", "expected_result": "发现 CART_SERVICE_ADDR=43.139.187.221:30099。"}, {"cmd": "kubectl --context=aws get deploy new-orderservice -n seat-1 -o jsonpath='{.spec.template.spec.containers[0].env}'", "purpose": "检查 new-orderservice 的环境变量配置。", "expected_result": "发现 CART_SERVICE_ADDR=43.139.187.221:30099。"}], "success_criteria": ["checkoutservice 的 CART_SERVICE_ADDR 恢复为 43.139.187.221:30007。", "new-orderservice 的 CART_SERVICE_ADDR 恢复为 43.139.187.221:30007。", "结账流程恢复正常。", "日志中不再出现 connection refused 错误。"]}, "proposed_resolution": {"suggestion": "将 checkoutservice 和 new-orderservice 的 CART_SERVICE_ADDR 均恢复为正确端口 30007。两处配置必须同时修复，只修一个只能恢复一条链路。建议对跨云服务地址配置做定期一致性检查。", "fix_steps": ["执行 kubectl --context=aliyun set env deploy/checkoutservice CART_SERVICE_ADDR=43.139.187.221:30007 -n seat-1。", "执行 kubectl --context=aws set env deploy/new-orderservice CART_SERVICE_ADDR=43.139.187.221:30007 -n seat-1。", "等待两个服务 rollout 完成。", "验证结账和下单功能恢复正常。"]}, "confidence": 0.96, "special_points": [{"point": "故障不在腾讯云 cartservice 本身，而是上游服务的配置指向了错误端口", "score": 4}, {"point": "两处错误配置分布在不同云（阿里云 + AWS），需要跨云排查", "score": 5}, {"point": "购物车操作正常但结账失败——cartservice 本身没坏，是调用方配置错误", "score": 4}, {"point": "100%失败而非概率性——端口 30099 无服务监听，每次连接必被拒绝", "score": 3}, {"point": "没有使用 Chaos Mesh，是纯配置变更导致的故障，更贴近生产真实场景", "score": 4}], "tool_process": {"ideal_call_count": 12, "valuable_actions": ["kubectl logs checkoutservice 查看连接错误日志", "kubectl logs new-orderservice 查看连接错误日志", "kubectl get svc cartservice 确认实际端口", "kubectl get deploy checkoutservice -o yaml 检查环境变量", "kubectl get deploy new-orderservice -o yaml 检查环境变量", "kubectl set env deploy/checkoutservice CART_SERVICE_ADDR 恢复正确地址", "kubectl set env deploy/new-orderservice CART_SERVICE_ADDR 恢复正确地址"]}}, "rubrics": {"positive_points_total": 95, "negative_points_total": -53, "rubrics": [{"criterion": "描述了结账100%失败、浏览和购物车正常、pod Running 的现象", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 checkoutservice 日志并看到连接 30099 被拒绝", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 new-orderservice 日志并看到连接 30099 被拒绝", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 tencent cartservice 实际端口为 30007", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 alicloud checkoutservice 的 CART_SERVICE_ADDR", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "检查了 aws new-orderservice 的 CART_SERVICE_ADDR", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "准确识别两处错误都指向 43.139.187.221:30099", "points": 6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出正确值应为 43.139.187.221:30007", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别多云配置漂移：faulty_clouds=unknown，affected 覆盖 alicloud/aws/tencent", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪故障链：上游地址错端口→连接拒绝→结账链路中断", "points": 6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解“购物车正常但结账失败”是调用方配置错而非 cartservice 故障", "points": 5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出修复 checkoutservice 地址命令", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出修复 new-orderservice 地址命令", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应结构清晰：现象→证据→根因→修复→验证", "points": 12, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "模型能识别多点配置同时漂移且需同时修复", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型排查覆盖三云链路证据", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型基于日志+svc端口+env配置做证据闭环", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "误判为 cartservice 崩溃或前端 bug", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只查一处配置就下全局结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "遗漏任一错误配置（只修一边）", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "建议用 chaos 方向排查并忽略配置漂移", "points": -6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "定位到错误云或错误端口", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "把100%失败说成概率性抖动", "points": -5, "tags": ["level:example", "axis:normality"]}, {"criterion": "只重启服务不改配置", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 12 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 15 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 18 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 12}, "prompt": "今天开始，结账完全失败。选好商品点结账，页面一直转圈最后报错说下单失败。但奇怪的是，浏览商品没问题，加购物车也正常，购物车里的东西也能看到。就是最后一步结账不行。试了好多次都不行，不是偶发的。"}
{"canary": "productcatalog-cascade", "tags": ["定位", "跨云一致性与容量稳定性", "DNS 解析故障", "级联故障", "hard", "multi-cloud"], "case": {"id": "productcatalog-cascade", "title": "商品服务级联故障", "inject_script": "./challenge-25-productcatalog-cascade/inject.sh", "recover_script": "./challenge-25-productcatalog-cascade/recover.sh", "fault_phenomenon": "商品列表突然打不开了，页面直接报错说服务不可用。同时推荐内容也刷不出来，结账页面的价格显示异常。购物车本身能打开但看不到商品价格。前端页面能正常加载。刷新好多次也不管用。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，或通过终端工具检查服务日志和端到端业务表现。"}, "ideal_answer": {"fault_info": {"root_cause": "腾讯云 productcatalogservice 的 DNS 解析被注入故障（DNSChaos, action: error），导致商品服务完全不可用。单根因级联故障：productcatalogservice 是多个上游服务的依赖，DNS 故障导致商品列表、推荐、结账价格显示全部异常。", "faulty_clouds": ["tencent"], "affected_clouds": ["tencent", "aliyun"], "fault_location": {"module": "productcatalogservice, dns", "file_path": null, "function_or_config": "DNSChaos: productcatalog-cascade-dns", "description": "Chaos Mesh DNSChaos 对 productcatalogservice 注入 DNS 解析错误。productcatalogservice 被 frontend（阿里云）、checkoutservice（阿里云）、recommendationservice（阿里云）三个服务依赖，DNS 故障导致整条商品链路级联失败。"}}, "reasoning_process": {"observed_symptoms": ["商品列表打不开，报服务不可用。", "推荐内容刷不出来。", "结账页面价格显示异常。", "购物车能打开但看不到商品价格。", "前端页面本身能加载。"], "key_evidence": [{"source": "kubectl get pods -n seat-1 --context=tencent", "content": "productcatalogservice pod 状态 Running。", "conclusion": "pod 本身没崩，问题在 DNS 解析层。"}, {"source": "kubectl get dnschaos -n seat-1 --context=tencent", "content": "存在 productcatalog-cascade-dns 实验，action=error。", "conclusion": "productcatalogservice 的 DNS 解析被注入故障。"}, {"source": "kubectl logs deploy/frontend -n seat-1 --context=aliyun", "content": "日志显示连接 productcatalogservice 失败。", "conclusion": "阿里云 frontend 无法访问腾讯云的 productcatalogservice。"}, {"source": "kubectl exec -it <pod> -- nslookup productcatalogservice", "content": "nslookup 返回错误或超时。", "conclusion": "DNS 解析失败，确认是 DNS 层问题。"}], "causal_chain": ["DNSChaos 注入 productcatalogservice DNS 解析错误 → 阿里云 frontend/checkoutservice/recommendationservice 无法连接 productcatalogservice → 商品列表不可用 + 推荐失败 + 价格异常。单根因，多路级联。"]}, "verification_method": {"verification_commands": [{"cmd": "kubectl --context=tencent get dnschaos -n seat-1", "purpose": "查看 DNSChaos 实验。", "expected_result": "存在 productcatalog-cascade-dns 实验。"}, {"cmd": "kubectl --context=tencent get pods -n seat-1 | grep productcatalog", "purpose": "确认 productcatalogservice pod 本身正常。", "expected_result": "pod Running。"}, {"cmd": "kubectl --context=aliyun logs deploy/frontend -n seat-1 --tail=20", "purpose": "查看 frontend 连接 productcatalogservice 的错误日志。", "expected_result": "出现连接 productcatalogservice 失败的日志。"}, {"cmd": "kubectl --context=aliyun exec -it deploy/frontend -n seat-1 -- nslookup productcatalogservice", "purpose": "验证 DNS 解析是否失败。", "expected_result": "nslookup 返回错误或超时。"}], "success_criteria": ["DNSChaos 实验被删除。", "DNS 解析恢复正常。", "商品列表、推荐、价格显示功能恢复正常。"]}, "proposed_resolution": {"suggestion": "删除 DNSChaos 实验。单个根因修复即可完全恢复。如果 DNS 缓存导致问题持续，可等待缓存过期或重启 CoreDNS。建议配置 DNS 解析监控。", "fix_steps": ["执行 kubectl --context=tencent delete dnschaos productcatalog-cascade-dns -n seat-1。", "等待 DNS 缓存过期或重启 CoreDNS。", "验证商品列表、推荐、价格显示功能恢复正常。"]}, "confidence": 0.95, "special_points": [{"point": "单根因级联：productcatalogservice DNS 故障 → 三个上游服务同时受影响", "score": 5}, {"point": "productcatalogservice 是核心依赖，被 frontend/checkoutservice/recommendationservice 三个服务调用", "score": 5}, {"point": "pod Running 但 DNS 解析失败是典型 DNS 故障模式", "score": 4}, {"point": "需要使用 nslookup/dig 直接验证 DNS 解析", "score": 4}, {"point": "购物车能打开但看不到价格——因为 cartservice 不依赖 productcatalogservice，但 checkoutservice 依赖", "score": 4}], "tool_process": {"ideal_call_count": 13, "valuable_actions": ["kubectl get dnschaos 查看 DNSChaos 实验", "kubectl get pods 确认 productcatalogservice pod 状态", "kubectl logs frontend 查看连接错误日志", "kubectl exec nslookup 验证 DNS 解析", "kubectl delete dnschaos 删除实验并验证恢复"]}}, "rubrics": {"positive_points_total": 98, "negative_points_total": -46, "rubrics": [{"criterion": "描述了观察到的故障现象：商品列表不可用、推荐失败、价格异常，购物车能打开但无价格", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 frontend 或 checkoutservice 日志，确认连接 productcatalogservice 失败", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 kubectl get dnschaos 查看了三云的 DNSChaos 实验", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 productcatalogservice pod 状态，确认 pod 本身未崩溃", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 nslookup 或 dig 直接验证了 DNS 解析结果", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "提供了来自至少两朵不同云的直接证据（如腾讯云 DNSChaos + 阿里云 frontend 日志）", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 productcatalogservice 是多个上游服务的核心依赖，DNS 故障导致多路级联", "points": 6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 DNSChaos 的具体参数（action: error）和部署云（腾讯云）", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪级联链：productcatalogservice DNS 故障 → frontend/checkoutservice/recommendationservice 全部受影响", "points": 6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解为什么购物车能打开但看不到价格：cartservice 不依赖 productcatalogservice，但价格显示依赖", "points": 5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "理解 pod Running 但 DNS 解析失败是 DNS 故障的典型特征", "points": 5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "理解 DNS 缓存可能导致故障恢复后仍有延迟，需要等待缓存过期或重启 CoreDNS", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出了具体可执行的修复步骤：删除腾讯云 DNSChaos 实验", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了具体修复命令：kubectl --context=tencent delete dnschaos productcatalog-cascade-dns -n seat-1", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复步骤：等待 DNS 缓存过期或重启 CoreDNS", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了合理的后续建议（如配置 DNS 监控、productcatalogservice 高可用）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复步骤 → 验证", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "模型能够识别级联故障模式：核心服务 DNS 故障 → 多个上游服务同时失败", "points": 5, "tags": ["level:cluster", "cluster:cascading_failure_diagnosis", "axis:context_awareness"]}, {"criterion": "模型的排查过程覆盖了腾讯云的 productcatalogservice 和阿里云的 frontend/checkoutservice", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型的排查过程基于实际证据（DNSChaos 实验、nslookup 结果、日志），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "把问题归因成 productcatalogservice 代码 Bug 或 pod 崩溃", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "误判为网络问题或多个独立故障，而实际是单根因 DNS 级联", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "没有检查 DNSChaos 实验就下结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只查看阿里云入口，遗漏了腾讯云 DNSChaos 证据", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只建议重启 productcatalogservice pod，而没有删除 DNSChaos 实验", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "定位到错误云集群（如把 DNSChaos 定位到阿里云而非腾讯云）", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 13 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 16 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 19 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}], "ideal_call_count": 13}, "prompt": "商品列表突然打不开了，页面直接报错说服务不可用。同时推荐内容也刷不出来，结账页面的价格显示异常。购物车本身能打开但看不到商品价格。前端页面能正常加载。刷新好多次也不管用。"}
{"canary": "mysql-pool-exhaustion", "tags": ["度量", "中间件与依赖", "连接池耗尽", "慢查询", "CPU 压力", "hard", "multi-cloud"], "case": {"id": "mysql-pool-exhaustion", "title": "结账链路排队与下游退化", "inject_script": "./challenge-26-mysql-pool-exhaustion/inject.sh", "recover_script": "./challenge-26-mysql-pool-exhaustion/recover.sh", "fault_phenomenon": "高峰期结账经常卡住，有时直接 500，低峰期偶尔也会撞上。商品浏览和购物车基本正常。问题不是每次都炸，但连续刷新几次大概率会碰到，而且越刷新越慢。另外商品推荐有时候也卡一下，但不确定是不是同一个问题。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，并结合日志和端到端业务表现排查。"}, "ideal_answer": {"fault_info": {"root_cause": "三层故障叠加 + 一个干扰项：(1) NetworkChaos 在 checkoutservice -> mysql-orders 之间注入轻度网络延迟，(2) checkoutservice 的 DB_MAX_CONNECTIONS 被收紧到 8，(3) StressChaos 对 mysql-orders 施加 CPU 压力。三层叠加后，结账请求更容易在连接等待和慢查询阶段堆积，最终触发超时。另外存在一个干扰项：recommendationservice 被注入 300ms 延迟，但 recommendationservice 不走 mysql-orders，与结账超时无直接因果关系。", "faulty_clouds": ["alicloud"], "affected_clouds": ["alicloud"], "fault_location": {"module": "chaos-mesh + k8s-deployment", "file_path": null, "function_or_config": "NetworkChaos: mysql-orders-slow-net, StressChaos: mysql-cpu-stress, deploy/checkoutservice: DB_MAX_CONNECTIONS=8, NetworkChaos: recommendation-net-delay（干扰项）", "description": "网络延迟 + 连接池收紧 + CPU 压力三层叠加。干扰项 recommendationservice 延迟需要被识别为非根因。"}}, "reasoning_process": {"observed_symptoms": ["高峰期结账经常卡住或 500，低峰期偶尔也会撞上。", "商品浏览和购物车基本正常。", "连续刷新几次大概率会碰到，越刷新越慢。", "商品推荐有时候也卡一下。", "mysql-orders pod Running，无重启。"], "key_evidence": [{"source": "MCP/K8s: alicloud seat-1 get networkchaos", "content": "mysql-orders-slow-net: checkoutservice -> mysql-orders delay 100ms, jitter 20ms", "conclusion": "结账到 MySQL 的往返被拉长。"}, {"source": "MCP/K8s: alicloud seat-1 get stresschaos", "content": "mysql-cpu-stress: mysql-orders CPU 70%, 2 workers", "conclusion": "MySQL 侧处理能力下降，查询更容易堆积。"}, {"source": "MCP/K8s: alicloud seat-1 get deploy checkoutservice -o yaml", "content": "DB_MAX_CONNECTIONS=8", "conclusion": "连接池被收紧，等待更容易累积。"}, {"source": "MCP/K8s: alicloud seat-1 get pods -l app=mysql-orders", "content": "mysql-orders Running, 0 restarts", "conclusion": "不是 OOM/CrashLoop，而是性能退化。"}, {"source": "日志或 APM span", "content": "checkoutservice 出现连接等待、DB timeout 或 pool wait 变长", "conclusion": "直接证据指向连接池等待和下游慢查询叠加。"}, {"source": "MCP/K8s: alicloud seat-1 get networkchaos", "content": "recommendation-net-delay: recommendationservice delay 300ms, jitter 80ms", "conclusion": "干扰项。recommendationservice 不走 mysql-orders，与结账超时无直接因果关系，但会让商品推荐变慢。"}], "causal_chain": ["NetworkChaos 让每次 DB 往返变慢。", "StressChaos 让 MySQL 查询处理能力下降。", "DB_MAX_CONNECTIONS=8 让高峰时连接等待开始放大。", "连接等待 + 慢查询叠加后，checkout 请求堆积。", "最终部分请求超过结账超时阈值，表现为 500 或卡顿。", "商品浏览和购物车不走 mysql-orders，所以基本正常。", "干扰项：recommendationservice 延迟会影响商品推荐，但不会导致结账超时。"]}, "proposed_resolution": {"suggestion": "三层故障都要处理，干扰项可一并清理。建议恢复 checkoutservice 的连接池上限，移除 MySQL 侧 CPU 压力和网络延迟，并补连接池等待和慢查询监控。", "fix_steps": ["kubectl --context=alicloud -n seat-1 delete networkchaos mysql-orders-slow-net", "kubectl --context=alicloud -n seat-1 delete stresschaos mysql-cpu-stress", "kubectl --context=alicloud -n seat-1 set env deploy/checkoutservice DB_MAX_CONNECTIONS-", "kubectl --context=alicloud -n seat-1 delete networkchaos recommendation-net-delay", "建议恢复后观察结账成功率和超时率。"]}, "confidence": 0.92, "special_points": [{"point": "三层叠加需要理解连接等待和慢查询的放大关系", "score": 5}, {"point": "mysql-orders Pod Running 无重启，不是 OOM/CrashLoop，是性能退化", "score": 3}, {"point": "只有结账受影响：商品浏览和购物车不走 mysql-orders", "score": 3}, {"point": "StressChaos 是 pod-level CPU stress，不是 node-level，容易被忽略", "score": 4}], "tool_process": {"ideal_call_count": 12, "valuable_actions": ["MCP/K8s 查看 alicloud networkchaos", "MCP/K8s 查看 alicloud stresschaos", "MCP/K8s 查看 checkoutservice 环境变量", "MCP/K8s 查看 mysql-orders pod 状态", "日志或 APM 查看 checkoutservice DB 等待和 timeout"]}}, "rubrics": {"positive_points_total": 104, "negative_points_total": -58, "rubrics": [{"criterion": "描述了观察到的故障现象（高峰期结账卡住/500，越刷新越慢，商品推荐偶尔也卡）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 checkoutservice 日志或 APM span，发现连接等待、DB timeout 或 pool wait 变长", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 alicloud 的 NetworkChaos 实验列表", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 checkoutservice -> mysql-orders 存在网络延迟", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 NetworkChaos 的具体参数（delay 100ms, jitter 20ms）和实验名称（mysql-orders-slow-net）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了 alicloud 的 StressChaos 实验列表", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 mysql-orders 存在 pod-level CPU 压力", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 StressChaos 的具体参数（CPU 70%, 2 workers）和作用范围（pod-level）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了 checkoutservice 环境变量", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 DB_MAX_CONNECTIONS 被收紧", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 DB_MAX_CONNECTIONS=8", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "确认 mysql-orders pod Running 无重启，排除 OOM/CrashLoop", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "正确追踪因果链：网络延迟 + CPU 压力 + 连接池收紧 -> 排队 -> 超时", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解排队/等待导致结账超时，而不是把它归为服务崩溃", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "解释了为什么只有结账受影响（商品浏览和购物车不走 mysql-orders）", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出了修复建议：删除 NetworkChaos mysql-orders-slow-net", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：删除 StressChaos mysql-cpu-stress", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：恢复 DB_MAX_CONNECTIONS 默认值", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了后续建议（连接池等待监控、慢查询告警）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "说明 DB_MAX_CONNECTIONS 被收紧后，请求会进入连接等待队列而非立即失败", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "说明网络延迟、CPU 压力和连接池收紧三者都需要处理，单独处理一层不足以闭环", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "识别出 recommendationservice 的 NetworkChaos 是干扰项，与结账超时无直接因果关系", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出干扰项的参数（recommendation-net-delay, delay 300ms, jitter 80ms）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "模型能够把 checkoutservice、mysql-orders 和连接池限制串成一条证据链", "points": 5, "tags": ["level:cluster", "cluster:queueing_reasoning", "axis:accuracy"]}, {"criterion": "模型会跨云对比，确认异常范围主要落在阿里云结账链路", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型不被 mysql-orders Running 的表象误导，能继续检查配置和延迟", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（NetworkChaos、StressChaos、env、logs/APM），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "响应按清晰结构组织：观察现象 -> 收集证据 -> 定位根因 -> 修复建议", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 12 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 15 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "将三层故障错误归因为单一根因（如只看到 NetworkChaos 就下结论）", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "遗漏 mysql-orders 的 StressChaos", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "把 mysql-orders 的 pod-level CPU stress 误判为 node-level 资源问题", "points": -4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未检查 checkoutservice 环境变量就下结论，遗漏 DB_MAX_CONNECTIONS", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误判为 mysql-orders OOM/CrashLoop，而实际 pod 正常运行", "points": -6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只建议修复一层就认为可以解决", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "仅查看单一来源或单一云就给出全局诊断", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 18 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "把概率性故障说成 100% 确定性失败，或给出完全不合理的根因判断", "points": -4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "被干扰项误导，将 recommendationservice 延迟误判为结账超时的根因", "points": -7, "tags": ["level:example", "axis:context_awareness"]}], "ideal_call_count": 12}, "prompt": "高峰期结账经常卡住，有时直接 500，低峰期偶尔也会撞上。商品浏览和购物车基本正常。问题不是每次都炸，但连续刷新几次大概率会碰到，而且越刷新越慢。另外商品推荐有时候也卡一下，但不确定是不是同一个问题。"}
{"canary": "redis-cache-miss-amplification", "tags": ["度量", "中间件与依赖", "缓存雪崩", "I/O 延迟", "网络延迟", "hard", "multi-cloud"], "case": {"id": "redis-cache-miss-amplification", "title": "缓存回源放大与超时边界", "inject_script": "./challenge-27-redis-cache-miss-amplification/inject.sh", "recover_script": "./challenge-27-redis-cache-miss-amplification/recover.sh", "fault_phenomenon": "最近好多人说加购物车和结账的时候页面卡住不动，有时候等好几秒才出来，有时候直接超时。逛商品的时候偶尔也会卡一下，价格有时候加载不出来。刷新之后能好一会儿，但很快又变慢。不是每次都这样，但碰到的概率挺高的。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，并结合日志、RUM 和端到端业务表现排查。"}, "ideal_answer": {"fault_info": {"root_cause": "四层故障叠加：(1) NetworkChaos 在 cartservice -> redis-cart 之间注入网络延迟和抖动，(2) cartservice 的 REDIS_CONNECT_TIMEOUT 贴近延迟边界，(3) CACHE_TTL 偏短，导致频繁回源，(4) IOChaos 对 redis-cart 施加磁盘 IO 延迟。单次请求不一定必死，但在回源放大、网络抖动和磁盘延迟叠加时，会更容易卡在 Redis 调用上。", "faulty_clouds": ["tencent"], "affected_clouds": ["tencent"], "fault_location": {"module": "chaos-mesh + k8s-deployment", "file_path": null, "function_or_config": "NetworkChaos: redis-cart-latency, IOChaos: redis-cart-io-delay, deploy/cartservice: CACHE_TTL=2s, REDIS_CONNECT_TIMEOUT=260ms", "description": "回源放大 + 边界超时 + 磁盘IO延迟。核心不是 Redis 挂了，而是部分请求跨过了超时边界。IOChaos 让 Redis 写入变慢，放大了回源时的延迟。"}}, "reasoning_process": {"observed_symptoms": ["用户反馈部分页面偶发卡顿，集中在加商品、查看购物车和结账环节。", "等待时间从几秒到十几秒不等，不是所有请求都慢。", "商品浏览偶尔也会卡，价格有时加载不出来。", "刷新后有时稍好但很快又变慢。", "Redis pod Running，看起来并没有挂。"], "key_evidence": [{"source": "MCP/K8s: tencent seat-1 get networkchaos", "content": "redis-cart-latency: cartservice -> redis-cart delay 170ms, jitter 70ms", "conclusion": "Redis 调用存在延迟和抖动。"}, {"source": "MCP/K8s: tencent seat-1 get iochaos", "content": "redis-cart-io-delay: redis-cart /data 目录 IO 延迟 80ms, 60% 操作", "conclusion": "Redis 写入变慢，放大了回源时的延迟。"}, {"source": "MCP/K8s: tencent seat-1 get deploy cartservice -o yaml", "content": "CACHE_TTL=2s, REDIS_CONNECT_TIMEOUT=260ms", "conclusion": "TTL 偏短且超时边界偏紧，容易放大回源和等待。"}, {"source": "MCP/K8s: tencent seat-1 get pods -l app=redis-cart", "content": "redis-cart Running", "conclusion": "不能把问题简单归因为 Redis 挂掉。"}, {"source": "日志或 RUM/APM", "content": "Redis connection timeout、回源增多或 cart 请求 p95 拉长", "conclusion": "证据指向回源放大后的超时边界，而不是单点宕机。"}], "causal_chain": ["TTL 偏短 -> 更多请求需要回源。", "NetworkChaos 把 Redis 往返拉长，并带来抖动。", "IOChaos 让 Redis 写入变慢，放大了回源时的延迟。", "REDIS_CONNECT_TIMEOUT 贴近边界 -> 一部分请求跨过超时线。", "于是购物车和相关页面整体变慢，偶发超时或卡顿。", "Redis Pod 仍 Running，所以不能把问题简化为 Redis down。"]}, "proposed_resolution": {"suggestion": "建议同时清掉网络延迟、Redis 压力和 cartservice 的边界配置，并补充缓存命中率和 Redis timeout 监控。", "fix_steps": ["kubectl --context=tencent -n seat-1 delete networkchaos redis-cart-latency", "kubectl --context=tencent -n seat-1 delete iochaos redis-cart-io-delay", "kubectl --context=tencent -n seat-1 set env deploy/cartservice CACHE_TTL-", "kubectl --context=tencent -n seat-1 set env deploy/cartservice REDIS_CONNECT_TIMEOUT-"]}, "confidence": 0.92, "special_points": [{"point": "TTL 偏短会放大回源，而不是单纯让缓存更快失效", "score": 5}, {"point": "Redis Pod Running 不代表链路正常，不能把问题当成 Redis down", "score": 4}, {"point": "超时配置贴近边界时，抖动会把一部分请求推过阈值", "score": 4}, {"point": "IOChaos 让 Redis 写入变慢，放大了回源时的延迟", "score": 4}, {"point": "缓存问题必须结合日志或 RUM/APM 才能确认是回源放大还是服务宕机", "score": 3}], "tool_process": {"ideal_call_count": 10, "valuable_actions": ["MCP/K8s 查看 tencent networkchaos", "MCP/K8s 查看 tencent iochaos", "MCP/K8s 查看 cartservice env", "MCP/K8s 查看 redis-cart pod 状态", "日志或 RUM/APM 查看 Redis timeout 与 p95"]}}, "rubrics": {"positive_points_total": 97, "negative_points_total": -53, "rubrics": [{"criterion": "描述了观察到的故障现象（部分页面偶发卡顿，集中在加商品、购物车和结账，等待时间几秒到十几秒不等）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 cartservice 日志或 APM，发现 Redis connection timeout 或明显的等待变长", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 tencent 的 NetworkChaos 实验列表", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 cartservice -> redis-cart 存在网络延迟", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 NetworkChaos 的具体参数（delay 170ms, jitter 70ms）和实验名称（redis-cart-latency）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了 tencent 的 IOChaos 实验列表", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 redis-cart 存在磁盘 IO 延迟", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 IOChaos 的具体参数（delay 80ms, percent 60%）和作用目录（/data）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了 cartservice 环境变量", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 REDIS_CONNECT_TIMEOUT 被设为边界值", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 REDIS_CONNECT_TIMEOUT=260ms", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 CACHE_TTL 被设为偏短值", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 CACHE_TTL=2s", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "确认 redis-cart pod Running，不能简单归因于 Redis 宕机", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "说明 CACHE_TTL=2s 会提高回源频率，从而放大 Redis 调用压力", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "说明 delay+jitter+Redis IO 延迟会让部分请求超过 REDIS_CONNECT_TIMEOUT=260ms", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "确认 redis-cart Running 后仍继续检查链路延迟、压力和 cartservice 配置", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "解释了为什么购物车和相关页面会一起变慢，而不是只有一个请求失败", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出了修复建议：删除 NetworkChaos redis-cart-latency", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：删除 IOChaos redis-cart-io-delay", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：恢复 CACHE_TTL 和 REDIS_CONNECT_TIMEOUT", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了后续建议（缓存命中率监控、Redis timeout 告警）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 -> 收集证据 -> 定位根因 -> 修复建议", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "模型能够把回源放大、边界超时和磁盘IO延迟串成一条证据链", "points": 5, "tags": ["level:cluster", "cluster:timeout_boundary_reasoning", "axis:accuracy"]}, {"criterion": "模型会跨云对比，确认异常主要落在腾讯云的购物车链路", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型不被 redis-cart Running 的表象误导，能够从配置和日志继续排查", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（NetworkChaos、IOChaos、env、logs/APM/RUM），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 10 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 12 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 15 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "误以为 Redis 本身挂了，而实际是边界超时 + 回源放大 + IO延迟叠加", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "遗漏 REDIS_CONNECT_TIMEOUT=260ms", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "遗漏 CACHE_TTL=2s", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未理解超时边界和抖动的关系，仅判断为慢", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查环境变量就下结论", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只修复一层就认为解决，未同时处理网络、压力和配置", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "仅查看单一来源或单一云就给出全局诊断", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "把概率性故障说成 100% 确定性失败，或给出完全不合理的根因判断", "points": -4, "tags": ["level:example", "axis:accuracy"]}], "ideal_call_count": 10}, "prompt": "最近好多人说加购物车和结账的时候页面卡住不动，有时候等好几秒才出来，有时候直接超时。逛商品的时候偶尔也会卡一下，价格有时候加载不出来。刷新之后能好一会儿，但很快又变慢。不是每次都这样，但碰到的概率挺高的。"}
{"canary": "cpu-throttle-cascade", "tags": ["度量", "容器与工作负载", "CPU限流", "网络延迟", "hard", "multi-cloud"], "case": {"id": "cpu-throttle-cascade", "title": "命名空间级 CPU 压力与跨服务退化", "inject_script": "./challenge-28-cpu-throttle-cascade/inject.sh", "recover_script": "./challenge-28-cpu-throttle-cascade/recover.sh", "fault_phenomenon": "有一段时间了，商城整体变慢：加购物车要等好几秒，商品列表偶尔刷不出来，价格换算也卡。不是完全不能用，就是慢。试了刷新、换了浏览器都没用。商品推荐有时候也卡一下，但不确定是不是同一个问题。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，并结合日志、资源指标和端到端业务表现排查。"}, "ideal_answer": {"fault_info": {"root_cause": "两层故障叠加 + 一个干扰项：(1) 腾讯云 seat-1 命名空间内存在 mode: all 的 StressChaos，使多个 Pod 同时消耗 CPU，表现为命名空间内多服务整体资源紧张，(2) currencyservice 同时被注入 550ms 级网络延迟，(3) 阿里云 frontend 被注入 200ms 延迟（干扰项，与腾讯云整体慢无直接因果关系）。CPU 压力导致腾讯云入口整体退化，currencyservice 延迟进一步放大货币换算和调用链等待。", "faulty_clouds": ["tencent"], "affected_clouds": ["tencent"], "fault_location": {"module": "chaos-mesh", "file_path": null, "function_or_config": "StressChaos: tencent-namespace-cpu-stress (mode: all), NetworkChaos: currency-service-delay (currencyservice delay 550ms, jitter 120ms), NetworkChaos: frontend-net-delay（干扰项，阿里云 frontend delay 200ms, jitter 60ms）", "description": "多 pod CPU pressure + currencyservice 延迟 + 干扰项 frontend 延迟。关键是先判断影响面，再区分整体资源压力和单服务延迟放大器，并识别干扰项。"}}, "reasoning_process": {"observed_symptoms": ["腾讯云入口整体变慢，多个业务动作都卡。", "商品推荐有时候也卡一下。", "所有 Pod Running，无明显重启。", "currencyservice 相关操作更明显。"], "key_evidence": [{"source": "MCP/K8s: tencent seat-1 get stresschaos", "content": "tencent-namespace-cpu-stress: mode all, CPU 75%, 2 workers", "conclusion": "腾讯云 seat-1 多个 Pod 同时被施加 CPU 压力。"}, {"source": "MCP/K8s: tencent top pods / top nodes", "content": "腾讯云多个 pod CPU 使用率或 throttling 明显升高，必要时可结合节点 CPU 指标判断整体资源压力", "conclusion": "影响面不是单个服务。"}, {"source": "MCP/K8s: tencent seat-1 get networkchaos", "content": "currency-service-delay: currencyservice delay 550ms, jitter 120ms", "conclusion": "currencyservice 响应被额外拉长。"}, {"source": "MCP/K8s: alicloud/aws 对比 pods 或业务入口", "content": "其他云资源和入口表现基本正常", "conclusion": "异常范围主要在腾讯云。"}, {"source": "日志/APM/RUM", "content": "腾讯云多个服务 p95 拉长，currencyservice 相关 span 更慢", "conclusion": "整体 CPU 压力和单服务延迟同时存在。"}, {"source": "MCP/K8s: alicloud seat-1 get networkchaos", "content": "frontend-net-delay: frontend delay 200ms, jitter 60ms", "conclusion": "干扰项。frontend 在阿里云，与腾讯云整体慢无直接因果关系。"}], "causal_chain": ["StressChaos mode: all 让腾讯云 seat-1 多个 Pod 同时消耗 CPU。", "多个服务响应变慢，入口表现为整体退化。", "currencyservice 叠加 550ms 级延迟。", "货币换算和依赖 currencyservice 的路径更慢。", "Pod Running 说明不是 CrashLoop，而是资源和延迟退化。", "干扰项：阿里云 frontend 延迟会影响阿里云入口，但不会导致腾讯云整体变慢。"]}, "proposed_resolution": {"suggestion": "建议清理腾讯云的 StressChaos 和 currencyservice NetworkChaos，以及阿里云的干扰项 frontend NetworkChaos。后续补充跨云入口 p95、Pod CPU throttling 和 currencyservice 依赖延迟告警。", "fix_steps": ["kubectl --context=tencent -n seat-1 delete stresschaos tencent-namespace-cpu-stress", "kubectl --context=tencent -n seat-1 delete networkchaos currency-service-delay", "kubectl --context=alicloud -n seat-1 delete networkchaos frontend-net-delay", "建议恢复后观察腾讯云 top nodes/top pods 和 RUM/APM p95。"]}, "confidence": 0.93, "special_points": [{"point": "mode: all 代表影响面广，不能当成单 pod 小问题", "score": 5}, {"point": "所有腾讯云服务都慢时，应先判断影响面，再定位单服务放大器", "score": 4}, {"point": "currencyservice 延迟是放大器，不是唯一根因", "score": 4}, {"point": "Pod Running 不等于服务健康，资源压力仍会造成级联退化", "score": 4}, {"point": "阿里云 frontend 延迟是干扰项，与腾讯云整体慢无直接因果关系", "score": 3}], "tool_process": {"ideal_call_count": 11, "valuable_actions": ["MCP/K8s 查看 tencent stresschaos", "MCP/K8s 查看 tencent networkchaos", "MCP/K8s 查看 tencent top pods/top nodes", "跨云对比 alicloud/aws 的入口或 pod 状态", "MCP/K8s 查看 alicloud networkchaos（识别干扰项）", "RUM/APM 查看腾讯云入口和 currencyservice p95"]}}, "rubrics": {"positive_points_total": 104, "negative_points_total": -57, "rubrics": [{"criterion": "描述了观察到的故障现象（腾讯云入口整体慢，商品推荐偶尔也卡，Pod Running）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 RUM、APM 或日志，确认腾讯云多个服务 p95 同时拉长", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 tencent 的 StressChaos 实验列表", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出腾讯云 seat-1 存在 mode: all 的多 pod CPU 压力", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 StressChaos 的具体参数（CPU 75%, 2 workers, mode: all）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "使用 top pods 或等价资源指标确认腾讯云多个 pod CPU 压力升高", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 top nodes 或等价资源指标辅助确认腾讯云整体资源压力", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 tencent 的 NetworkChaos 实验列表", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 currencyservice 存在额外响应延迟", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 NetworkChaos 的具体参数（delay 550ms, jitter 120ms）和作用对象（currencyservice）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "跨云对比阿里云/AWS，确认异常主要落在腾讯云", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 alicloud 的 NetworkChaos 实验列表", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 frontend 延迟是干扰项，与腾讯云整体慢无直接因果关系", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出干扰项的参数（frontend-net-delay, delay 200ms, jitter 60ms）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "确认所有 Pod Running 无明显重启，排除 CrashLoop/OOM 作为主因", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "结合多个腾讯云服务 p95 同时拉长和 top 证据，判断异常指向整体资源压力", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "说明 currencyservice 延迟是相关调用链的放大器，而不是腾讯云整体变慢的唯一根因", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确追踪因果链：多 pod CPU pressure -> 整体退化；currencyservice delay -> 局部路径更慢", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出了修复建议：删除 StressChaos tencent-namespace-cpu-stress", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：删除 NetworkChaos currency-service-delay", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：删除 NetworkChaos frontend-net-delay（干扰项）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了后续建议（Pod CPU throttling 告警、跨云入口延迟监控）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 -> 收集证据 -> 定位根因 -> 修复建议", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "模型能够区分整体资源压力和单服务延迟放大器", "points": 5, "tags": ["level:cluster", "cluster:namespace_vs_service_fault_isolation", "axis:accuracy"]}, {"criterion": "模型的排查过程覆盖腾讯云、阿里云和 AWS，能够用跨云对比收敛范围", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型不被 Pod Running 的表象误导，能够继续检查资源压力和延迟", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（StressChaos、NetworkChaos、top、RUM/APM），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 11 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 13 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "误以为是单个服务问题，未从整体资源压力角度排查", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "遗漏 StressChaos（最关键的整体压力来源）", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未区分 mode: all 的影响面，把它当作单 pod 小故障", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "遗漏 currencyservice NetworkChaos 这个延迟放大器", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未使用 top pods/top nodes 或等价资源证据就下结论", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "仅查看单一云就给出全局诊断", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 16 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "把概率性故障说成 100% 确定性失败，或给出完全不合理的根因判断", "points": -4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "被干扰项误导，将 frontend 延迟误判为腾讯云整体慢的根因", "points": -7, "tags": ["level:example", "axis:context_awareness"]}], "ideal_call_count": 11}, "prompt": "有一段时间了，商城整体变慢：加购物车要等好几秒，商品列表偶尔刷不出来，价格换算也卡。不是完全不能用，就是慢。试了刷新、换了浏览器都没用。商品推荐有时候也卡一下，但不确定是不是同一个问题。"}
{"canary": "rum-apm-joint", "tags": ["定位", "可观测系统自身", "RUM数据异常", "指标缺失", "网络延迟", "hard", "multi-cloud"], "case": {"id": "rum-apm-joint", "title": "RUM + APM 联合排查：页面退化与链路断点", "inject_script": "./challenge-29-rum-apm-joint/inject.sh", "recover_script": "./challenge-29-rum-apm-joint/recover.sh", "fault_phenomenon": "运营反馈商品详情页最近打开特别慢，图片加载半天出不来，有时候直接白屏。但首页和购物车页都挺正常的。用户投诉集中在商品页，其他页面没什么反馈。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，并结合 RUM、APM 和日志做联合排查。RUM App Key: QVWbXcJbj1ZeOYQ0zJ，APM Endpoint: ap-guangzhou.apm.tencentcs.com:4319。"}, "ideal_answer": {"fault_info": {"root_cause": "三层故障叠加：(1) NetworkChaos 在 productcatalogservice 注入 400ms 级延迟，(2) checkoutservice 的 ENABLE_TRACING 被设为 0，导致 APM 链路断点，(3) RUM ConfigMap 只对 /product/ 路由降低采样率，商品页数据变稀疏。必须同时借助 RUM 和 APM 才能把页面异常和后端服务对上。", "faulty_clouds": ["unknown"], "affected_clouds": ["tencent", "alicloud"], "fault_location": {"module": "chaos-mesh + k8s-deployment + configmap", "file_path": null, "function_or_config": "NetworkChaos: productcatalog-net-delay (delay 400ms, jitter 80ms), deploy/checkoutservice: ENABLE_TRACING=0, ConfigMap: rum-nginx-config /product/ sample=0.25", "description": "网络延迟 + 链路断点 + 商品页采样降低。单靠 RUM 或单靠 APM 都不够，必须把两边证据合并。"}}, "reasoning_process": {"observed_symptoms": ["RUM 中商品详情页 FCP/LCP 飙升。", "首页和购物车页基本正常。", "APM 中商品相关 span 变慢。", "APM 中 checkoutservice 的调用链缺少子 span。", "商品页数据比其他页面更稀疏。"], "key_evidence": [{"source": "RUM 控制台", "content": "/product/* 页面 FCP/LCP 上升，API 耗时拉长", "conclusion": "商品页是异常集中点。"}, {"source": "APM 控制台", "content": "productcatalogservice span 明显变慢", "conclusion": "后端商品链路存在延迟。"}, {"source": "APM 控制台", "content": "checkoutservice 链路只有入口 span，缺少子 span", "conclusion": "trace 传播断了。"}, {"source": "MCP/K8s: tencent seat-1 get networkchaos", "content": "productcatalog-net-delay: productcatalogservice delay 400ms, jitter 80ms", "conclusion": "商品服务侧存在网络延迟。"}, {"source": "MCP/K8s: alicloud seat-1 get deploy checkoutservice -o yaml", "content": "ENABLE_TRACING=0", "conclusion": "APM 链路断点来自 tracing 关闭。"}, {"source": "MCP/K8s: alicloud seat-1 get configmap rum-nginx-config -o yaml", "content": "/product/ 路由 sample=0.25", "conclusion": "商品页 RUM 采样降低，导致数据更稀疏。"}], "causal_chain": ["productcatalogservice 的延迟让商品页 API 变慢。", "ENABLE_TRACING=0 让 APM 无法把 checkoutservice 和下游 span 串起来。", "RUM 采样只对商品页降低，数据稀疏但仍能看到异常。", "RUM 先告诉你是哪个页面，APM 再告诉你是哪个服务。", "只有把两边证据合并，才能闭环到真正的故障点。"]}, "proposed_resolution": {"suggestion": "建议同时恢复 productcatalog 延迟、checkoutservice tracing 和 RUM 商品页采样。后续补充 RUM 采样审计、APM trace 传播检查和 ConfigMap 变更审计。", "fix_steps": ["kubectl --context=tencent -n seat-1 delete networkchaos productcatalog-net-delay", "kubectl --context=alicloud -n seat-1 set env deploy/checkoutservice ENABLE_TRACING=1", "kubectl --context=alicloud -n seat-1 apply -f <恢复后的 rum-nginx-config，使 /product/ 路由 sample 回到 1>", "kubectl --context=alicloud -n seat-1 rollout restart deploy/frontend-rum-proxy", "建议恢复后复看 /product/ 的 RUM 和 APM 数据"]}, "confidence": 0.93, "special_points": [{"point": "必须同时用 RUM 和 APM 才能把页面异常和后端服务对应起来", "score": 5}, {"point": "ENABLE_TRACING=0 会让 APM 链路断点，不是简单的服务慢", "score": 4}, {"point": "RUM 只对商品页降采样，意味着数据稀疏但仍可诊断", "score": 4}, {"point": "productcatalog 本身的网络延迟是页面退化的主要来源", "score": 3}], "tool_process": {"ideal_call_count": 11, "valuable_actions": ["MCP/K8s 查看 tencent networkchaos", "MCP/K8s 查看 alicloud checkoutservice env", "MCP/K8s 查看 alicloud rum-nginx-config", "RUM 控制台查看 /product/ 页面", "APM 控制台查看 checkoutservice 和 productcatalogservice span"]}}, "rubrics": {"positive_points_total": 98, "negative_points_total": -57, "rubrics": [{"criterion": "描述了观察到的故障现象（商品详情页 RUM 变差，首页和购物车基本正常）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 RUM 数据，识别出 /product/* 页面 FCP/LCP 上升", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "准确给出 /product/ 页面 FCP/LCP 异常对应的后端服务（productcatalogservice 延迟），而非误判为浏览器或前端资源问题", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了 APM 数据，发现 productcatalogservice span 变慢", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 APM 数据，发现 checkoutservice 链路缺少子 span", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 tencent 的 NetworkChaos 实验列表", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 productcatalogservice 存在网络延迟", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 NetworkChaos 的实验名称、作用对象和参数（productcatalog-net-delay, productcatalogservice, delay 400ms, jitter 80ms）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了 checkoutservice 环境变量", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 ENABLE_TRACING 被设为 0，导致链路断点", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 ENABLE_TRACING=0", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了 RUM ConfigMap 配置", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 /product/ 路由 sample 被降低", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 sample=0.25（商品页数据更稀疏）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "说明 RUM 用于定位异常页面，APM 用于定位后端服务，二者需要联合闭环", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "说明 ENABLE_TRACING=0 会导致 trace context 不传播，从而出现 APM 子 span 缺失", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "说明 /product/ sample=0.25 会让商品页数据稀疏，但剩余样本仍能反映趋势", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "理解只有商品页异常，而首页和购物车页基本正常", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出了修复建议：删除 NetworkChaos productcatalog-net-delay", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：恢复 ENABLE_TRACING=1", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：恢复 RUM 商品页 sample", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：重启 frontend-rum-proxy 使 ConfigMap 生效", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 -> 收集证据 -> 定位根因 -> 修复建议", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "模型能够同时利用 RUM 和 APM 完成页面到服务的闭环诊断", "points": 5, "tags": ["level:cluster", "cluster:rum_apm_joint_diagnosis", "axis:context_awareness"]}, {"criterion": "模型的排查过程覆盖腾讯云和阿里云，能够跨云收集证据", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型不被链路断点误导，仍能通过 RUM 和独立 span 推断调用关系", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（RUM、APM、NetworkChaos、ConfigMap），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 11 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 13 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 16 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "仅用 APM 或仅用 RUM 就下结论，未联合排查", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "遗漏 ENABLE_TRACING=0（链路断点根因）", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "遗漏 RUM sample 降低", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误以为 productcatalogservice pod 本身有问题，而实际故障在网络层", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 ConfigMap 就下结论", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "仅查看单一云的证据就给出跨云诊断，未覆盖腾讯云和阿里云", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "发现 RUM 采样降低后直接放弃 RUM 数据，未理解剩余样本仍可诊断", "points": -5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "把概率性故障说成 100% 确定性失败，或给出完全不合理的根因判断", "points": -3, "tags": ["level:example", "axis:accuracy"]}], "ideal_call_count": 11}, "prompt": "运营反馈商品详情页最近打开特别慢，图片加载半天出不来，有时候直接白屏。但首页和购物车页都挺正常的。用户投诉集中在商品页，其他页面没什么反馈。"}
{"canary": "payment-cascade", "tags": ["定位", "跨云一致性与容量稳定性", "网络延迟", "超时重试", "烟雾弹", "hard", "multi-cloud"], "case": {"id": "payment-cascade", "title": "支付慢点与购物车冒烟：根因与干扰", "inject_script": "./challenge-30-payment-cascade/inject.sh", "recover_script": "./challenge-30-payment-cascade/recover.sh", "fault_phenomenon": "下单时经常卡很久才响应，有时直接报 500。浏览商品基本正常，购物车操作偶尔也慢。首页打开速度还行。不是每次都坏，但走到结账就容易撞上，刷新重试后有时更慢。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，并结合 RUM、APM、日志和端到端业务表现排查。"}, "ideal_answer": {"fault_info": {"root_cause": "一主因 + 一冒烟 + 一放大器 + 一已自愈残留：(1) AWS paymentservice 被注入 450ms 级响应延迟，是结账路径主慢点，(2) 腾讯云 cartservice 被注入 280ms 级延迟，是购物车慢的冒烟干扰，(3) 阿里云 checkoutservice 的 CHECKOUT_TIMEOUT_MS 被收紧到 1400，(4) AWS new-orderservice 存在一个已自愈的 StressChaos 残留（duration=1s 已过期，但 chaos 对象仍在）。结账链路在支付慢点、购物车延迟和额外开销叠加后越过 timeout，导致部分下单超时。", "faulty_clouds": ["unknown"], "affected_clouds": ["alicloud", "tencent", "aws"], "fault_location": {"module": "chaos-mesh + k8s-deployment", "file_path": null, "function_or_config": "NetworkChaos: payment-service-delay (AWS, delay 450ms, jitter 80ms), NetworkChaos: cart-service-delay (Tencent, delay 280ms, jitter 80ms), deploy/checkoutservice: CHECKOUT_TIMEOUT_MS=1400, StressChaos: order-cpu-stress-residual (AWS, 已自愈)", "description": "payment 是主慢点，cart 是冒烟干扰，checkout timeout 是放大器。已自愈的 StressChaos 残留可作为排查历史线索。"}}, "reasoning_process": {"observed_symptoms": ["下单经常卡很久或 500。", "浏览商品基本正常。", "购物车偶尔也慢，但不是所有请求都失败。", "刷新重试后有时更慢。"], "key_evidence": [{"source": "MCP/K8s: aws seat-1 get networkchaos", "content": "payment-service-delay: paymentservice delay 450ms, jitter 80ms", "conclusion": "支付服务是结账路径的主慢点。"}, {"source": "MCP/K8s: tencent seat-1 get networkchaos", "content": "cart-service-delay: cartservice delay 280ms, jitter 80ms", "conclusion": "购物车链路也慢，但量级更小，属于冒烟干扰。"}, {"source": "MCP/K8s: alicloud seat-1 get deploy checkoutservice -o yaml", "content": "CHECKOUT_TIMEOUT_MS=1400", "conclusion": "结账超时预算被收紧。"}, {"source": "APM 或日志", "content": "checkoutservice -> paymentservice span 明显大于 checkoutservice -> cartservice span", "conclusion": "payment 是主要瓶颈，cart 是次要慢点。"}, {"source": "RUM 或端到端业务表现", "content": "商品浏览正常，结账链路 p95/p99 升高", "conclusion": "故障集中在下单路径。"}, {"source": "MCP/K8s: aws seat-1 get stresschaos", "content": "order-cpu-stress-residual: new-orderservice CPU 80%, 2 workers, 但 StartTime 和 RecoverTime 表明已过期", "conclusion": "已自愈的 StressChaos 残留。不是当前根因，但说明 new-orderservice 曾经受过 CPU 压力。"}], "causal_chain": ["paymentservice 延迟让结账必经路径变慢。", "cartservice 延迟造成购物车冒烟，也会增加结账前置等待。", "CHECKOUT_TIMEOUT_MS=1400 使原本还能容忍的延迟预算变紧。", "payment + cart + 其他开销叠加后超过 timeout。", "超时后的刷新/重试会重复打下游，造成体感更慢。", "因此 cart 不是主根因，payment 才是结账超时的关键慢点。", "已自愈的 StressChaos 残留不是当前根因，但说明 new-orderservice 曾经受过 CPU 压力，可作为排查历史线索。"]}, "proposed_resolution": {"suggestion": "建议同时处理 payment 主慢点、cart 冒烟慢点和 checkout 超时预算。后续补充跨云 checkout 链路 p95/p99、重试退避和 timeout budget 告警。", "fix_steps": ["kubectl --context=aws -n seat-1 delete networkchaos payment-service-delay", "kubectl --context=tencent -n seat-1 delete networkchaos cart-service-delay", "kubectl --context=alicloud -n seat-1 set env deploy/checkoutservice CHECKOUT_TIMEOUT_MS-（或按默认配置恢复为 CHECKOUT_TIMEOUT_MS=2000）", "kubectl --context=aws -n seat-1 delete stresschaos order-cpu-stress-residual（已自愈，但 chaos 对象仍在集群中，建议清理）", "建议恢复后观察结账成功率和 payment/cart span 分布。"]}, "confidence": 0.93, "special_points": [{"point": "区分根因（payment）和冒烟（cart）是核心难点", "score": 5}, {"point": "checkout timeout 是放大器，不是唯一根因", "score": 4}, {"point": "刷新重试会放大下游压力，解释体感更慢", "score": 3}, {"point": "必须跨三个集群收集证据", "score": 5}, {"point": "已自愈的 StressChaos 残留不是当前根因，但可作为排查历史线索", "score": 4}], "tool_process": {"ideal_call_count": 10, "valuable_actions": ["MCP/K8s 查看 aws networkchaos", "MCP/K8s 查看 tencent networkchaos", "MCP/K8s 查看 alicloud checkoutservice env", "APM 对比 payment 和 cart span", "RUM 或端到端观察 checkout p95/p99"]}}, "rubrics": {"positive_points_total": 104, "negative_points_total": -58, "rubrics": [{"criterion": "描述了观察到的故障现象（结账卡顿/500，购物车偶尔慢，商品浏览基本正常）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了日志或 APM，确认 checkout 路径存在超时和下游等待", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 AWS 的 NetworkChaos 实验列表", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 paymentservice 存在主慢点延迟", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 payment-service-delay 的具体参数（delay 450ms, jitter 80ms）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了腾讯云的 NetworkChaos 实验列表", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 cartservice 存在冒烟延迟", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 cart-service-delay 的具体参数（delay 280ms, jitter 80ms）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了 checkoutservice 环境变量", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 CHECKOUT_TIMEOUT_MS 被收紧", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 CHECKOUT_TIMEOUT_MS=1400", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确区分根因（payment 主慢点）和冒烟（cart 次要慢点）", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "说明即使 cart 延迟被修复，payment 主慢点仍会让结账接近或越过超时预算", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "说明 CHECKOUT_TIMEOUT_MS=1400 如何把 payment/cart 延迟叠加后的边界请求变成 500", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "说明刷新重试会重复调用 payment/cart 下游，从而放大等待和失败概率", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "正确追踪因果链：payment 延迟 + cart 冒烟 + timeout 收紧 -> checkout 超时", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "给出了修复建议：删除 payment-service-delay", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：删除 cart-service-delay", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：恢复 CHECKOUT_TIMEOUT_MS=2000 或默认值", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了后续建议（重试退避、跨云 timeout budget、端到端 p95/p99 监控）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出已自愈的 StressChaos 残留（order-cpu-stress-residual），并判断其非当前根因", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出已自愈残留的参数（CPU 80%, 2 workers, duration=1s 已过期）和作用对象（new-orderservice）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解已自愈残留可作为排查历史线索，但不影响当前故障判断", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "响应按清晰结构组织：观察现象 -> 收集证据 -> 定位根因 -> 修复建议", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "模型能够正确区分根因和冒烟，不被购物车也慢误导", "points": 5, "tags": ["level:cluster", "cluster:root_vs_smoke_differentiation", "axis:accuracy"]}, {"criterion": "模型的排查过程覆盖 AWS、腾讯云和阿里云，能够跨云收集证据", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型能够结合 APM span 量级和 timeout 配置判断主次关系", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（NetworkChaos、env、APM/RUM），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 10 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 12 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 15 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "将 cart 280ms 误判为主根因，未区分根因和冒烟", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "遗漏 CHECKOUT_TIMEOUT_MS 收紧（放大器）", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "遗漏两个 NetworkChaos 中的任何一个", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未检查 checkoutservice 环境变量就下结论", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未理解重试放大效应", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "仅查看单一云就给出全局诊断", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "把概率性故障说成 100% 确定性失败，或给出完全不合理的根因判断", "points": -4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "被已自愈的 StressChaos 残留误导，将其误判为当前根因", "points": -7, "tags": ["level:example", "axis:context_awareness"]}], "ideal_call_count": 10}, "prompt": "下单时经常卡很久才响应，有时直接报 500。浏览商品基本正常，购物车操作偶尔也慢。首页打开速度还行。不是每次都坏，但走到结账就容易撞上，刷新重试后有时更慢。"}
{"canary": "gateway-crosscloud-timeout", "tags": ["定位", "跨云一致性与容量稳定性", "跨云网络延迟", "网关超时阈值过低", "网络丢包", "hard", "multi-cloud"], "case": {"id": "gateway-crosscloud-timeout", "title": "跨云级联超时：网关超时配置收紧+重试禁用", "inject_script": "./challenge-31-gateway-crosscloud-timeout/inject.sh", "recover_script": "./challenge-31-gateway-crosscloud-timeout/recover.sh", "fault_phenomenon": "有用户反馈访问商城时偶尔打不开，刷新一下又好了。大部分时候浏览商品没问题，但走到下单那一步更容易出错。不是所有人都遇到，时好时坏。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，并结合 RUM、服务日志和端到端业务表现排查。注意：本题有多个 Chaos 对象和 env 配置变化，需要判断哪个是主因、哪个是干扰。"}, "ideal_answer": {"fault_info": {"root_cause": "腾讯云 new-gatewayservice 到 AWS new-orderservice 外部入口（3.26.241.84:30070）的跨云调用被注入 400±150ms 延迟和 12% 丢包，同时 gateway 的 UPSTREAM_TIMEOUT 被收紧到 500ms 且 MAX_RETRIES=0。延迟+抖动可能达到 550ms，超过 500ms 超时阈值，导致偶发连接超时。StressChaos（全命名空间 CPU 压力）是干扰项，不是主因。", "faulty_clouds": ["unknown"], "affected_clouds": ["tencent", "aws", "alicloud"], "fault_location": {"module": "gateway + network", "file_path": null, "function_or_config": "NetworkChaos: gw-to-aws-order-delay (externalTargets=3.26.241.84, 400ms±150ms) + gw-to-aws-order-loss (externalTargets=3.26.241.84, 12%), env: UPSTREAM_TIMEOUT=500ms, MAX_RETRIES=0", "description": "腾讯云 gateway 到 AWS orderservice 的跨云调用延迟+丢包叠加上超时配置收紧，导致偶发 502/504。StressChaos 是干扰项。"}}, "reasoning_process": {"observed_symptoms": ["RUM 中接口错误率和耗时出现尖刺，但不是所有页面都失败。", "腾讯云 gateway 相关接口更容易出现 502/504。", "Pod 基本 Running，Deployment 可用副本正常。", "seat-1 中有多个 Chaos 对象，需要判断主因和干扰。"], "key_evidence": [{"source": "RUM 控制台", "content": "入口/API 维度出现 5xx 和耗时尖刺，用户侧影响集中在网关转发链路。", "conclusion": "先确认真实用户受影响范围。"}, {"source": "kubectl --context=tencent -n seat-1 logs deploy/new-gatewayservice", "content": "日志中出现 upstream timed out、connection timeout 等跨云调用错误。", "conclusion": "网关访问下游时超时，而不是前端静态资源问题。"}, {"source": "kubectl --context=tencent -n seat-1 get networkchaos", "content": "gw-to-aws-order-delay / gw-to-aws-order-loss 作用于 new-gatewayservice 到外部目标 3.26.241.84（AWS new-orderservice NodePort）的流量：400ms±150ms + 丢包 12%", "conclusion": "gateway 到 AWS orderservice 有显著延迟和丢包。"}, {"source": "kubectl --context=tencent -n seat-1 get deploy new-gatewayservice -o yaml", "content": "UPSTREAM_TIMEOUT=500ms, MAX_RETRIES=0", "conclusion": "超时配置比延迟+抖动的上限（550ms）短，且不重试。"}, {"source": "kubectl --context=tencent -n seat-1 get stresschaos", "content": "tencent-global-cpu-stress (mode: all, load: 50)", "conclusion": "这是干扰项，CPU 压力不是主因。"}], "causal_chain": ["NetworkChaos 注入 gateway -> AWS orderservice 外部入口（externalTargets=3.26.241.84）延迟 400±150ms + 丢包 12%。", "env 配置 UPSTREAM_TIMEOUT=500ms，但延迟+抖动可能达到 550ms。", "env 配置 MAX_RETRIES=0，失败不重试。", "并发请求时，部分请求超过 500ms 超时阈值。", "用户侧表现为 gateway 502/504。", "StressChaos 是干扰项，容易被误判为主因。"]}, "proposed_resolution": {"suggestion": "删除 NetworkChaos 对象，恢复 gateway env 配置。长期建议包括合理设置超时阈值（应大于延迟+抖动上限）、启用重试机制、监控跨云调用延迟和丢包率。", "fix_steps": ["kubectl --context=tencent -n seat-1 delete networkchaos gw-to-aws-order-delay gw-to-aws-order-loss", "kubectl --context=tencent -n seat-1 set env deploy/new-gatewayservice UPSTREAM_TIMEOUT- MAX_RETRIES-", "观察 RUM 中 gateway 5xx 和耗时回落。"]}, "confidence": 0.92, "special_points": [{"point": "必须识别 StressChaos 是干扰项而非主因", "score": 5}, {"point": "需要对比延迟值（400±150ms）和超时值（500ms）才能定位主因", "score": 5}, {"point": "MAX_RETRIES=0 禁用重试是放大器", "score": 4}, {"point": "跨云调用的延迟+丢包叠加效应", "score": 4}], "tool_process": {"ideal_call_count": 10, "valuable_actions": ["RUM 查看入口/API 5xx 分布", "查看 new-gatewayservice 日志", "kubectl get networkchaos", "kubectl get deploy 查看 env 配置", "对比延迟值和超时值"]}}, "rubrics": {"positive_points_total": 97, "negative_points_total": -56, "ideal_call_count": 10, "rubrics": [{"criterion": "描述了观察到的故障现象（入口/API 偶发 502/504，刷新可能恢复，商品浏览多数正常）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 RUM 确认真实用户侧 5xx/接口耗时尖刺和影响范围", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了腾讯云 new-gatewayservice 日志，发现跨云连接超时或 reset", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了腾讯云 NetworkChaos 实验列表", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 gw-to-aws-order-delay（延迟 400ms, jitter 150ms）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 NetworkChaos 的具体参数（delay 400ms, jitter 150ms, correlation 70）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 gw-to-aws-order-loss（丢包 12%）", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 NetworkChaos loss 的具体参数（loss 12%, correlation 50）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了腾讯云 StressChaos 实验列表", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 tencent-global-cpu-stress（mode: all, load: 50）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 new-gatewayservice 环境变量", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 UPSTREAM_TIMEOUT=500ms", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 MAX_RETRIES=0", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "对比延迟值（400±150ms=550ms）和超时值（500ms），判断超时配置不合理", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别 StressChaos 是干扰项，CPU 压力不是主因", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "解释延迟+抖动（550ms）超过超时阈值（500ms）导致偶发失败", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "解释 MAX_RETRIES=0 禁用重试是放大器", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确指出根因是跨云延迟/抖动/丢包与 gateway timeout/retry 配置叠加，而不是 StressChaos CPU 压力", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "说明延迟、丢包和超时收紧三者叠加才能导致问题，单独处理一层不足以闭环", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出了修复建议：删除 NetworkChaos gw-to-aws-order-delay", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：删除 NetworkChaos gw-to-aws-order-loss", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：恢复 UPSTREAM_TIMEOUT 和 MAX_RETRIES", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 -> 收集证据 -> 定位根因 -> 修复建议", "points": 4, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "模型能够把 gateway、NetworkChaos、env 配置串成一条证据链", "points": 4, "tags": ["level:cluster", "cluster:queueing_reasoning", "axis:accuracy"]}, {"criterion": "模型会跨云对比，确认异常范围主要落在腾讯云 gateway 到 AWS 链路", "points": 4, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型不被 StressChaos 误导，能继续检查 NetworkChaos 和 env 配置", "points": 4, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（RUM、NetworkChaos、env、logs），而非猜测", "points": 4, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "模型能正确处理多层叠加故障而非只停在单点", "points": 4, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 10 次）", "points": 3, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 13 次）", "points": 1, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 15 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "只看 StressChaos 就下结论是 CPU 压力问题", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只看 NetworkChaos 就下结论，不对比延迟值和超时值", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 env 配置（UPSTREAM_TIMEOUT、MAX_RETRIES）", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "遗漏 NetworkChaos gw-to-aws-order-loss", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "把问题误判为 CDN 配置或 frontend 问题", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只重启 gateway Pod，不处理 Chaos 和 env", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "把概率性超时说成 100% 确定性失败", "points": -2, "tags": ["level:example", "axis:normality"]}, {"criterion": "定位到错误云（非腾讯云）", "points": -5, "tags": ["level:example", "axis:accuracy"]}]}, "prompt": "有用户反馈访问商城时偶尔打不开，刷新一下又好了。大部分时候浏览商品没问题，但走到下单那一步更容易出错。不是所有人都遇到，时好时坏。"}
{"canary": "emptydir-log-eviction", "tags": ["发现", "存储与资源", "emptyDir 写爆磁盘", "日志爆量挤占磁盘", "hard", "multi-cloud"], "case": {"id": "emptydir-log-eviction", "title": "日志风暴：emptyDir 写满导致 Pod 驱逐", "inject_script": "./challenge-32-emptydir-log-eviction/inject.sh", "recover_script": "./challenge-32-emptydir-log-eviction/recover.sh", "fault_phenomenon": "今天开始结账功能不稳定，有时候能成功，有时候直接报错。页面能打开，商品也能看，就是最后一步结账容易出问题。刷新几次偶尔能成功。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，并结合 RUM、K8s events 和日志表现排查。注意：本题的表象在 checkout 路径，但根因需要看 Pod 的 volume 和 env 配置。"}, "ideal_answer": {"fault_info": {"root_cause": "阿里云 checkoutservice 被配置了 debug 日志级别并输出到文件，同时挂载了一个 30Mi 的 emptyDir volume 到 /var/log/checkout，并添加 log-filler sidecar 持续向该目录写入日志。StressChaos 让 checkoutservice 产生更多资源压力/日志背景，debug 文件日志和 log-filler 很快把 30Mi emptyDir 写满，导致 Pod 被 eviction due to ephemeral storage。NetworkChaos（checkoutservice -> mysql 80ms 延迟）是干扰项。", "faulty_clouds": ["alicloud"], "affected_clouds": ["alicloud"], "fault_location": {"module": "checkoutservice + storage", "file_path": null, "function_or_config": "emptyDir sizeLimit=30Mi + volumeMount=/var/log/checkout + log-filler sidecar + LOG_LEVEL=debug + LOG_OUTPUT=file + LOG_PATH=/var/log/checkout + StressChaos", "description": "checkoutservice 的 debug 文件日志和 log-filler sidecar 把已挂载到 /var/log/checkout 的 30Mi emptyDir 写满，Pod 被 eviction。NetworkChaos 是干扰项。"}}, "reasoning_process": {"observed_symptoms": ["RUM 中 checkout 相关接口错误率升高。", "checkoutservice Pod 有时变成 Evicted 或 CrashLoopBackOff。", "kubectl describe pod 能看到 evicted due to ephemeral storage 事件。", "APM 链路并不总是完整。"], "key_evidence": [{"source": "RUM 控制台", "content": "checkout 相关接口错误率升高。", "conclusion": "用户侧影响集中在 checkout 路径。"}, {"source": "kubectl --context=alicloud -n seat-1 describe pod -l app=checkoutservice", "content": "evicted due to ephemeral storage", "conclusion": "Pod 被驱逐是因为临时存储耗尽。"}, {"source": "kubectl --context=alicloud -n seat-1 get deploy checkoutservice -o yaml", "content": "emptyDir volume (sizeLimit=30Mi, mountPath=/var/log/checkout) + LOG_LEVEL=debug + LOG_OUTPUT=file + LOG_PATH=/var/log/checkout + log-filler sidecar", "conclusion": "日志输出到 30Mi 的 emptyDir，debug 级别会产生大量日志。"}, {"source": "kubectl --context=alicloud -n seat-1 get stresschaos", "content": "checkout-log-stress (CPU 2 workers, load: 65)", "conclusion": "StressChaos 让 checkoutservice 产生更多请求/日志。"}, {"source": "kubectl --context=alicloud -n seat-1 get networkchaos", "content": "checkout-mysql-delay (80ms)", "conclusion": "这是干扰项，网络延迟不是主因。"}], "causal_chain": ["checkoutservice 被配置 LOG_LEVEL=debug + LOG_OUTPUT=file + LOG_PATH=/var/log/checkout。", "checkoutservice 挂载了 sizeLimit=30Mi 的 emptyDir volume 到 /var/log/checkout。", "StressChaos 和 log-filler sidecar 让 checkoutservice 的日志目录持续写入。", "debug 文件日志和 log-filler 很快把 30Mi emptyDir 写满。", "K8s 驱逐 Pod (evicted due to ephemeral storage)。", "NetworkChaos 是干扰项，容易被误判为网络问题。"]}, "proposed_resolution": {"suggestion": "删除 Chaos 对象，恢复 checkoutservice 配置（移除 emptyDir volume、LOG_LEVEL、LOG_OUTPUT）。长期建议包括合理设置日志级别、监控 emptyDir 使用率、设置合理的 ephemeral storage limit。", "fix_steps": ["kubectl --context=alicloud -n seat-1 delete stresschaos checkout-log-stress", "kubectl --context=alicloud -n seat-1 delete networkchaos checkout-mysql-delay", "kubectl --context=alicloud -n seat-1 set env deploy/checkoutservice LOG_LEVEL- LOG_OUTPUT- LOG_PATH-", "kubectl --context=alicloud -n seat-1 patch deploy/checkoutservice --type=json -p '<移除 log-filler sidecar 与 log-volume volumeMount>'", "kubectl --context=alicloud -n seat-1 set volume deploy/checkoutservice --remove --name=log-volume", "kubectl --context=alicloud -n seat-1 rollout restart deploy/checkoutservice"]}, "confidence": 0.93, "special_points": [{"point": "必须识别 NetworkChaos 是干扰项而非主因", "score": 5}, {"point": "需要看 Pod events 才能发现 eviction 原因", "score": 5}, {"point": "需要回溯 deploy 配置发现 emptyDir + 日志级别", "score": 5}, {"point": "理解 debug 日志 + StressChaos 叠加导致存储耗尽", "score": 4}], "tool_process": {"ideal_call_count": 10, "valuable_actions": ["RUM 查看 checkout 错误率", "kubectl describe pod 查看 eviction 原因", "kubectl get deploy 查看 volume 和 env 配置", "kubectl get stresschaos,networkchaos", "判断主因和干扰"]}}, "rubrics": {"positive_points_total": 98, "negative_points_total": -55, "ideal_call_count": 10, "rubrics": [{"criterion": "描述了观察到的故障现象（checkout 失败、Pod 有时 Evicted/CrashLoopBackOff）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 RUM 确认用户侧 checkout 错误率升高", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "kubectl describe pod 发现 evicted due to ephemeral storage", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 checkoutservice Deployment 的 volume 配置", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 emptyDir volume (sizeLimit=30Mi)", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 emptyDir 的 sizeLimit=30Mi", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了 checkoutservice 环境变量", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 LOG_LEVEL=debug", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 LOG_OUTPUT=file", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了阿里云 StressChaos 实验列表", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 checkout-log-stress（CPU 2 workers, load 65）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了阿里云 NetworkChaos 实验列表", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 checkout-mysql-delay（80ms）", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别 NetworkChaos 是干扰项，网络延迟不是主因", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "解释 debug 日志 + StressChaos 叠加导致 emptyDir 写满", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "解释 emptyDir 写满导致 Pod 被 eviction 的机制", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确指出根因是存储压力，而不是网络延迟或 CPU 压力", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "说明 debug 日志级别、StressChaos 和 emptyDir 三者叠加才能导致问题", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出了修复建议：删除 StressChaos checkout-log-stress", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：删除 NetworkChaos checkout-mysql-delay", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：移除 emptyDir volume、恢复 LOG_LEVEL/LOG_OUTPUT", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 -> 收集证据 -> 定位根因 -> 修复建议", "points": 4, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "模型能够把 Pod eviction、emptyDir、日志级别、StressChaos 串成一条证据链", "points": 5, "tags": ["level:cluster", "cluster:queueing_reasoning", "axis:accuracy"]}, {"criterion": "模型不被 NetworkChaos 误导，能继续检查 volume 和 env 配置", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（RUM、describe pod、deploy 配置、Chaos），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "模型能正确处理多层叠加故障而非只停在单点", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "模型能够跨云排查，确认异常范围主要落在阿里云 checkoutservice", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 10 次）", "points": 3, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 13 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 15 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "只看 NetworkChaos 就下结论是网络问题", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 Pod events（evicted due to ephemeral storage）", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未检查 deploy 的 volume 配置（emptyDir）", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未检查 env 配置（LOG_LEVEL、LOG_OUTPUT）", "points": -7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误判为 OOM/CPU 限流", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只重启 Pod，不处理 Chaos 和配置", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "把概率性失败说成 100% 确定性崩溃", "points": -4, "tags": ["level:example", "axis:normality"]}]}, "prompt": "今天开始结账功能不稳定，有时候能成功，有时候直接报错。页面能打开，商品也能看，就是最后一步结账容易出问题。刷新几次偶尔能成功。"}
{"canary": "clock-skew-payment-signature", "tags": ["定位", "跨云一致性与容量稳定性", "跨云网络延迟", "时间窗口与时区不一致", "hard", "multi-cloud"], "case": {"id": "clock-skew-payment-signature", "title": "跨云时间漂移：支付签名间歇失效", "inject_script": "./challenge-33-clock-skew-payment-signature/inject.sh", "recover_script": "./challenge-33-clock-skew-payment-signature/recover.sh", "fault_phenomenon": "用户反馈下单时偶尔提示\"验证失败\"或\"请重试\"，不是每次都坏。浏览商品、加购物车都正常，就是最后提交订单那一步会撞上。重启服务也没用。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。本题必须结合 RUM 和 APM：RUM 用于确认结账用户侧影响，APM 用于定位 paymentservice-aws-direct 的异常链路，再结合 K8s/节点时间证据闭环。"}, "ideal_answer": {"fault_info": {"root_cause": "AWS paymentservice Pod 被 TimeChaos 注入时间漂移，导致支付签名/JWT/时间窗口校验失败，并造成 APM 中 paymentservice-aws-direct span 时间戳异常。根因是跨云时间不一致，不是 Secret 本身过期，也不是 checkoutservice 重启能解决的问题。", "faulty_clouds": ["aws"], "affected_clouds": ["alicloud", "aws"], "fault_location": {"module": "chaos-mesh + paymentservice runtime time", "file_path": null, "function_or_config": "TimeChaos: payment-clock-skew, selector app=paymentservice, timeOffset=-10m", "description": "AWS paymentservice 进程时间被偏移，导致依赖时间窗口的支付校验失败。RUM 和 APM 都是必需证据：RUM 确认结账影响，APM 定位 paymentservice 时间异常。"}}, "reasoning_process": {"observed_symptoms": ["RUM 中结账接口错误率升高，商品浏览和购物车基本正常。", "APM 中 paymentservice-aws-direct 相关 span 集中异常。", "日志错误像 token expired、not yet valid、timestamp skew 或 signature window invalid。", "checkoutservice 重启后问题仍存在。", "RUM 请求时间与 APM span 时间戳存在分钟级差异。"], "key_evidence": [{"source": "RUM 控制台", "content": "错误集中在 checkout/payment 路径，普通浏览链路基本正常。", "conclusion": "用户侧影响集中在结账支付，而非全站故障。"}, {"source": "APM 控制台", "content": "paymentservice-aws-direct span 出现错误、时间戳错位或与上游 span 时间关系异常。", "conclusion": "异常集中在 AWS paymentservice，而不是前端或商品服务。"}, {"source": "kubectl --context=aws -n seat-1 get timechaos", "content": "payment-clock-skew 作用于 app=paymentservice，timeOffset=-10m。", "conclusion": "paymentservice 进程时间被注入偏移。"}, {"source": "RUM/APM 时间戳对比", "content": "RUM 请求时间与 APM span 时间存在分钟级差异，paymentservice-aws-direct span 时间戳异常。", "conclusion": "跨云时间不一致。"}, {"source": "paymentservice 日志", "content": "出现 signature expired / timestamp skew / token not yet valid 类错误。", "conclusion": "支付失败与时间窗口校验相关。"}], "causal_chain": ["TimeChaos 修改 AWS paymentservice 进程看到的时间。", "checkoutservice 发起支付请求时，签名或 token 时间戳按真实时间生成。", "paymentservice 用偏移后的时间校验，判断请求过期或尚未生效。", "支付失败导致结账失败。", "APM span 时间关系异常，RUM 中结账接口错误率升高。", "RUM 请求时间与 APM span 时间戳存在分钟级差异。"]}, "proposed_resolution": {"suggestion": "删除 TimeChaos，恢复 paymentservice 时间。长期建议增加跨云 NTP/时钟偏移监控、签名校验错误分类告警、APM span 时间异常检测。", "fix_steps": ["kubectl --context=aws -n seat-1 delete timechaos payment-clock-skew", "等待 paymentservice 进程时间恢复，并对比 alicloud/tencent/aws 关键 Pod date 或 APM span 时间戳，确认三云时间恢复一致。"]}, "confidence": 0.9, "special_points": [{"point": "必须使用 RUM 确认用户侧影响集中在结账支付", "score": 4}, {"point": "必须使用 APM 定位到 paymentservice-aws-direct 时间/错误异常", "score": 5}, {"point": "不能把 Secret/token 本身误判为根因", "score": 4}, {"point": "RUM/APM 时间戳对比是闭环证据", "score": 4}], "tool_process": {"ideal_call_count": 12, "valuable_actions": ["RUM 查看 checkout/payment 错误率", "APM 查看 paymentservice-aws-direct span", "kubectl get timechaos", "RUM/APM 时间戳对比", "查看 paymentservice 日志中的时间窗口错误"]}}, "rubrics": {"positive_points_total": 103, "negative_points_total": -60, "ideal_call_count": 12, "rubrics": [{"criterion": "描述结账支付失败且商品/购物车基本正常的现象", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 RUM 确认错误集中在 checkout/payment 用户路径", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 APM 查看 paymentservice-aws-direct span", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "从 APM 中识别 paymentservice 时间戳错位或错误集中", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看 paymentservice 日志并识别 timestamp skew/token time window 类错误", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看 AWS TimeChaos 实验列表", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 payment-clock-skew 作用于 paymentservice", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 TimeChaos 的具体参数（timeOffset=-10m）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "通过 RUM/APM 时间戳对比，发现 AWS paymentservice 时间偏移", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "对比 RUM 请求时间与 APM span 时间，确认时间戳不一致", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确指出根因是 AWS paymentservice 时间漂移，不是 Secret 本身过期", "points": 8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "解释时间漂移如何导致签名/JWT/时间窗口校验失败", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "说明 checkoutservice 发起请求时签名按真实时间生成，但 paymentservice 用偏移时间校验", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "说明重启 checkoutservice 不能解决问题（时间偏移在 paymentservice 侧）", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "理解 APM span 时间戳异常是时间偏移的副作用，不是独立故障", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出了修复建议：删除 TimeChaos payment-clock-skew", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了恢复步骤：等待时间恢复 + 对比三云 Pod date", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了后续建议（跨云 NTP/时钟偏移监控、签名校验错误分类告警）", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 -> 收集证据 -> 定位根因 -> 修复建议", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 12 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 15 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 18 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "模型能够跨云比对时间，用 RUM/APM 时间戳作为闭环证据", "points": 5, "tags": ["level:cluster", "cluster:cross_cloud_diagnosis", "axis:completeness"]}, {"criterion": "模型不被 Secret/token 过期的表象误导，能继续检查时间偏移", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（RUM、APM、TimeChaos），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "模型理解时间偏移对签名/校验的影响机制，而非停留在表面错误", "points": 5, "tags": ["level:cluster", "cluster:time_sensitivity", "axis:accuracy"]}, {"criterion": "仅用 K8s/Chaos 对象下结论，未使用 RUM 和 APM", "points": -10, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只使用 RUM，不使用 APM 定位 paymentservice", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误判为 Secret/token 真实过期并建议轮换密钥", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只重启 checkoutservice 或 paymentservice 就认为能解决", "points": -6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未做跨云时间比对，直接假设时间一致", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "遗漏 TimeChaos 对象，未检查 chaos 实验列表", "points": -4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "仅查看单一云就给出全局诊断", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "把概率性故障说成 100% 确定性失败，或给出完全不合理的根因判断", "points": -4, "tags": ["level:example", "axis:normality"]}]}, "prompt": "用户反馈下单时偶尔提示\"验证失败\"或\"请重试\"，不是每次都坏。浏览商品、加购物车都正常，就是最后提交订单那一步会撞上。重启服务也没用。"}
{"canary": "redis-hotkey-eviction", "tags": ["定位", "中间件与依赖", "缓存热点 key", "I/O 延迟", "连接池耗尽", "hard", "multi-cloud"], "case": {"id": "redis-hotkey-eviction", "title": "Redis 热 key + I/O 延迟：eviction 导致购物车偶发丢失", "inject_script": "./challenge-34-redis-hotkey-eviction/inject.sh", "recover_script": "./challenge-34-redis-hotkey-eviction/recover.sh", "fault_phenomenon": "用户说购物车不太靠谱：有时候加了商品刷新一下就没了，有时候结账提示购物车是空的。不是每次都这样，概率性的。商品浏览和支付都没问题。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，并结合 RUM、服务日志和端到端业务表现排查。注意：本题有 IOChaos、NetworkChaos 和 env 配置变化三层因素，需要对比延迟值、超时配置和 eviction 策略才能定位完整根因。购物车 key 丢失与 Redis eviction 策略直接相关。"}, "ideal_answer": {"fault_info": {"root_cause": "腾讯云 redis-cart 被注入 IOChaos（/data 目录 I/O 延迟 250ms），同时 cartservice 的 REDIS_CONNECT_TIMEOUT 被收紧到 180ms、CACHE_TTL=0（禁用/绕过本地缓存），redis-cart 本体被真实设置为 maxmemory=32mb、maxmemory-policy=allkeys-lru，并添加 redis-hotkey-writer sidecar 持续写入购物车热 key 触发 eviction。Redis I/O 延迟（250ms）超过连接超时（180ms），导致偶发超时失败；allkeys-lru 在内存压力下驱逐购物车 key，导致购物车数据丢失。NetworkChaos（cartservice -> redis-cart 30ms 延迟）是干扰项。", "faulty_clouds": ["tencent"], "affected_clouds": ["tencent", "alicloud"], "fault_location": {"module": "redis + cartservice", "file_path": null, "function_or_config": "IOChaos: redis-io-delay (250ms) + cartservice env: REDIS_CONNECT_TIMEOUT=180ms, CACHE_TTL=0 + redis-cart args: --maxmemory 32mb --maxmemory-policy allkeys-lru + redis-hotkey-writer sidecar", "description": "Redis I/O 延迟超过连接超时 + redis-cart maxmemory-policy/allkeys-lru 配置和 hotkey writer 触发 eviction，导致购物车操作偶发失败和数据丢失。NetworkChaos 是干扰项。"}}, "reasoning_process": {"observed_symptoms": ["RUM 中 cart API 错误率和耗时波动。", "购物车商品偶发消失或结账提示 cart empty。", "商品浏览和支付链路大多正常。", "cartservice 日志出现超时错误，但不是每次都失败。"], "key_evidence": [{"source": "RUM 控制台", "content": "购物车相关 API 错误率和耗时波动，影响集中在 cart 路径。", "conclusion": "用户侧异常集中在购物车。"}, {"source": "kubectl --context=tencent -n seat-1 logs deploy/cartservice", "content": "出现 timeout、connection timed out 等超时错误。", "conclusion": "购物车操作因为超时而失败。"}, {"source": "kubectl --context=tencent -n seat-1 get iochaos", "content": "redis-io-delay (delay: 250ms, volumePath: /data)", "conclusion": "Redis I/O 被注入 250ms 延迟。"}, {"source": "kubectl --context=tencent -n seat-1 get deploy cartservice -o jsonpath='{.spec.template.spec.containers[0].env}'", "content": "REDIS_CONNECT_TIMEOUT=180ms, CACHE_TTL=0；redis-cart 启动参数包含 --maxmemory 32mb --maxmemory-policy allkeys-lru，并存在 redis-hotkey-writer sidecar", "conclusion": "连接超时（180ms）比 I/O 延迟（250ms）短，缓存被禁用，且 eviction 策略被改为 allkeys-lru。"}, {"source": "kubectl --context=tencent -n seat-1 get networkchaos", "content": "cart-redis-net-delay (30ms)", "conclusion": "这是干扰项，网络延迟不是主因。"}], "causal_chain": ["IOChaos 注入 redis-cart /data 目录 I/O 延迟 250ms。", "cartservice 的 REDIS_CONNECT_TIMEOUT 被收紧到 180ms。", "CACHE_TTL=0 禁用缓存，每次操作都访问 Redis（放大热 key 效应）。", "redis-cart 本体启用 allkeys-lru，使 Redis 在内存压力下驱逐购物车 key。", "Redis I/O 延迟（250ms）超过连接超时（180ms）→ 偶发超时失败。", "热 key 被 eviction 驱逐 → 购物车数据丢失。", "NetworkChaos 是干扰项，容易被误判为网络问题。"]}, "proposed_resolution": {"suggestion": "删除 IOChaos 和 NetworkChaos 对象，恢复 cartservice env 配置（包括 REDIS_MAXMEMORY_POLICY）。长期建议包括合理设置连接超时（应大于 I/O 延迟上限）、监控 Redis I/O 延迟和 eviction 指标、避免禁用缓存、合理配置 maxmemory-policy。", "fix_steps": ["kubectl --context=tencent -n seat-1 delete iochaos redis-io-delay", "kubectl --context=tencent -n seat-1 delete networkchaos cart-redis-net-delay", "kubectl --context=tencent -n seat-1 patch deploy/redis-cart --type=json -p '<移除 redis-hotkey-writer sidecar 和 redis-cart maxmemory/allkeys-lru args>'", "kubectl --context=tencent -n seat-1 set env deploy/cartservice REDIS_CONNECT_TIMEOUT- CACHE_TTL-", "kubectl --context=tencent -n seat-1 rollout restart deploy/cartservice", "kubectl --context=tencent -n seat-1 rollout restart deploy/redis-cart"]}, "confidence": 0.92, "special_points": [{"point": "必须识别 NetworkChaos 是干扰项而非主因", "score": 5}, {"point": "需要对比 IOChaos 延迟值（250ms）和 env 超时值（180ms）", "score": 5}, {"point": "CACHE_TTL=0 禁用缓存是放大器", "score": 4}, {"point": "识别 REDIS_MAXMEMORY_POLICY=allkeys-lru 与热 key eviction 的关系", "score": 5}, {"point": "理解 I/O 延迟 + 超时竞态 + eviction 策略的三层叠加效应", "score": 5}], "tool_process": {"ideal_call_count": 10, "valuable_actions": ["RUM 查看 cart API", "查看 cartservice 日志", "kubectl get iochaos", "kubectl get deploy 查看 env 配置", "对比 I/O 延迟值和超时值", "识别 eviction 策略配置"]}}, "rubrics": {"positive_points_total": 98, "negative_points_total": -53, "ideal_call_count": 10, "rubrics": [{"criterion": "描述了观察到的故障现象（购物车操作偶发失败、cart empty、商品浏览和支付正常）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 RUM 确认异常集中在 cart API", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 cartservice 日志并发现超时错误", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了腾讯云 IOChaos 实验列表", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 redis-io-delay（delay 250ms, volumePath /data）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 IOChaos 的具体参数（delay 250ms, percent 80）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了 cartservice 环境变量", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 REDIS_CONNECT_TIMEOUT=180ms", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 CACHE_TTL=0", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 redis-cart 启动参数 --maxmemory 32mb --maxmemory-policy allkeys-lru", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解 redis-cart allkeys-lru 策略和 hotkey writer 内存压力会驱逐购物车 key，导致购物车数据丢失", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了腾讯云 NetworkChaos 实验列表", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 cart-redis-net-delay（30ms）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "对比 I/O 延迟值（250ms）和超时值（180ms），判断超时配置不合理", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别 NetworkChaos 是干扰项，30ms 网络延迟不是主因", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "解释 I/O 延迟（250ms）超过连接超时（180ms）导致偶发失败", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "解释 CACHE_TTL=0 放大 Redis 访问；redis-cart maxmemory=32mb/allkeys-lru 和 hotkey writer 在内存压力下驱逐购物车 key", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确指出根因是 I/O 延迟+超时竞态，而不是网络延迟", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "说明 I/O 延迟、超时配置和缓存禁用三者叠加才能导致问题", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出了修复建议：删除 IOChaos redis-io-delay", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：删除 NetworkChaos cart-redis-net-delay", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了修复建议：恢复 REDIS_CONNECT_TIMEOUT、CACHE_TTL，并移除 redis-cart maxmemory/allkeys-lru 参数和 hotkey writer sidecar", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 -> 收集证据 -> 定位根因 -> 修复建议", "points": 4, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "模型能够把 IOChaos、env 配置、NetworkChaos 串成一条证据链", "points": 5, "tags": ["level:cluster", "cluster:queueing_reasoning", "axis:accuracy"]}, {"criterion": "模型不被 NetworkChaos 误导，能继续检查 IOChaos 和 env 配置", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（RUM、IOChaos、env、logs），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "模型能正确处理多层叠加故障而非只停在单点", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 10 次）", "points": 3, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 13 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 15 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "只看 NetworkChaos 就下结论是网络延迟问题", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 IOChaos 对象", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未检查 env/启动参数配置（REDIS_CONNECT_TIMEOUT、CACHE_TTL、redis-cart maxmemory-policy）", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "把本题答成缓存穿透或 TTL 过短", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只重启 cartservice 或 redis-cart，不处理 Chaos 和配置", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "把概率性超时说成 100% 确定性失败", "points": -4, "tags": ["level:example", "axis:normality"]}, {"criterion": "定位到错误云（非腾讯云）", "points": -5, "tags": ["level:example", "axis:accuracy"]}]}, "prompt": "用户说购物车不太靠谱：有时候加了商品刷新一下就没了，有时候结账提示购物车是空的。不是每次都这样，概率性的。商品浏览和支付都没问题。"}
{"canary": "rum-apm-regional-reversal", "tags": ["定位", "可观测系统自身", "RUM数据异常", "指标缺失", "网络延迟", "hard", "multi-cloud"], "case": {"id": "rum-apm-regional-reversal", "title": "RUM/APM 联合定位：商品页延迟与噪声干扰", "inject_script": "./challenge-35-rum-apm-regional-reversal/inject.sh", "recover_script": "./challenge-35-rum-apm-regional-reversal/recover.sh", "fault_phenomenon": "运营发现有一批用户反馈商品详情页打开特别慢，等好久才加载出来。但这批用户还没走到结账，只是在看商品。首页和购物车页没收到投诉。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。本题必须依靠 RUM 与 APM 联合排查：RUM 用于确认受影响页面和用户切片，APM 用于确认后端首个慢点。K8s/Chaos 对象中有多个噪声干扰，必须结合 APM 服务列表才能区分根因和噪声。"}, "ideal_answer": {"fault_info": {"root_cause": "腾讯云 productcatalogservice 被注入 450ms 网络延迟，导致商品详情页 API 响应变慢，RUM 中 /product/* 页面 LCP/TTFB 升高。另外 3 个 NetworkChaos 对象（exp-a1b2 adservice、exp-e5f6 shippingservice、exp-g7h8 emailservice）是噪声，它们命中的是 seat-1 中真实存在的服务，但这些服务不在商品页调用链上，且未接入 APM，因此 APM 拓扑中看不到它们。选手必须 RUM→APM→kubectl 三步联合定位。", "faulty_clouds": ["unknown"], "affected_clouds": ["alicloud", "tencent"], "fault_location": {"module": "chaos-mesh + observability", "file_path": null, "function_or_config": "NetworkChaos: exp-x7k2 (根因), exp-a1b2/exp-e5f6/exp-g7h8 (噪声)", "description": "4 个 NetworkChaos 对象全部指向 seat-1 真实服务，但只有 productcatalogservice 在商品页 APM 链路上。其余 3 个服务（adservice/shippingservice/emailservice）未接入 APM，不在商品页链路，是噪声干扰。"}}, "reasoning_process": {"observed_symptoms": ["RUM 中 /product/* 页面 LCP 和 TTFB 明显升高，首页和购物车页基本正常。", "APM 中 productcatalogservice-tencent-direct 的 span 耗时约 450ms。", "APM 拓扑中只有 5 个服务（frontend/productcatalogservice/checkoutservice/recommendationservice/paymentservice），噪声服务不在其中。", "kubectl get networkchaos 显示 4 个对象，全部指向 seat-1 中真实存在的服务。"], "key_evidence": [{"source": "RUM 控制台（页面维度）", "content": "/product/* 页面 LCP ~10s+、TTFB ~9s+，首页和购物车页基本正常。", "conclusion": "用户影响集中在商品详情页，不是全站或结账链路。"}, {"source": "RUM 控制台（用户/地域切片）", "content": "商品页 API 慢与 LCP 变差同步出现，异常集中在商品页用户。", "conclusion": "需要从用户侧切片确认影响范围。"}, {"source": "APM 拓扑", "content": "frontend-aliyun-direct → productcatalogservice-tencent-direct 链路耗时高（~450ms），其他链路基本正常。", "conclusion": "商品页后端首个慢点是 productcatalogservice。"}, {"source": "APM 服务列表", "content": "只有 5 个服务注册：frontend、productcatalogservice、checkoutservice、recommendationservice、paymentservice。adservice/shippingservice/emailservice 未接入 APM。", "conclusion": "APM 中看不到噪声服务，只能看到 productcatalogservice 异常。"}, {"source": "kubectl -n seat-1 get networkchaos（三云汇总）", "content": "4 个对象：exp-x7k2 (productcatalogservice, 450ms, tencent)、exp-a1b2 (adservice, 380ms, alicloud)、exp-e5f6 (shippingservice, 320ms, alicloud)、exp-g7h8 (emailservice, 400ms, tencent)。", "conclusion": "4 个 chaos 都指向真实服务，但只有 productcatalogservice 在 APM 商品页链路上。"}], "causal_chain": ["exp-x7k2 对 productcatalogservice 注入 450ms 网络延迟。", "frontend 访问 productcatalogservice 时被延迟拖慢，商品详情页 API 响应变慢。", "RUM 中 /product/* 页面 LCP/TTFB 升高，用户感知商品页加载慢。", "APM 中 productcatalogservice-tencent-direct span 耗时约 450ms，是商品链路首个慢点。", "exp-a1b2/exp-e5f6/exp-g7h8 命中 adservice/shippingservice/emailservice，这些服务不在商品页链路且未接入 APM，APM 拓扑中看不到。", "因此根因是 exp-x7k2，其余 3 个是噪声干扰。"]}, "proposed_resolution": {"suggestion": "删除 productcatalogservice 的根因 NetworkChaos。后续建议将关键服务接入 APM、建立 RUM 页面维度与 APM 拓扑联合看板。", "fix_steps": ["kubectl --context=tencent -n seat-1 delete networkchaos exp-x7k2", "可选：删除其余噪声 chaos（exp-a1b2、exp-e5f6、exp-g7h8）。"]}, "confidence": 0.95, "special_points": [{"point": "必须先用 RUM 确认受影响页面是 /product/*，不能只看全局指标", "score": 5}, {"point": "必须用 APM 定位 productcatalogservice 是商品链路首个慢点", "score": 5}, {"point": "4 个 chaos 对象全部指向真实服务，无法通过 kubectl 排除噪声", "score": 5}, {"point": "噪声服务（adservice/shippingservice/emailservice）未接入 APM，APM 拓扑中看不到", "score": 4}], "tool_process": {"ideal_call_count": 12, "valuable_actions": ["RUM 按页面维度查看 /product/* LCP/TTFB", "RUM 按用户/地域切片确认影响范围", "APM 查看服务列表（确认哪些服务接入了 APM）", "APM 查看拓扑中 frontend → productcatalogservice 链路", "kubectl get networkchaos（三云）查看所有 chaos 对象", "对比 APM 服务列表和 chaos 对象的 target app，锁定唯一在链路上的根因"]}}, "rubrics": {"positive_points_total": 104, "negative_points_total": -57, "ideal_call_count": 12, "rubrics": [{"criterion": "描述商品详情页加载慢，用户反馈集中在商品页而非首页或购物车", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 RUM 按页面维度查看，确认 /product/* 的 LCP/TTFB 明显升高", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 RUM 按用户/地域/来源切片，确认异常集中在商品页用户群", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 APM 查看服务列表，确认哪些服务已接入 APM（仅 5 个）", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 APM 查看拓扑，发现 frontend → productcatalogservice 链路耗时高", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别 productcatalogservice-tencent-direct span 耗时约 450ms，是商品链路首个慢点", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "使用 kubectl 查看三云 NetworkChaos 实验列表", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "观察到 4 个 NetworkChaos 对象，全部指向 seat-1 中真实存在的服务", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别出 exp-x7k2 对应 productcatalogservice（结合 APM 证据）", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "准确给出 exp-x7k2 的具体参数（delay 450ms, jitter 80ms, 腾讯云）", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "识别 exp-a1b2/exp-e5f6/exp-g7h8 分别命中 adservice/shippingservice/emailservice", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "理解噪声服务（adservice/shippingservice/emailservice）未接入 APM，APM 拓扑中看不到它们", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "说明 productcatalogservice 延迟 → 商品 API 慢 → LCP/TTFB 变差的因果链", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "解释噪声服务不在商品页调用链上，其延迟不影响 /product/* 页面", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "给出修复建议：删除 exp-x7k2（优先）", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出修复建议：可选删除其余噪声 chaos 对象", "points": 2, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：观察现象 → 收集证据 → 定位根因 → 修复建议", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "排查效率：总工具调用次数 ≤ 12（ideal_call_count × 1）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ 15（ideal_call_count × 1.25）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > 18（ideal_call_count × 1.5）", "points": -5, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "模型能够同时利用 RUM（页面维度）和 APM（服务拓扑）完成联合定位", "points": 5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "模型不被 3 个噪声 chaos 对象误导，能结合 APM 服务列表区分根因和噪声", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（RUM、APM、NetworkChaos），而非猜测", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "模型能识别出噪声服务未接入 APM，因此 APM 中看不到它们是正常的而非异常", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "模型能够同时利用 RUM 和 APM 完成从页面到服务的闭环诊断", "points": 5, "tags": ["level:cluster", "cluster:rum_apm_joint_diagnosis", "axis:context_awareness"]}, {"criterion": "模型不被噪声 chaos 对象和未接入 APM 的服务误导，能用 RUM/APM 证据区分根因", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "模型的排查过程基于实际证据（RUM、APM、NetworkChaos），而非猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "面对多个指向真实服务的 chaos 对象，模型能结合 APM 服务列表逐个排除而非猜测", "points": 5, "tags": ["level:cluster", "cluster:noise_filtering_capability", "axis:context_awareness"]}, {"criterion": "未使用 RUM，仅看 APM 或 K8s/Chaos 对象下结论", "points": -12, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未使用 APM，仅看 RUM 或 K8s 下结论", "points": -10, "tags": ["level:example", "axis:completeness"]}, {"criterion": "仅依赖 kubectl 查看 chaos 对象列表直接下结论，未做 RUM/APM 关联验证", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "把 paymentservice 或 recommendationservice 误判为商品页根因", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "把噪声服务（adservice/shippingservice/emailservice）的 chaos 误判为实际故障源", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "没有解释页面维度（RUM）和服务拓扑（APM）之间的对应关系", "points": -5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "把概率性故障说成 100% 确定性失败，或给出完全不合理的根因判断", "points": -4, "tags": ["level:example", "axis:normality"]}]}, "prompt": "运营发现有一批用户反馈商品详情页打开特别慢，等好久才加载出来。但这批用户还没走到结账，只是在看商品。首页和购物车页没收到投诉。"}
{"canary": "api-schema-compat", "tags": ["定位", "配置与密钥", "ConfigMap 内容错误", "hard", "multi-cloud"], "case": {"id": "api-schema-compat", "title": "RUM JS Error 激增：API Schema 兼容陷阱", "inject_script": "./challenge-36-api-schema-compat/inject.sh", "recover_script": "./challenge-36-api-schema-compat/recover.sh", "fault_phenomenon": "运营发现商品详情页打开后内容是空的，页面框架在但商品名称、价格、描述全部缺失。有时候会弹出脚本错误提示。首页商品卡片也显示为空。最近没有发过版本。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态。本题必须结合 RUM（JS Error 分析）和后端服务配置排查：RUM 用于确认报错页面和错误类型，K8s 配置用于定位后端变更根因。注意：APM 接口返回 200 不代表数据格式正确。"}, "ideal_answer": {"fault_info": {"root_cause": "腾讯云 productcatalogservice 的 products.json 被 ConfigMap 覆盖，字段名被篡改（name→productName, priceUsd→price, categories→tags, description→desc）。Go protobuf 的 jsonpb.Unmarshal 忽略未知字段，导致商品名、价格、描述全部为空值。前端 JS 拿到空数据，页面显示空白或脚本错误，RUM 中 JS Error 飙升。", "faulty_clouds": ["tencent"], "affected_clouds": ["tencent", "alicloud"], "fault_location": {"module": "productcatalogservice", "file_path": "products.json（通过 ConfigMap 挂载覆盖）", "function_or_config": "ConfigMap: products-wrong-schema — 字段名被篡改", "description": "products.json 字段名变更导致 protobuf 解析丢弃数据，商品信息为空。"}}, "reasoning_process": {"observed_symptoms": ["RUM 中 JS Error 突然飙升，商品详情页和首页商品卡片显示为空。", "页面能打开但商品名称、价格、描述全部缺失。", "APM 后端接口返回 200，没有服务端错误。", "前端最近没有改过代码。"], "key_evidence": [{"source": "RUM 控制台", "content": "JS Error 激增，集中在商品列表和详情页，错误类型为 TypeError（如 Cannot read property 'xxx' of undefined）。", "conclusion": "前端 JS 拿到的商品数据为空，解析时报错。"}, {"source": "APM / 后端日志", "content": "productcatalogservice 接口返回 200，响应时间正常。", "conclusion": "后端没有报错，但返回的商品数据字段可能为空。"}, {"source": "kubectl --context=tencent -n seat-1 get deploy productcatalogservice -o yaml", "content": "发现 volumeMount 挂载了 ConfigMap products-wrong-schema 到 /usr/src/app/products.json", "conclusion": "products.json 被 ConfigMap 覆盖，这是根因。"}, {"source": "kubectl --context=tencent -n seat-1 get cm products-wrong-schema -o yaml", "content": "字段名被篡改：name→productName, priceUsd→price, categories→tags, description→desc", "conclusion": "protobuf jsonpb.Unmarshal 忽略未知字段，商品数据全部丢失。"}], "causal_chain": ["products-wrong-schema ConfigMap 被创建，字段名被篡改。", "productcatalogservice 被 patch 挂载该 ConfigMap，覆盖原始 products.json。", "Go protobuf jsonpb.Unmarshal 解析时忽略未知字段（productName, price, tags, desc）。", "商品的 name、priceUsd、categories、description 全部为空值。", "gRPC 接口返回空商品数据，前端 JS 解析时报 TypeError。", "阿里云前端通过腾讯云网关调用 productcatalogservice，因此阿里云入口的商品数据也受影响。", "RUM 中 JS Error 飙升，页面商品信息显示不出来。"]}, "proposed_resolution": {"suggestion": "移除 ConfigMap 挂载，恢复原始 products.json。长期建议：商品数据配置变更需要经过前后端联调验证，不能只改后端数据格式。", "fix_steps": ["kubectl --context=tencent -n seat-1 patch deploy productcatalogservice --type='json' -p='[{\"op\":\"remove\",\"path\":\"/spec/template/spec/containers/0/volumeMounts/0\"},{\"op\":\"remove\",\"path\":\"/spec/template/spec/volumes/0\"}]'（需先确认 volumeMounts[0] 和 volumes[0] 确实是 products-volume，否则应按名称查找索引后再删除）", "kubectl --context=tencent -n seat-1 delete configmap products-wrong-schema", "kubectl --context=tencent -n seat-1 rollout restart deploy/productcatalogservice"]}, "confidence": 0.92, "special_points": [{"point": "APM 返回 200 不代表数据内容正确，需要区分'服务可用'和'数据完整'", "score": 6}, {"point": "RUM JS Error 是前端症状，根因在后端 ConfigMap 覆盖了 products.json", "score": 5}, {"point": "需要检查 deployment 的 volumeMount 才能发现 ConfigMap 挂载", "score": 5}, {"point": "protobuf jsonpb.Unmarshal 的行为：忽略未知字段，不报错", "score": 4}], "tool_process": {"ideal_call_count": 9, "valuable_actions": ["RUM 查看 JS Error 详情", "APM 查看 productcatalogservice 链路", "kubectl get deploy 查看 volumeMount", "kubectl get cm products-wrong-schema 查看内容", "对比原始 products.json 和篡改版本的字段名差异"]}}, "rubrics": {"positive_points_total": 105, "negative_points_total": -49, "ideal_call_count": 9, "rubrics": [{"criterion": "描述了观察到的故障现象（RUM JS Error 飙升、商品信息显示为空、TypeError）", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 RUM 确认 JS Error 集中在商品列表和详情页", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 JS Error 的具体类型和堆栈信息", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "查看了 APM 中 productcatalogservice 的接口状态", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "发现 APM 接口返回 200 但 RUM 显示 JS Error（数据内容为空）", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了 productcatalogservice 的 deployment 配置", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "发现 volumeMount 挂载了 ConfigMap 覆盖 products.json", "points": 6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了 products-wrong-schema ConfigMap 的内容", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出字段名被篡改（name→productName, priceUsd→price, categories→tags, description→desc）", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解 protobuf jsonpb.Unmarshal 忽略未知字段导致数据丢失的机制", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "区分了前端症状（JS Error）和后端根因（ConfigMap 覆盖 products.json）", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "查看了 NetworkChaos 对象并识别为干扰项（50ms 延迟不影响功能）", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "对比了不同页面的受影响情况（商品列表 vs 商品详情 vs 购物车）", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "理解 ConfigMap 挂载到文件路径的工作原理", "points": 3, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "不被 APM 200 响应误导，继续排查后端数据配置", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "排查过程基于 RUM + APM + K8s ConfigMap 的联合证据", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "面对 RUM JS Error + APM 200 + NetworkChaos 多信号时能正确隔离各故障", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "识别出数据格式兼容性问题（字段名变更导致解析失败）而非单纯的代码 bug", "points": 5, "tags": ["level:cluster", "cluster:api_compatibility_understanding", "axis:accuracy"]}, {"criterion": "给出了修复建议：移除 ConfigMap 挂载，恢复原始 products.json", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "修复建议包含具体操作：patch 移除 volumeMount + 删除 ConfigMap + rollout restart", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "修复建议说明预期恢复效果：移除 ConfigMap 挂载后商品信息应恢复，RUM JS Error 应回落", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了后续建议：数据配置变更需前后端联调验证", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：现象 → 证据 → 根因 → 修复", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 9 次）", "points": 3, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 12 次）", "points": 1, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 14 次）", "points": -3, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "误判为前端代码 bug（实际是后端 ConfigMap 覆盖 products.json）", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 productcatalogservice 的 volumeMount 配置", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只看 APM 200 就排除后端问题", "points": -6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "误判为网络问题或 CDN 问题", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "定位到错误云（非腾讯云）", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未使用 RUM 确认 JS Error 影响范围", "points": -4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "把概率性问题说成 100% 确定性失败", "points": -3, "tags": ["level:example", "axis:normality"]}, {"criterion": "忽略 NetworkChaos 干扰项，将其误认为主因", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未理解 protobuf 忽略未知字段的机制", "points": -4, "tags": ["level:example", "axis:accuracy"]}]}, "prompt": "运营发现商品详情页打开后内容是空的，页面框架在但商品名称、价格、描述全部缺失。有时候会弹出脚本错误提示。首页商品卡片也显示为空。最近没有发过版本。"}
{"canary": "frontend-origin-degradation", "tags": ["定位", "混沌与题面治理", "多故障叠加", "故障+烟雾弹", "hard", "single-cloud"], "case": {"id": "frontend-origin-degradation", "title": "前端 Origin 网络退化：资源加载失败与噪声干扰", "inject_script": "./challenge-37-frontend-origin-degradation/inject.sh", "recover_script": "./challenge-37-frontend-origin-degradation/recover.sh", "fault_phenomenon": "用户反馈首页和商品页打开特别慢，要等好几秒才出来，有时候图片和样式加载不出来，页面白屏或部分内容显示异常。部分入口的问题比其他入口更明显。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，并结合 RUM、APM 和日志做联合排查。RUM App Key: QVWbXcJbj1ZeOYQ0zJ，APM 实例 ID: apm-zAVfXX13n。本题需要结合 RUM/APM/MCP 多源数据综合判断。"}, "ideal_answer": {"fault_info": {"root_cause": "阿里云 frontend 被注入 NetworkChaos 延迟 600ms+抖动 200ms + 丢包 8%，导致前端 origin 响应严重退化。RUM 显示 LCP/TTFB 飙升，静态资源加载失败或出现资源错误。MCP 同时可见多个 chaos 对象（catalog 380ms、recommend 420ms、checkout-db 300ms、CPU 压力），均为噪声，不是前端资源加载失败的主因。", "faulty_clouds": ["alicloud"], "affected_clouds": ["alicloud"], "fault_location": {"module": "frontend", "file_path": null, "function_or_config": "NetworkChaos: origin-frontend-delay (600ms) + origin-frontend-loss (8%)", "description": "前端 origin 服务网络延迟和丢包，导致页面加载超时、白屏和静态资源加载失败。"}}, "reasoning_process": {"observed_symptoms": ["首页和商品页加载极慢，LCP 从 1.2s 飙升到 5s+。", "部分静态资源加载失败或出现资源错误，RUM 资源成功率下降。", "阿里云入口最明显。"], "key_evidence": [{"source": "RUM", "content": "LCP 5s+，TTFB 升高，静态资源成功率下降或资源错误增加。", "conclusion": "前端加载链路存在严重延迟和失败。"}, {"source": "APM", "content": "frontend-aliyun-direct 响应延迟 600ms+，拓扑显示前端是瓶颈。", "conclusion": "APM 确认前端服务是延迟源头。"}, {"source": "MCP", "content": "6 个 chaos 对象分布在不同服务和云上。", "conclusion": "无法仅靠 chaos 列表判断根因，需要 RUM/APM 交叉验证。"}], "causal_chain": ["NetworkChaos origin-frontend-delay 注入 600ms 延迟到 frontend。", "NetworkChaos origin-frontend-loss 叠加 8% 丢包，部分资源请求完全失败。", "RUM 监测到 LCP/TTFB 飙升和资源加载失败。", "其他 chaos（catalog/recommend/checkout-db/cpu）是噪声，不是前端资源加载失败的主因。"]}, "proposed_resolution": {"suggestion": "删除 origin-frontend-delay 和 origin-frontend-loss 两个 NetworkChaos 对象。", "fix_steps": ["kubectl --context=alicloud -n seat-1 delete networkchaos origin-frontend-delay", "kubectl --context=alicloud -n seat-1 delete networkchaos origin-frontend-loss"]}, "confidence": 0.9, "special_points": [{"point": "MCP 可见多个 chaos，必须用 RUM/APM 区分前端 origin 根因和后台噪声", "score": 6}, {"point": "RUM LCP/TTFB 是定位前端问题的关键指标", "score": 5}, {"point": "APM 拓扑可确认 frontend 是延迟源头", "score": 5}, {"point": "噪声 chaos 分布在不同云和服务，增加排查难度", "score": 4}], "tool_process": {"ideal_call_count": 12, "valuable_actions": ["RUM 查看 LCP/TTFB/资源成功率", "APM 查看 frontend 响应延迟和拓扑", "MCP 查看 chaos 对象列表", "MCP 查看 frontend deploy 状态", "对比各 chaos 对象影响的服务是否在 RUM/APM 异常链路上"]}}, "rubrics": {"positive_points_total": 97, "negative_points_total": -52, "ideal_call_count": 12, "rubrics": [{"criterion": "描述了观察到的故障现象（页面加载慢、资源加载失败/资源错误、LCP 飙升）", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 RUM 查看了 LCP/TTFB/资源成功率等性能指标", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 RUM 确认前端加载链路存在严重延迟和资源失败", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "使用 APM 查看了 frontend 服务的响应延迟", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 APM 拓扑确认 frontend 是延迟源头", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "使用 MCP 查看了 chaos 对象列表", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 6 个 chaos 对象中 origin-frontend-delay/loss 是导致 RUM 异常的根因", "points": 6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确区分了噪声 chaos（catalog/recommend/checkout-db/cpu）与根因", "points": 5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "通过 RUM 资源成功率下降和 APM 响应延迟联合定位到 frontend 是瓶颈，而非仅凭单一数据源", "points": 5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "理解前端 origin 退化会导致 RUM 资源加载失败/资源错误（不是 CDN 问题）", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "排查过程基于 RUM + APM + MCP 的证据链，而非仅靠 chaos 列表", "points": 6, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "面对 6 个 chaos 对象能正确隔离根因和噪声", "points": 5, "tags": ["level:cluster", "cluster:noise_filtering_capability", "axis:accuracy"]}, {"criterion": "理解前端延迟+丢包会导致资源加载失败、白屏或资源错误", "points": 5, "tags": ["level:cluster", "cluster:origin_degradation_understanding", "axis:accuracy"]}, {"criterion": "给出了修复建议：删除 origin-frontend-delay 和 origin-frontend-loss", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "修复建议包含具体 kubectl delete 命令", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "说明了噪声 chaos 不需要删除（不影响主路径）", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "响应按清晰结构组织：现象 → 证据 → 根因 → 修复", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "分析了 TTFB 与 LCP 的变化关系，确认前端 origin 延迟是性能下降的主因", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "注意到噪声 chaos 的跨云分布（catalog 在腾讯云），利用云分布信息辅助排除噪声", "points": 4, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 12 次）", "points": 3, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 15 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 18 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "误判为单纯 CDN 配置问题（实际是前端 origin 网络退化）", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只看 chaos 列表就下结论，不使用 RUM/APM 交叉验证", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "误判噪声 chaos（catalog/recommend/checkout-db/cpu）为根因", "points": -6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "定位到错误云（非阿里云）", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "误判为数据库或 Redis 问题", "points": -6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "把所有 6 个 chaos 都当成需要修复的问题", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未使用 RUM 工具查看前端性能数据", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未使用 APM 工具查看服务拓扑", "points": -5, "tags": ["level:example", "axis:completeness"]}]}, "prompt": "用户反馈首页和商品页打开特别慢，要等好几秒才出来，有时候图片和样式加载不出来，页面白屏或部分内容显示异常。部分入口的问题比其他入口更明显。"}
{"canary": "cdn-origin-403", "tags": ["定位", "配置与密钥", "ConfigMap 内容错误", "medium", "single-cloud"], "case": {"id": "cdn-origin-403", "title": "CDN 回源陷阱：静态资源 403 与 RUM 资源错误", "inject_script": "./challenge-38-cdn-origin-403/inject.sh", "recover_script": "./challenge-38-cdn-origin-403/recover.sh", "fault_phenomenon": "用户反馈商城页面打开后样式全乱了，图片全都不显示，但商品数据还在、购物车也能用。刷新偶尔能恢复但很快又不行。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。本题需要结合 RUM 数据和 K8s 配置排查。"}, "ideal_answer": {"fault_info": {"root_cause": "阿里云 frontend-rum-proxy 的 nginx ConfigMap 被修改，/static/ 路径被配置为返回 403 Forbidden（模拟 CDN 回源 Host 不匹配）。同时 frontend 被注入 StressChaos CPU 压力（干扰项）。静态资源加载全部失败，但 API 接口正常。", "faulty_clouds": ["alicloud"], "affected_clouds": ["alicloud"], "fault_location": {"module": "frontend-rum-proxy (nginx ConfigMap)", "file_path": null, "function_or_config": "ConfigMap: rum-nginx-config — location /static/ { return 403 }", "description": "nginx 配置中 /static/ 路径返回 403，导致所有静态资源加载失败。"}}, "reasoning_process": {"observed_symptoms": ["页面样式错乱，图片加载不出来。", "商品列表数据能正常获取，购物车操作正常。", "RUM 中资源错误率飙升，大量 /static/ 路径返回 403。", "API 接口全部正常。"], "key_evidence": [{"source": "RUM 控制台", "content": "资源错误集中在 /static/ 路径，返回 403 Forbidden。", "conclusion": "静态资源被主动拒绝，不是网络问题。"}, {"source": "kubectl --context=alicloud -n seat-1 get cm rum-nginx-config -o yaml", "content": "location /static/ { return 403 \"Forbidden: Host mismatch\"; }", "conclusion": "nginx 配置被修改，/static/ 路径返回 403。"}, {"source": "kubectl --context=alicloud -n seat-1 get stresschaos", "content": "frontend-cpu-stress (CPU 50%)", "conclusion": "这是干扰项，CPU 压力不是静态资源 403 的原因。"}], "causal_chain": ["rum-nginx-config ConfigMap 被修改，/static/ 路径返回 403。", "frontend-rum-proxy 重启后新配置生效。", "所有静态资源请求（CSS/JS/图片）被 nginx 返回 403。", "页面样式错乱、图片加载失败。", "API 请求走 / 路径正常代理到 frontend:8080，不受影响。", "StressChaos 是干扰项。"]}, "proposed_resolution": {"suggestion": "恢复 rum-nginx-config 的原始配置，移除 /static/ 的 403 规则。长期建议：CDN 回源 Host 配置需要与源站 nginx 配置保持一致，变更时需同步验证。", "fix_steps": ["kubectl --context=alicloud -n seat-1 apply -f - <<'EOF'\napiVersion: v1\nkind: ConfigMap\nmetadata:\n  name: rum-nginx-config\n  namespace: seat-1\ndata:\n  default.conf: |\n    server {\n        listen 80;\n        location / {\n            proxy_pass http://frontend:8080;\n            sub_filter '</head>' '\n        <script src=\"https://tam.cdn-go.cn/aegis-sdk/latest/aegis.min.js\"></script>\n        <script>\n          if (typeof Aegis === \"function\") {\n            var aegis = new Aegis({\n              id: \"QVWbXcJbj1ZeOYQ0zJ\",\n              uin: \"seat1\",\n              reportApiSpeed: true,\n              reportAssetSpeed: true,\n              spa: true,\n              hostUrl: \"https://rumt-zh.com\",\n              sample: 1\n            });\n          }\n        </script>\n    </head>';\n            sub_filter_once on;\n            sub_filter_types text/html;\n            proxy_set_header Host $host;\n            proxy_set_header X-Real-IP $remote_addr;\n            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;\n        }\n    }\nEOF", "kubectl --context=alicloud -n seat-1 rollout restart deploy/frontend-rum-proxy"]}, "confidence": 0.92, "special_points": [{"point": "API 正常但静态资源异常，需要区分两类请求路径", "score": 5}, {"point": "403 是主动拒绝而非超时/网络问题", "score": 4}, {"point": "需要查看 ConfigMap 才能发现 nginx 配置变更", "score": 5}, {"point": "StressChaos 是干扰项", "score": 4}], "tool_process": {"ideal_call_count": 8, "valuable_actions": ["RUM 查看资源错误", "curl 验证 /static/ 路径", "kubectl get cm rum-nginx-config", "kubectl get stresschaos"]}}, "rubrics": {"positive_points_total": 106, "negative_points_total": -54, "ideal_call_count": 8, "rubrics": [{"criterion": "描述了观察到的故障现象（样式错乱、图片加载失败、API 正常）", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 RUM 确认资源错误集中在 /static/ 路径", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "发现 /static/ 路径返回 403 Forbidden", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "区分了静态资源请求（失败）和 API 请求（正常）", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了 rum-nginx-config ConfigMap", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "发现 location /static/ { return 403 } 配置", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "查看了 StressChaos 对象并识别为干扰项", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "理解 403 是主动拒绝而非网络超时", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解 nginx 配置变更 → 静态资源 403 的因果链", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解 location /static/ 匹配规则和 return 403 的含义", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "对比了不同资源路径的访问情况（/static/ vs /api/）", "points": 3, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 403 Forbidden 是服务端主动返回而非网络层问题", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "理解 StressChaos CPU 50% 不会导致静态资源 403", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "不被 StressChaos 误导，定位到 ConfigMap 变更", "points": 5, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "排查过程基于 RUM 资源错误 + ConfigMap + StressChaos 的证据链", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "面对 ConfigMap 变更 + StressChaos 多故障时能正确隔离主因和干扰项", "points": 5, "tags": ["level:cluster", "cluster:multi_fault_handling", "axis:accuracy"]}, {"criterion": "识别出静态资源 403 是 nginx 配置问题而非 CDN/网络问题", "points": 5, "tags": ["level:cluster", "cluster:static_resource_403_understanding", "axis:accuracy"]}, {"criterion": "给出了修复建议：恢复 rum-nginx-config 原始配置", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "修复建议包含具体操作：kubectl apply 原始 ConfigMap + rollout restart", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "修复建议说明预期恢复效果：恢复 ConfigMap 后 /static/ 资源不再返回 403，页面样式和图片应恢复", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "给出了后续建议：CDN 回源 Host 需与源站配置一致", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "响应按清晰结构组织：现象 → 证据 → 根因 → 修复", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 8 次）", "points": 3, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 10 次）", "points": 1, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 12 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "误判为前端代码 bug（实际是 nginx 配置变更）", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 ConfigMap 配置", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只看 StressChaos 就下结论是 CPU 压力问题", "points": -6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "误判为 CDN 故障或网络问题", "points": -6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "定位到错误云（非阿里云）", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未使用 RUM 确认资源错误路径", "points": -4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "把 403 说成超时或网络抖动", "points": -4, "tags": ["level:example", "axis:normality"]}, {"criterion": "忽略 StressChaos 干扰项，将其误认为主因", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未理解 nginx location 匹配规则，无法定位配置问题", "points": -5, "tags": ["level:example", "axis:accuracy"]}]}, "prompt": "用户反馈商城页面打开后样式全乱了，图片全都不显示，但商品数据还在、购物车也能用。刷新偶尔能恢复但很快又不行。"}
{"canary": "timeout-budget-trap", "tags": ["定位", "混沌与题面治理", "多故障叠加", "故障+烟雾弹", "hard", "single-cloud"], "case": {"id": "timeout-budget-trap", "title": "超时预算陷阱：购物车 Redis 超时与噪声干扰", "inject_script": "./challenge-39-timeout-budget-trap/inject.sh", "recover_script": "./challenge-39-timeout-budget-trap/recover.sh", "fault_phenomenon": "用户反馈购物车页面偶发超时，结账流程卡住。有时候加购商品失败，有时候结账页面白屏。问题不是每次都复现，时好时坏。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，并结合 RUM、APM 和日志做联合排查。RUM App Key: QVWbXcJbj1ZeOYQ0zJ，APM 实例 ID: apm-zAVfXX13n。本题需要结合 RUM/APM/MCP 多源数据综合判断。"}, "ideal_answer": {"fault_info": {"root_cause": "腾讯云 cartservice 被注入 NetworkChaos 延迟 500ms，同时环境变量 REDIS_CONNECT_TIMEOUT=100ms（超时预算比注入延迟短，Redis 操作必然超时）。cartservice 不在 APM 注册中，但 checkoutservice（在 APM 注册）调用 cartservice，因此 APM 拓扑显示 checkoutservice 响应变慢。MCP 同时可见 3 个噪声对象（adservice 400ms、emailservice 350ms、inventoryservice CPU 压力），均不在购物车/结账主故障链路上。", "faulty_clouds": ["tencent"], "affected_clouds": ["tencent"], "fault_location": {"module": "cartservice", "file_path": null, "function_or_config": "NetworkChaos: cart-conn-pool-exhaustion (500ms) + env: REDIS_CONNECT_TIMEOUT=100ms", "description": "购物车服务网络延迟 + Redis 连接超时预算过短，导致购物车操作超时。"}}, "reasoning_process": {"observed_symptoms": ["购物车页面偶发超时，加购商品失败。", "结账流程卡住，页面白屏。", "多个服务间歇性延迟，看起来像网络抖动，但购物车/结账链路最明显。"], "key_evidence": [{"source": "RUM", "content": "购物车和结账页面 TTFB 升高，API 调用超时。", "conclusion": "购物车/结账链路存在性能问题。"}, {"source": "APM", "content": "checkoutservice 响应延迟增加，拓扑显示其调用 cartservice 时延迟异常。", "conclusion": "checkoutservice 的下游 cartservice 是瓶颈。"}, {"source": "MCP", "content": "cartservice REDIS_CONNECT_TIMEOUT=100ms，cart-conn-pool-exhaustion chaos 500ms。", "conclusion": "超时 100ms < 延迟 500ms，Redis 操作必然超时。"}], "causal_chain": ["NetworkChaos cart-conn-pool-exhaustion 注入 500ms 延迟到 cartservice。", "REDIS_CONNECT_TIMEOUT=100ms 使 Redis 连接在延迟注入下必然超时。", "checkoutservice（APM 可见）调用 cartservice 时被阻塞，APM 显示 checkoutservice 延迟升高。", "其他 chaos（adservice/emailservice/inventoryservice）是噪声，不影响购物车/结账主链路。"]}, "proposed_resolution": {"suggestion": "删除 cart-conn-pool-exhaustion NetworkChaos，恢复 REDIS_CONNECT_TIMEOUT 环境变量。", "fix_steps": ["kubectl --context=tencent -n seat-1 delete networkchaos cart-conn-pool-exhaustion", "kubectl --context=tencent -n seat-1 set env deploy/cartservice REDIS_CONNECT_TIMEOUT-"]}, "confidence": 0.9, "special_points": [{"point": "cartservice 不在 APM，但通过 checkoutservice 间接可见", "score": 6}, {"point": "超时值 < 延迟值是关键诊断线索", "score": 5}, {"point": "多个 chaos 对象需要 RUM/APM 才能区分根因", "score": 5}, {"point": "噪声分布在不同云和服务，增加排查难度", "score": 4}], "tool_process": {"ideal_call_count": 14, "valuable_actions": ["RUM 查看购物车/结账页面性能", "APM 查看 checkoutservice 延迟和拓扑", "MCP 查看 chaos 对象列表", "MCP 查看 cartservice deploy env", "对比超时值和延迟值的大小关系"]}}, "rubrics": {"positive_points_total": 99, "negative_points_total": -56, "ideal_call_count": 14, "rubrics": [{"criterion": "描述了观察到的故障现象（购物车超时、结账卡住、多服务间歇延迟）", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "使用 RUM 查看了购物车/结账页面的 TTFB 和 API 错误率", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 RUM 确认购物车/结账链路存在性能问题", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "使用 APM 查看了 checkoutservice 的延迟和调用拓扑", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 APM 拓扑发现 checkoutservice 调用 cartservice 时延迟异常", "points": 6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "使用 MCP 查看了 chaos 对象列表和 deploy env", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "识别出 cart-conn-pool-exhaustion（500ms）+ REDIS_CONNECT_TIMEOUT=100ms 是根因", "points": 6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "发现超时值（100ms）< 延迟值（500ms），Redis 操作必然超时", "points": 6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确区分了噪声 chaos（ad/email/inventory）与根因", "points": 5, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "理解 cartservice 不在 APM 但通过 checkoutservice 间接可观测", "points": 5, "tags": ["level:cluster", "cluster:rum_apm_joint_diagnosis", "axis:accuracy"]}, {"criterion": "排查过程基于 RUM + APM + MCP 的证据链", "points": 6, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "面对多个 chaos + 1 个 env 能正确隔离根因和噪声", "points": 5, "tags": ["level:cluster", "cluster:noise_filtering_capability", "axis:accuracy"]}, {"criterion": "理解超时配置与网络延迟的大小关系决定故障表现", "points": 5, "tags": ["level:cluster", "cluster:timeout_budget_analysis", "axis:accuracy"]}, {"criterion": "给出了修复建议：删除 chaos + 恢复 REDIS_CONNECT_TIMEOUT", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "修复建议包含具体 kubectl 命令", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "说明了噪声 chaos 不需要删除", "points": 4, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "响应按清晰结构组织：现象 → 证据 → 根因 → 修复", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "分析了 NetworkChaos 的 jitter 参数（180ms），理解实际延迟是一个范围（320ms-680ms），而非固定值", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "系统性检查了三朵云的 chaos 对象，没有只查单云就下结论", "points": 4, "tags": ["level:example", "axis:completeness"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 14 次）", "points": 3, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 18 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 21 次）", "points": -5, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "误判为普通网络抖动、DNS 问题或端口耗尽", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只看 chaos 列表就下结论，不使用 RUM/APM 交叉验证", "points": -7, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "误判噪声 chaos（ad/email/inventory）为根因", "points": -6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "定位到错误云（非腾讯云）", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未检查 cartservice 的 env 配置", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "把所有噪声 chaos 都当成需要修复的主因", "points": -5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "未使用 RUM 工具", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "未使用 APM 工具", "points": -5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "误判为数据库连接池耗尽或真实端口耗尽", "points": -5, "tags": ["level:example", "axis:accuracy"]}]}, "prompt": "用户反馈购物车页面偶发超时，结账流程卡住。有时候加购商品失败，有时候结账页面白屏。问题不是每次都复现，时好时坏。"}
{"canary": "systematic-fault-hunt", "tags": ["发现", "混沌与题面治理", "多故障叠加", "hard", "multi-cloud"], "case": {"id": "systematic-fault-hunt", "title": "多云系统性故障清点：独立故障点发现率测试", "inject_script": "./challenge-40-systematic-fault-hunt/inject.sh", "recover_script": "./challenge-40-systematic-fault-hunt/recover.sh", "fault_phenomenon": "系统出现大面积异常：页面打开变慢、商品页和购物车间歇失败、结账和下单链路变慢、支付偶尔报错。多个业务功能同时出现问题，不是单一故障。请系统性排查三朵云环境中的所有异常配置和故障点，尽可能完整列出你发现的每个独立问题。", "public_case_info": "该题基于部署在三朵云（阿里云、腾讯云、AWS）上的电商微服务系统。可访问入口：阿里云前端 http://8.163.35.175:31366/、腾讯云网关 http://43.139.187.221:30076/、AWS 订单服务 http://3.26.241.84:30070/。可使用 MCP K8s 工具（只读）查看三云集群状态，并结合 RUM、APM 做系统性排查。本题重点考察多云、多服务、多资源类型故障的覆盖式发现能力。请输出你发现的每个独立故障点，包括云、资源类型、资源名、目标服务或配置、异常值/异常行为和判断依据。(这道题故障有点多,注入时间稍微长一丢丢)"}, "ideal_answer": {"fault_info": {"root_cause": "本题为多云系统性故障发现率测试，共注入 30 个独立故障点，无单一根因。故障分布在三朵云、多个服务、多种 Kubernetes/Chaos Mesh 资源类型中。", "faulty_clouds": ["unknown"], "affected_clouds": ["alicloud", "tencent", "aws"], "fault_location": {"module": "多云电商微服务系统", "file_path": null, "function_or_config": "30 个独立故障点：NetworkChaos×11 + StressChaos×3 + IOChaos×1 + TimeChaos×2（Chaos合计17）+ Deployment Env×6 + Deployment Resource×4 + ConfigMap/Volume×2 + ReadinessProbe×1", "description": "该题不要求寻找唯一根因，而是评估 agent 在多云、多资源类型环境中的系统性故障发现覆盖率。"}}, "reasoning_process": {"observed_symptoms": ["前端页面整体响应变慢，RUM 性能指标异常且采样量下降。", "商品页、推荐、购物车、结账和下单链路均出现间歇延迟或失败。", "支付链路出现时序相关异常。", "多个服务存在资源受限、探针异常、配置被覆盖或环境变量异常。"], "expected_investigation_strategy": ["先通过 RUM/APM 获取影响面：前端、商品、购物车、结账、下单、支付均异常，判断不是单一根因。", "对 alicloud、tencent、aws 三个 context 批量扫描 Chaos Mesh 对象：NetworkChaos、IOChaos、TimeChaos、StressChaos。", "对三云所有 deployment 批量扫描 env、resources、readinessProbe/livenessProbe、volumeMounts/volumes。", "检查关键 ConfigMap，尤其是 alicloud 的 rum-nginx-config 和 tencent 的 products-wrong-schema。", "按云、资源类型、资源名、目标服务、异常值去重汇总，避免把同一个故障重复计数。"], "fault_list": [{"id": 1, "cloud": "alicloud", "type": "NetworkChaos", "name": "c40-ali-checkout-latency", "target": "checkoutservice", "abnormal_value": "delay 500ms", "impact": "结账链路延迟升高"}, {"id": 2, "cloud": "alicloud", "type": "NetworkChaos", "name": "c40-ali-frontend-latency", "target": "frontend", "abnormal_value": "delay 450ms", "impact": "前端入口 TTFB/LCP 升高"}, {"id": 3, "cloud": "alicloud", "type": "NetworkChaos", "name": "c40-ali-recommend-loss", "target": "recommendationservice", "abnormal_value": "loss 8%", "impact": "推荐请求间歇失败"}, {"id": 4, "cloud": "alicloud", "type": "StressChaos", "name": "c40-ali-frontend-cpu-stress", "target": "frontend", "abnormal_value": "CPU load 70", "impact": "前端服务 CPU 压力"}, {"id": 5, "cloud": "alicloud", "type": "StressChaos", "name": "c40-ali-mysql-cpu-stress", "target": "mysql-orders", "abnormal_value": "CPU load 60", "impact": "订单存储链路抖动"}, {"id": 6, "cloud": "alicloud", "type": "Deployment Env", "name": "frontend", "target": "AD_SERVICE_ADDR", "abnormal_value": "127.0.0.1:1", "impact": "广告服务地址错误"}, {"id": 7, "cloud": "alicloud", "type": "Deployment Env", "name": "checkoutservice", "target": "ORDER_SERVICE_ADDR", "abnormal_value": "3.26.241.84:1", "impact": "结账调用订单服务端口错误"}, {"id": 8, "cloud": "alicloud", "type": "Deployment Resource", "name": "frontend", "target": "resources.limits.cpu", "abnormal_value": "20m", "impact": "前端 CPU 被限流"}, {"id": 9, "cloud": "alicloud", "type": "Deployment Resource", "name": "checkoutservice", "target": "resources.limits.cpu", "abnormal_value": "25m", "impact": "结账服务 CPU 被限流"}, {"id": 10, "cloud": "alicloud", "type": "ConfigMap", "name": "rum-nginx-config", "target": "RUM sample", "abnormal_value": "sample: 0.01", "impact": "RUM 上报采样骤降"}, {"id": 11, "cloud": "tencent", "type": "NetworkChaos", "name": "c40-tx-productcatalog-latency", "target": "productcatalogservice", "abnormal_value": "delay 450ms", "impact": "商品页加载延迟"}, {"id": 12, "cloud": "tencent", "type": "NetworkChaos", "name": "c40-tx-cart-latency", "target": "cartservice", "abnormal_value": "delay 500ms", "impact": "购物车操作超时"}, {"id": 13, "cloud": "tencent", "type": "NetworkChaos", "name": "c40-tx-currency-loss", "target": "currencyservice", "abnormal_value": "loss 6%", "impact": "货币换算间歇失败"}, {"id": 14, "cloud": "tencent", "type": "NetworkChaos", "name": "c40-tx-email-latency", "target": "emailservice", "abnormal_value": "delay 350ms", "impact": "通知链路延迟"}, {"id": 15, "cloud": "tencent", "type": "IOChaos", "name": "c40-tx-redis-io-latency", "target": "redis-cart", "abnormal_value": "/data latency 200ms, percent 80", "impact": "购物车 Redis IO 延迟"}, {"id": 16, "cloud": "tencent", "type": "Deployment Env", "name": "cartservice", "target": "CACHE_TTL", "abnormal_value": "0", "impact": "购物车缓存失效"}, {"id": 17, "cloud": "tencent", "type": "Deployment Env", "name": "cartservice", "target": "REDIS_CONNECT_TIMEOUT", "abnormal_value": "100ms", "impact": "Redis 连接超时预算过短"}, {"id": 18, "cloud": "tencent", "type": "Deployment Resource", "name": "productcatalogservice", "target": "resources.limits.cpu", "abnormal_value": "20m", "impact": "商品服务 CPU 被限流"}, {"id": 19, "cloud": "tencent", "type": "StressChaos", "name": "c40-tx-currency-cpu-stress", "target": "currencyservice", "abnormal_value": "CPU load 60", "impact": "currencyservice CPU 压力"}, {"id": 20, "cloud": "tencent", "type": "ConfigMap/Volume", "name": "products-wrong-schema", "target": "productcatalogservice /usr/src/app/products.json", "abnormal_value": "字段名 name/priceUsd/categories 被改为 productName/price/tags", "impact": "商品数据解析为空"}, {"id": 21, "cloud": "aws", "type": "NetworkChaos", "name": "c40-aws-orders-latency", "target": "new-orderservice", "abnormal_value": "delay 800ms", "impact": "下单链路延迟"}, {"id": 22, "cloud": "aws", "type": "NetworkChaos", "name": "c40-aws-inventory-latency", "target": "new-inventoryservice", "abnormal_value": "delay 450ms", "impact": "库存检查延迟"}, {"id": 23, "cloud": "aws", "type": "NetworkChaos", "name": "c40-aws-riskcontrol-loss", "target": "new-riskcontrolservice", "abnormal_value": "loss 7%", "impact": "风控链路间歇失败"}, {"id": 24, "cloud": "aws", "type": "NetworkChaos", "name": "c40-aws-notification-latency", "target": "new-notificationservice", "abnormal_value": "delay 350ms", "impact": "订单通知延迟"}, {"id": 25, "cloud": "aws", "type": "TimeChaos", "name": "c40-aws-payment-clock-skew", "target": "paymentservice", "abnormal_value": "+60s", "impact": "支付签名/时间窗口异常"}, {"id": 26, "cloud": "aws", "type": "TimeChaos", "name": "c40-aws-orders-clock-skew", "target": "new-orderservice", "abnormal_value": "-45s", "impact": "订单时间戳异常"}, {"id": 27, "cloud": "aws", "type": "Deployment Env", "name": "new-orderservice", "target": "MAX_RETRIES", "abnormal_value": "0", "impact": "下游调用无重试"}, {"id": 28, "cloud": "aws", "type": "Deployment Env", "name": "paymentservice", "target": "SIGNATURE_TTL", "abnormal_value": "1s", "impact": "支付签名 TTL 过短"}, {"id": 29, "cloud": "aws", "type": "Deployment Resource", "name": "new-inventoryservice", "target": "resources.limits.cpu", "abnormal_value": "30m", "impact": "库存服务 CPU 被限流"}, {"id": 30, "cloud": "aws", "type": "ReadinessProbe", "name": "new-orderservice", "target": "readinessProbe.httpGet.path", "abnormal_value": "/bad-health", "impact": "new-orderservice readiness 失败"}]}, "proposed_resolution": {"suggestion": "按故障清单逐项恢复：删除所有 c40-* Chaos 对象；恢复被篡改的 deployment env、resource、readinessProbe；恢复 rum-nginx-config 采样率；移除 products-wrong-schema ConfigMap 挂载并删除 ConfigMap。选手只需给出恢复建议，不应要求其实际修改集群。", "fix_steps_summary": ["删除三云所有 c40-* NetworkChaos/StressChaos/IOChaos/TimeChaos。", "恢复 alicloud frontend/checkoutservice 的 env 和 CPU limits。", "恢复 alicloud rum-nginx-config 中 sample: 1。", "恢复 tencent cartservice env，productcatalogservice CPU limit，并删除 products-wrong-schema 挂载和 ConfigMap。", "恢复 aws new-orderservice/paymentservice env，new-inventoryservice CPU limit，new-orderservice readinessProbe。"]}, "confidence": 1.0, "special_points": [{"point": "本题为发现率测试，不存在单一根因；发现一个真实独立故障点就应得一次分。", "score": 8}, {"point": "30 个故障覆盖三云和多资源类型，不能只查 Chaos 对象。", "score": 8}, {"point": "需要对 deployment env/resource/probe/volumeMount 与 ConfigMap 做批量扫描。", "score": 6}, {"point": "输出必须去重，不能把同一个故障重复计数。", "score": 5}], "tool_process": {"ideal_call_count": 16, "valuable_actions": ["RUM/APM 总览确认多链路异常，不按单根因收敛。", "三云批量查看 NetworkChaos/IOChaos/TimeChaos/StressChaos。", "三云批量查看 deployment env/resources/probes/volumes。", "查看 alicloud rum-nginx-config 和 tencent products-wrong-schema。", "按云、资源类型、目标服务、异常值生成去重故障清单。"]}}, "rubrics": {"positive_points_total": 96, "negative_points_total": -46, "ideal_call_count": 16, "rubrics": [{"criterion": "发现故障点 1：alicloud / NetworkChaos c40-ali-checkout-latency / checkoutservice 延迟 500ms", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:1"]}, {"criterion": "发现故障点 2：alicloud / NetworkChaos c40-ali-frontend-latency / frontend 延迟 450ms", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:2"]}, {"criterion": "发现故障点 3：alicloud / NetworkChaos c40-ali-recommend-loss / recommendationservice 丢包 8%", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:3"]}, {"criterion": "发现故障点 4：alicloud / StressChaos c40-ali-frontend-cpu-stress / frontend CPU load 70", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:4"]}, {"criterion": "发现故障点 5：alicloud / StressChaos c40-ali-mysql-cpu-stress / mysql-orders CPU load 60", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:5"]}, {"criterion": "发现故障点 6：alicloud / frontend deployment / AD_SERVICE_ADDR 被改为 127.0.0.1:1", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:6"]}, {"criterion": "发现故障点 7：alicloud / checkoutservice deployment / ORDER_SERVICE_ADDR 被改为 3.26.241.84:1", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:7"]}, {"criterion": "发现故障点 8：alicloud / frontend deployment / CPU limit 被改为 20m", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:8"]}, {"criterion": "发现故障点 9：alicloud / checkoutservice deployment / CPU limit 被改为 25m", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:9"]}, {"criterion": "发现故障点 10：alicloud / ConfigMap rum-nginx-config / RUM sample 被改为 0.01", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:10"]}, {"criterion": "发现故障点 11：tencent / NetworkChaos c40-tx-productcatalog-latency / productcatalogservice 延迟 450ms", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:11"]}, {"criterion": "发现故障点 12：tencent / NetworkChaos c40-tx-cart-latency / cartservice 延迟 500ms", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:12"]}, {"criterion": "发现故障点 13：tencent / NetworkChaos c40-tx-currency-loss / currencyservice 丢包 6%", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:13"]}, {"criterion": "发现故障点 14：tencent / NetworkChaos c40-tx-email-latency / emailservice 延迟 350ms", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:14"]}, {"criterion": "发现故障点 15：tencent / IOChaos c40-tx-redis-io-latency / redis-cart /data IO 延迟 200ms", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:15"]}, {"criterion": "发现故障点 16：tencent / cartservice deployment / CACHE_TTL=0", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:16"]}, {"criterion": "发现故障点 17：tencent / cartservice deployment / REDIS_CONNECT_TIMEOUT=100ms", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:17"]}, {"criterion": "发现故障点 18：tencent / productcatalogservice deployment / CPU limit 被改为 20m", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:18"]}, {"criterion": "发现故障点 19：tencent / StressChaos c40-tx-currency-cpu-stress / currencyservice CPU load 60", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:19"]}, {"criterion": "发现故障点 20：tencent / products-wrong-schema ConfigMap 挂载覆盖 productcatalogservice 的 products.json，字段名被篡改", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:20"]}, {"criterion": "发现故障点 21：aws / NetworkChaos c40-aws-orders-latency / new-orderservice 延迟 800ms", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:21"]}, {"criterion": "发现故障点 22：aws / NetworkChaos c40-aws-inventory-latency / new-inventoryservice 延迟 450ms", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:22"]}, {"criterion": "发现故障点 23：aws / NetworkChaos c40-aws-riskcontrol-loss / new-riskcontrolservice 丢包 7%", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:23"]}, {"criterion": "发现故障点 24：aws / NetworkChaos c40-aws-notification-latency / new-notificationservice 延迟 350ms", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:24"]}, {"criterion": "发现故障点 25：aws / TimeChaos c40-aws-payment-clock-skew / paymentservice 时钟偏移 +60s", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:25"]}, {"criterion": "发现故障点 26：aws / TimeChaos c40-aws-orders-clock-skew / new-orderservice 时钟偏移 -45s", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:26"]}, {"criterion": "发现故障点 27：aws / new-orderservice deployment / MAX_RETRIES=0", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:27"]}, {"criterion": "发现故障点 28：aws / paymentservice deployment / SIGNATURE_TTL=1s", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:28"]}, {"criterion": "发现故障点 29：aws / new-inventoryservice deployment / CPU limit 被改为 30m", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:29"]}, {"criterion": "发现故障点 30：aws / new-orderservice deployment / readinessProbe path 被改为 /bad-health", "points": 2, "tags": ["level:example", "axis:accuracy", "fault:30"]}, {"criterion": "排查覆盖三朵云，没有只查单云或两朵云就下结论", "points": 2, "tags": ["level:cluster", "cluster:systematic_investigation", "axis:completeness"]}, {"criterion": "排查覆盖多类资源：Chaos、Deployment env、Deployment resource、ConfigMap/volumeMount、readinessProbe", "points": 3, "tags": ["level:cluster", "cluster:systematic_investigation", "axis:completeness"]}, {"criterion": "输出结构清晰，每个故障点包含云、资源类型、资源名或 deployment、目标服务/配置、异常值和依据", "points": 3, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "明确这是多故障发现率测试，不把 30 个独立故障强行合并为一个单一根因", "points": 2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "响应按清晰结构组织：每个故障点包含云、资源类型、资源名、目标服务、异常值和依据", "points": 5, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "面对三云多服务多资源类型的故障分布，能系统性覆盖而非随机抽查", "points": 5, "tags": ["level:cluster", "cluster:systematic_investigation", "axis:completeness"]}, {"criterion": "排查覆盖多类资源类型（Chaos、env、resource、configmap、probe），不局限于单一类型", "points": 5, "tags": ["level:cluster", "cluster:systematic_investigation", "axis:completeness"]}, {"criterion": "基于 MCP/RUM/APM 可观测证据逐一确认故障点，而非凭经验猜测", "points": 5, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1（≤ 16 次）", "points": 4, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 ≤ ideal_call_count × 1.25（≤ 20 次）", "points": 2, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "排查效率：总工具调用次数 > ideal_call_count × 1.5（> 24 次）", "points": -3, "tags": ["level:example", "axis:efficiency"]}, {"criterion": "编造不存在的故障、正常配置或正常服务异常：每项扣 2 分", "points": -2, "tags": ["level:example", "axis:normality"]}, {"criterion": "同一个故障重复计数：每项扣 2 分", "points": -2, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "统计故障总数错误：每多报或少报 1 个扣 1 分", "points": -1, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只查 Chaos 对象，未检查 env/resource/configmap/probe，导致非 Chaos 故障大量遗漏", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只查 deployment 配置，未检查 Chaos 对象，导致注入故障大量遗漏", "points": -8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "只查一朵云或两朵云，未覆盖全部三云", "points": -6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "答案要求或暗示选手实际执行恢复、修改集群、SSH 登录等超出只读能力的操作", "points": -4, "tags": ["level:example", "axis:normality"]}, {"criterion": "大量无依据猜测，未基于 MCP/RUM/APM 可观测证据，或把概率性现象说成绝对结论", "points": -5, "tags": ["level:example", "axis:normality"]}, {"criterion": "输出格式严重混乱，无法稳定区分真实故障点、推测项和恢复建议", "points": -3, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "排查效率异常低：大量重复查看同一资源或反复调用无关 RUM/APM 查询但没有产生新证据", "points": -4, "tags": ["level:example", "axis:efficiency", "axis:normality"]}]}, "prompt": "系统出现大面积异常：页面打开变慢、商品页和购物车间歇失败、结账和下单链路变慢、支付偶尔报错。多个业务功能同时出现问题，不是单一故障。请系统性排查三朵云环境中的所有异常配置和故障点，尽可能完整列出你发现的每个独立问题。"}
{"canary": "mc_origin_conntrack_cdn_smoke", "tags": ["定位", "网络与协议", "conntrack 耗尽", "CDN 5xx 风暴", "hard", "multi-cloud"], "case": {"id": "mc_origin_conntrack_cdn_smoke", "title": "多云入口间歇性 5xx", "inject_script": "./faults/mc_origin_conntrack_cdn_smoke/inject.sh", "recover_script": "./faults/mc_origin_conntrack_cdn_smoke/recover.sh", "fault_phenomenon": "用户访问商城时，页面或接口偶发 5xx，商品列表、评分等内容有时加载失败；刷新后可能恢复，故障不是每次都复现。", "public_case_info": "该题基于部署在阿里云与腾讯云上的 Stan's Robot Shop 微服务系统。可观察入口：多云全局入口 http://8.163.90.83:18080/，阿里云直连入口 http://8.163.90.83:18081/，腾讯云直连入口 http://43.139.215.81:18081/。平台提供只读 MCP 诊断能力；请根据题面现象和 MCP 证据定位故障根因、影响范围、证据链和恢复建议。", "order_id": 31, "submission_enabled": true, "ai_analysis_visible": true}, "ideal_answer": {"fault_info": {"root_cause": "阿里云 origin CVM 到本地 Robot Shop edge 的 18081 新连接被连接跟踪/短连接压力保护规则随机拒绝，导致全局 gateway 命中阿里云 origin 时出现 502；网关和后端容器本身不是根因。", "faulty_clouds": ["aliyun"], "affected_clouds": ["aliyun"], "fault_location": {"module": "origin-host-network", "file_path": "/opt/mc-robot-shop/state/conntrack-saturation.json", "function_or_config": "iptables OUTPUT rule with mc_origin_conntrack_cdn_smoke marker", "description": "故障在阿里云 origin 主机网络层，表现为全局 gateway 到本机 edge:18081 的新连接被随机拒绝。"}}, "reasoning_process": {"observed_symptoms": ["全局入口间歇 502，成功和失败交替。", "容器状态大多正常。", "腾讯云直连入口相对正常。"], "key_evidence": [{"source": "multi-cloud-ssh.conntrack(aliyun)", "content": "可见 conntrack 计数、18081 socket 状态和 mc_origin_conntrack_cdn_smoke 连接拒绝规则。", "conclusion": "故障发生在阿里云 origin 主机连接路径。"}, {"source": "multi-cloud-ssh.curl_loop(aliyun, global, /)", "content": "全局入口重复请求出现 200/502 交替。", "conclusion": "符合 origin 新连接被随机拒绝后的上层 5xx。"}, {"source": "multi-cloud-ssh.probe(tencent)", "content": "腾讯云直连入口和 API 探测正常。", "conclusion": "排除腾讯云整体应用故障。"}], "causal_chain": ["阿里云 origin 主机 18081 新连接被随机拒绝。", "全局 gateway 命中阿里云 origin 时建连失败。", "Nginx 返回 502。", "腾讯云路径仍可用形成 CDN/gateway 烟雾弹。"]}, "verification_method": {"verification_commands": [{"cmd": "MCP: conntrack(aliyun)", "purpose": "检查 origin 主机连接跟踪和拒绝规则。", "expected_result": "恢复后无 mc_origin_conntrack_cdn_smoke 规则。"}, {"cmd": "MCP: curl_loop(aliyun, global, /, 20)", "purpose": "验证全局入口是否随机 502。", "expected_result": "恢复后连续成功。"}, {"cmd": "MCP: probe(aliyun)+probe(tencent)", "purpose": "验证两云入口。", "expected_result": "关键 API 均正常。"}], "success_criteria": ["连接拒绝规则消失。", "全局入口不再间歇 502。", "两云直连入口保持正常。"]}, "proposed_resolution": {"suggestion": "删除阿里云 origin 主机上的连接拒绝/conntrack 压力保护规则，恢复正常建连。", "fix_steps": ["删除带 mc_origin_conntrack_cdn_smoke 标记的 iptables 规则。", "清理状态摘要。", "重复探测全局入口和两云直连。", "增加 origin conntrack/短连接风暴告警。"]}, "confidence": 0.91}, "rubrics": {"positive_points_total": 111, "negative_points_total": -60, "rubrics": [{"criterion": "准确指出根因是阿里云 origin CVM 到本地 edge:18081 的新连接被 conntrack/短连接压力相关规则随机拒绝；只写 CDN、Nginx、gateway 抖动或后端服务异常不得命中本项。", "points": 22, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确写出 faulty_clouds 为阿里云、affected_clouds 为阿里云，并说明腾讯云直连正常，异常集中在阿里云承载的全局入口/origin 路径。", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "完整描述入口对比结果：多云全局入口间歇 502，阿里云直连和腾讯云直连基本 200；缺少直连对比不得命中本项。", "points": 10, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 MCP conntrack(aliyun) 拿到关键证据，并至少引用 nf_conntrack_count/nf_conntrack_max、edge:18081 socket 样本、带 mc_origin_conntrack_cdn_smoke 标记的 iptables 规则三者之一。", "points": 16, "tags": ["level:example", "axis:evidence"]}, {"criterion": "能解释为什么容器 Running、后端 API 偶尔 200、腾讯云直连正常是烟雾弹，不能据此排除阿里云 origin 主机网络层。", "points": 8, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "给出完整因果链：阿里云 origin 主机连接拒绝 -> 全局 gateway 命中该 upstream 时 502 -> 用户在全局入口看到间歇性 5xx。", "points": 12, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "明确排除近似误判：CDN 缓存/配置错误、Nginx upstream 漂移、数据库故障、腾讯云服务故障，并给出排除依据。", "points": 8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "恢复方案必须指向清理阿里云 origin 主机上的连接拒绝/iptables/conntrack 压力注入状态，而不是重启业务容器或改数据库。", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "恢复验证必须包含全局入口连续请求稳定、两云直连仍正常、conntrack/iptables 故障标记消失。", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "Agent 过程先通过平台提供的 MCP 工具 probe 或 curl_loop 做三入口对比，再进入 conntrack 证据验证；如果只调用 status/logs 不得命中本项。", "points": 7, "tags": ["level:example", "axis:process_quality"]}, {"criterion": "Agent 调用 conntrack(aliyun) 前无关 logs 调用不超过 3 次，且没有无目标地扫 Redis、MQ、数据库日志。", "points": 5, "tags": ["level:example", "axis:tool_efficiency"]}, {"criterion": "回答结构清晰，最终结论、证据、因果链、恢复和验证彼此一致。", "points": 3, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "提出针对 origin 主机连接压力的预防措施，例如连接数/conntrack 水位告警、全局入口按 upstream 分层探测、变更前网络规则审计。", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "错误地把根因判断为 CDN/gateway/Nginx upstream 配置、数据库或腾讯云应用服务故障，且没有指出阿里云 origin 主机 18081 连接拒绝。", "points": -14, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只描述服务重启、资源不足、网络波动或浏览器问题，没有任何 MCP conntrack/iptables/socket 证据链。", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "忽略多云全局入口与两云直连入口的关键差异，把单入口现象直接泛化成全局业务全挂。", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "建议重装系统、删除数据库、清空 Redis/RabbitMQ、关闭安全组等不必要破坏性操作。", "points": -8, "tags": ["level:example", "axis:safety"]}, {"criterion": "编造不存在的云厂商告警、日志、监控指标或命令输出作为关键证据。", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "Agent 总工具调用次数 >=20，或未通过 MCP 调用 conntrack(aliyun) 却反复调用 status/probe/logs 合计 >=12 次，并停留在表象方向。", "points": -10, "tags": ["level:example", "axis:tool_efficiency"]}]}, "prompt": "用户访问商城时，页面或接口偶发 5xx，商品列表、评分等内容有时加载失败；刷新后可能恢复，故障不是每次都复现。"}
{"canary": "mc_tencent_egress_degrade", "tags": ["定位", "网络与协议", "跨云网络延迟", "偶发超时", "hard", "multi-cloud"], "case": {"id": "mc_tencent_egress_degrade", "title": "跨云访问偶发超时", "inject_script": "./faults/mc_tencent_egress_degrade/inject.sh", "recover_script": "./faults/mc_tencent_egress_degrade/recover.sh", "fault_phenomenon": "用户访问商城时，页面和接口响应时快时慢，偶尔长时间等待或超时；同一操作重复尝试时可能有时成功、有时失败。", "public_case_info": "该题基于部署在阿里云与腾讯云上的 Stan's Robot Shop 微服务系统。可观察入口：多云全局入口 http://8.163.90.83:18080/，阿里云直连入口 http://8.163.90.83:18081/，腾讯云直连入口 http://43.139.215.81:18081/。平台提供只读 MCP 诊断能力；请根据题面现象和 MCP 证据定位故障根因、影响范围、证据链和恢复建议。", "order_id": 32, "submission_enabled": true, "ai_analysis_visible": true}, "ideal_answer": {"fault_info": {"root_cause": "腾讯云主机到阿里云 gateway 的回包方向被 tc netem 注入高延迟、抖动和丢包，导致阿里云全局入口访问腾讯云 upstream 时慢或超时；业务容器和 Nginx 配置本身正常。", "faulty_clouds": ["tencent"], "affected_clouds": ["aliyun"], "fault_location": {"module": "host-network", "file_path": "/opt/mc-robot-shop/state/netem-summary.json", "function_or_config": "tc qdisc/filter on Tencent default interface", "description": "故障作用在腾讯云默认网卡发往阿里云 gateway IP 的回包路径。"}}, "reasoning_process": {"observed_symptoms": ["全局入口一半请求慢或超时。", "腾讯云本机 probe 和直连入口正常。", "容器 Running，应用日志无稳定单服务报错。"], "key_evidence": [{"source": "multi-cloud-ssh.netem(tencent)", "content": "默认网卡存在匹配阿里云 IP 的 netem delay/loss 规则。", "conclusion": "跨云回包路径被退化。"}, {"source": "multi-cloud-ssh.curl_loop(aliyun, global, /api/catalogue/products)", "content": "全局入口重复请求耗时抖动或超时。", "conclusion": "故障在 gateway 访问远端 upstream 路径。"}, {"source": "multi-cloud-ssh.probe(tencent)", "content": "腾讯云本地直连 API 正常。", "conclusion": "排除腾讯云业务容器整体故障。"}], "causal_chain": ["tc filter 匹配阿里云 gateway IP。", "回包被施加延迟、抖动和丢包。", "全局 gateway 命中腾讯云 upstream 时慢或超时。", "腾讯云本地自检正常形成烟雾弹。"]}, "verification_method": {"verification_commands": [{"cmd": "MCP: netem(tencent)", "purpose": "查看 tc qdisc/filter。", "expected_result": "恢复后无 delay/loss 规则。"}, {"cmd": "MCP: curl_loop(aliyun, global, /api/catalogue/products, 20)", "purpose": "验证全局入口耗时。", "expected_result": "恢复后耗时稳定。"}, {"cmd": "MCP: probe(tencent)", "purpose": "确认腾讯云直连正常。", "expected_result": "本地服务正常。"}], "success_criteria": ["tc netem 规则被删除。", "全局入口访问腾讯云 upstream 不再慢或超时。", "两云直连入口正常。"]}, "proposed_resolution": {"suggestion": "删除腾讯云默认网卡上匹配阿里云 IP 的 tc netem qdisc/filter，并验证跨云路径恢复。", "fix_steps": ["定位腾讯云默认网卡。", "删除异常 tc qdisc/filter。", "从阿里云全局入口重复探测。", "增加分 upstream 延迟和单向路径探测。"]}, "confidence": 0.91}, "rubrics": {"positive_points_total": 115, "negative_points_total": -60, "rubrics": [{"criterion": "准确指出根因是腾讯云发往阿里云 gateway 的单向回包路径被 tc netem 注入延迟、抖动或丢包；只写普通跨云网络慢不得命中本项。", "points": 22, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确写出 faulty_clouds 为腾讯云、affected_clouds 为阿里云，并说明症状出现在阿里云全局入口访问腾讯云 upstream 的跨云路径，腾讯云直连基本正常。", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "用全局入口与腾讯云直连入口的对比说明：全局入口慢/超时，而腾讯云直连基本正常。", "points": 10, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 MCP netem(tencent) 拿到 qdisc/filter 或 netem-summary 证据，并能指出匹配目标是阿里云 gateway IP 或该特定跨云路径。", "points": 18, "tags": ["level:example", "axis:evidence"]}, {"criterion": "解释为什么腾讯云本机 curl 正常、容器 Running、应用日志少报错不能排除单向回包路径退化。", "points": 8, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "给出完整因果链：腾讯云回包方向退化 -> 阿里云 gateway 等待 upstream 响应变慢/超时 -> 全局入口间歇慢请求。", "points": 12, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "明确排除近似误判：业务容器 Down、DNS、阿里云 gateway 配置、普通随机网络波动，并给出排除依据。", "points": 8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "恢复方案必须指向清理腾讯云网卡上的 tc netem qdisc/filter，而不是只调大 Nginx 超时或重启服务。", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "恢复验证必须包含全局入口连续请求耗时恢复、腾讯云直连仍正常、netem(tencent) 无故障规则。", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "Agent 通过 MCP curl_loop/probe 做全局入口与腾讯直连对比后，再调用 netem(tencent) 收敛网络路径证据。", "points": 8, "tags": ["level:example", "axis:process_quality"]}, {"criterion": "Agent 总工具调用次数 <=15，且没有在业务日志、Redis、MQ 上做大量无关探索。", "points": 6, "tags": ["level:example", "axis:tool_efficiency"]}, {"criterion": "回答结构清晰，最终结论、证据、因果链、恢复和验证彼此一致。", "points": 3, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "提出针对单向跨云路径的预防措施，例如按方向探测、跨云链路 SLO、tc/路由规则审计。", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "错误地把根因判断为业务容器、阿里云 gateway 配置、DNS、负载均衡随机性或普通公网抖动，未指出腾讯云定向回包 netem。", "points": -14, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只描述网络慢或重试超时，没有任何 MCP netem/qdisc/filter 证据。", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "忽略腾讯云直连基本正常这一关键差异，把问题写成腾讯云整体不可用。", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "建议重装系统、关闭安全组、删除容器或清空业务数据等不必要破坏性操作。", "points": -8, "tags": ["level:example", "axis:safety"]}, {"criterion": "编造不存在的应用错误日志、云监控告警、链路追踪或命令输出作为关键证据。", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "Agent 总工具调用次数 >=20，或连续调用 curl_loop/probe/status 合计 >=12 次仍未通过 MCP 调用 netem(tencent)。", "points": -10, "tags": ["level:example", "axis:tool_efficiency"]}]}, "prompt": "用户访问商城时，页面和接口响应时快时慢，偶尔长时间等待或超时；同一操作重复尝试时可能有时成功、有时失败。"}
{"canary": "mc_api_schema_rum_js_error", "tags": ["定位", "配置与密钥", "RUM JS Error", "API Schema 不兼容", "hard", "multi-cloud"], "case": {"id": "mc_api_schema_rum_js_error", "title": "前端页面局部渲染异常", "inject_script": "./faults/mc_api_schema_rum_js_error/inject.sh", "recover_script": "./faults/mc_api_schema_rum_js_error/recover.sh", "fault_phenomenon": "用户打开商品相关页面时，页面局部区域可能空白、商品信息渲染异常或前端报错；页面其他区域可能仍能正常显示。", "public_case_info": "该题基于部署在阿里云与腾讯云上的 Stan's Robot Shop 微服务系统。可观察入口：多云全局入口 http://8.163.90.83:18080/，阿里云直连入口 http://8.163.90.83:18081/，腾讯云直连入口 http://43.139.215.81:18081/。平台提供只读 MCP 诊断能力；请根据题面现象和 MCP 证据定位故障根因、影响范围、证据链和恢复建议。", "order_id": 33, "submission_enabled": true, "ai_analysis_visible": true}, "ideal_answer": {"fault_info": {"root_cause": "腾讯云 edge 对 catalogue products API 暴露了不兼容 schema，把原本前端期望的商品数组改成 items/schemaVersion 包装对象，导致前端按旧契约解析失败并表现为 JS error。", "faulty_clouds": ["tencent"], "affected_clouds": ["aliyun", "tencent"], "fault_location": {"module": "api-contract", "file_path": "/opt/mc-robot-shop/edge/default.conf", "function_or_config": "default.schema-bad.conf /api/catalogue/products", "description": "故障在腾讯云入口暴露的 catalogue API 契约，不是浏览器代码或 catalogue 容器崩溃。"}}, "reasoning_process": {"observed_symptoms": ["商品页局部渲染失败，页面可见前端 JS error 或同类用户侧异常。", "catalogue 容器 Running，接口仍返回 200。", "响应体 schema 与前端预期不一致。"], "key_evidence": [{"source": "页面现象 / 多云入口访问结果", "content": "商品区域局部渲染失败，用户侧可见前端报错或渲染异常，但 catalogue 服务本身不一定 Down。", "conclusion": "用户侧异常更像是 API 契约变化导致的前端解析失败，而不是 catalogue 容器直接崩溃。"}, {"source": "multi-cloud-ssh.curl_loop(tencent, direct, /api/catalogue/products)", "content": "接口返回 200 但响应结构变成 items/schemaVersion 包装对象。", "conclusion": "后端返回成功不代表契约兼容。"}, {"source": "multi-cloud-ssh.edge(tencent) / diff_config(tencent)", "content": "edge 配置命中 /api/catalogue/products 的异常响应配置，直接改写了 catalogue products API 的返回结构。", "conclusion": "定位到腾讯云入口层的 API schema 变更。"}], "causal_chain": ["腾讯云 catalogue API 响应结构被变更。", "前端仍按旧 schema 读取商品字段。", "解析异常表现为 JS error。", "服务 200/Running 造成烟雾弹。"]}, "verification_method": {"verification_commands": [{"cmd": "MCP: curl_loop(tencent, direct, /api/catalogue/products, 3)", "purpose": "检查响应 schema。", "expected_result": "恢复后返回兼容结构。"}, {"cmd": "MCP: diff_config(tencent)", "purpose": "检查 edge 配置差异。", "expected_result": "恢复后 edge 与 baseline 无差异。"}], "success_criteria": ["edge 恢复 baseline。", "catalogue products API schema 恢复兼容。", "页面商品区域恢复正常渲染。"]}, "proposed_resolution": {"suggestion": "回滚腾讯云 edge 的 schema canary，恢复 catalogue API 与前端约定的响应结构。", "fix_steps": ["恢复 edge/default.conf.baseline。", "重启腾讯云 edge。", "验证 catalogue API schema 和页面渲染状态。", "增加 API contract test。"]}, "confidence": 0.91}, "rubrics": {"positive_points_total": 114, "negative_points_total": -60, "rubrics": [{"criterion": "准确指出根因是腾讯云 catalogue products API schema 与前端旧契约不兼容；只写前端 JS bug、RUM 告警或接口 500 不得命中本项。", "points": 22, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确写出 faulty_clouds 为腾讯云；affected_clouds 至少包含腾讯云。如果说明阿里云全局入口可能因负载均衡间接受影响，写成 [\"aliyun\", \"tencent\"] 也可接受，但必须指出直接异常实例在腾讯云。", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "完整描述现象：页面商品区域局部渲染异常，并伴随前端 JS error 或同类用户侧异常线索；同时说明 catalogue 服务不一定 Down。", "points": 8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "优先通过 MCP rum() 拿到 JS error/schema 线索；如果 RUM 暂时无数据，也可以通过页面现象或接口响应先建立前端异常与 /api/catalogue/products schema 异常的关联。", "points": 10, "tags": ["level:example", "axis:evidence"]}, {"criterion": "探测 /api/catalogue/products 并指出 items、schemaVersion、displayName、unitPrice 等字段形态与旧前端契约不兼容。", "points": 14, "tags": ["level:example", "axis:evidence"]}, {"criterion": "通过 MCP edge(tencent) 或 diff_config(tencent) 定位到 /api/catalogue/products 的异常 schema 配置。", "points": 12, "tags": ["level:example", "axis:evidence"]}, {"criterion": "给出完整因果链：API 契约变化 -> 前端按旧字段解析失败 -> 局部渲染异常和 JS error。", "points": 10, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "明确排除近似误判：前端代码发布、浏览器缓存、静态资源 403、catalogue 容器 Down、数据库故障。", "points": 8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "恢复方案必须指向还原腾讯云 edge/API schema 或兼容前端契约，而不是只清缓存或重启前端。", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "恢复验证必须包含 API 响应结构恢复、页面商品区域渲染恢复；如果环境中已有 RUM 数据，再补充验证 RUM 中的 JS/schema 错误摘要回落。", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "Agent 过程体现优先尝试 rum()，或直接查 API 响应结构，再用 edge/diff_config 验证配置来源。", "points": 7, "tags": ["level:example", "axis:process_quality"]}, {"criterion": "Agent 总工具调用次数 <=15，且已获得 schema 线索后没有继续大量探索 Redis、MQ、netem。", "points": 4, "tags": ["level:example", "axis:tool_efficiency"]}, {"criterion": "回答结构清晰，最终结论、证据、因果链、恢复和验证彼此一致。", "points": 3, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "错误地把根因判断为前端代码 bug、浏览器缓存、静态资源/CDN、catalogue 容器 Down 或数据库故障。", "points": -14, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只引用 RUM JS error 或页面异常，没有 API schema 响应字段证据。", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "忽略接口返回 200 但契约不兼容这一关键点，把 HTTP 状态码正常误判为后端正常。", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "建议删除数据库、清空缓存作为主要恢复手段，可能破坏业务数据或掩盖根因。", "points": -8, "tags": ["level:example", "axis:safety"]}, {"criterion": "编造不存在的前端发布记录、浏览器报错堆栈、云监控告警或接口返回内容作为证据。", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "Agent 总工具调用次数 >=20，或已获得 RUM/schema 线索后仍调用 redis/mq/netem/logs 等无关 MCP 工具合计 >=8 次。", "points": -10, "tags": ["level:example", "axis:tool_efficiency"]}]}, "prompt": "用户打开商品相关页面时，页面局部区域可能空白、商品信息渲染异常或前端报错；页面其他区域可能仍能正常显示。"}
{"canary": "mc_redis_hotkey_eviction", "tags": ["定位", "中间件与依赖", "Redis 热点 key", "缓存驱逐", "hard", "multi-cloud"], "case": {"id": "mc_redis_hotkey_eviction", "title": "购物车数据偶发丢失", "inject_script": "./faults/mc_redis_hotkey_eviction/inject.sh", "recover_script": "./faults/mc_redis_hotkey_eviction/recover.sh", "fault_phenomenon": "用户使用购物车时，已加入的商品偶发丢失、数量回退或刷新后状态不一致；加购操作不一定每次失败。", "public_case_info": "该题基于部署在阿里云与腾讯云上的 Stan's Robot Shop 微服务系统。可观察入口：多云全局入口 http://8.163.90.83:18080/，阿里云直连入口 http://8.163.90.83:18081/，腾讯云直连入口 http://43.139.215.81:18081/。平台提供只读 MCP 诊断能力；请根据题面现象和 MCP 证据定位故障根因、影响范围、证据链和恢复建议。", "order_id": 34, "submission_enabled": true, "ai_analysis_visible": true}, "ideal_answer": {"fault_info": {"root_cause": "腾讯云 Redis 被设置了过小 maxmemory 和 allkeys-lru 淘汰策略，同时注入大量热 key，触发 evicted_keys 增长，购物车/会话相关键被淘汰。", "faulty_clouds": ["tencent"], "affected_clouds": ["tencent"], "fault_location": {"module": "redis", "file_path": "/opt/mc-robot-shop/state/redis-summary.json", "function_or_config": "redis maxmemory-policy allkeys-lru", "description": "故障在腾讯云 Redis 内存与淘汰策略，不是 cart 服务崩溃。"}}, "reasoning_process": {"observed_symptoms": ["购物车状态偶发丢失或抖动。", "Redis 容器 Running，连接可用。", "evicted_keys 和热 key 写入信号异常。"], "key_evidence": [{"source": "multi-cloud-ssh.redis(tencent)", "content": "maxmemory 较小、policy=allkeys-lru、evicted_keys 增长。", "conclusion": "Redis 正在淘汰业务键。"}, {"source": "multi-cloud-ssh.probe(tencent)", "content": "cart metrics 和 Redis 容器仍可访问。", "conclusion": "排除 Redis Down。"}, {"source": "multi-cloud-ssh.diff_config(tencent)", "content": "state/redis-summary.json 记录 hot key/eviction fault active。", "conclusion": "定位到热 key 与淘汰策略。"}], "causal_chain": ["Redis maxmemory 被收紧且策略改成 allkeys-lru。", "热 key 写入制造内存压力。", "Redis 淘汰购物车/会话键。", "用户看到购物车丢失但容器健康。"]}, "verification_method": {"verification_commands": [{"cmd": "MCP: redis(tencent)", "purpose": "查看 Redis 策略和 evicted_keys。", "expected_result": "恢复后策略回 baseline。"}, {"cmd": "MCP: probe(tencent)", "purpose": "验证 cart/Redis 基础可用。", "expected_result": "cart metrics 正常。"}, {"cmd": "MCP: curl_loop(tencent, direct, /api/cart/metrics, 5)", "purpose": "检查 cart 路径稳定性。", "expected_result": "恢复后稳定成功。"}], "success_criteria": ["Redis policy 恢复 baseline。", "mc_hotkey 测试键清理。", "购物车相关接口稳定。"]}, "proposed_resolution": {"suggestion": "恢复腾讯云 Redis 内存策略，清理注入热 key，并验证 evicted_keys 不再增长。", "fix_steps": ["恢复 Redis maxmemory 和 maxmemory-policy。", "清理 mc_hotkey 测试键。", "验证 Redis INFO 和 cart metrics。", "增加 hot key、eviction、latency 监控。"]}, "confidence": 0.91}, "rubrics": {"positive_points_total": 106, "negative_points_total": -60, "rubrics": [{"criterion": "准确指出根因是腾讯云 Redis maxmemory/allkeys-lru 误配叠加热 key，导致购物车或会话相关键被 evicted；只写 Redis 慢或 cart 异常不得命中本项。", "points": 22, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确写出 faulty_clouds 为腾讯云、affected_clouds 为腾讯云，并说明全局入口只是可能命中腾讯云状态存储问题，不把阿里云写成根因或直接异常云。", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "完整描述现象：购物车数据偶发丢失/抖动，但 cart/API 未整体 Down。", "points": 8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 MCP redis(tencent) 引用 maxmemory、maxmemory-policy=allkeys-lru、evicted_keys、hot key 摘要中的至少两类证据。", "points": 18, "tags": ["level:example", "axis:evidence"]}, {"criterion": "能解释热 key 与 allkeys-lru/容量限制叠加，为什么会造成已有购物车数据被淘汰而不是请求全部失败。", "points": 12, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "能说明 cart 容器 Running、API 200、Redis 未 Down 是烟雾弹，不能排除 Redis 数据淘汰问题。", "points": 8, "tags": ["level:cluster", "cluster:misleading_signal_awareness", "axis:context_awareness"]}, {"criterion": "明确排除近似误判：cart 服务崩溃、Redis 容器 Down、浏览器缓存、用户会话前端问题。", "points": 8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "恢复方案必须指向还原 Redis maxmemory/maxmemory-policy、清理注入热 key 或扩大容量保护，而不是只重启 cart。", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "恢复验证必须包含 evicted_keys 不再快速增长、购物车操作稳定、cart/API 状态正常。", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "Agent 过程先通过 MCP probe/curl_loop 排除 cart/API 整体 Down，再调用 redis(tencent) 收集配置和统计证据。", "points": 6, "tags": ["level:example", "axis:process_quality"]}, {"criterion": "Agent 总工具调用次数 <=16，且没有反复查看 cart 日志替代 Redis 指标。", "points": 4, "tags": ["level:example", "axis:tool_efficiency"]}, {"criterion": "回答结构清晰，最终结论、证据、因果链、恢复和验证彼此一致。", "points": 3, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "错误地把根因判断为 cart 服务崩溃、用户 session、浏览器缓存、Redis Down 或普通缓存击穿，未指出 eviction 机制。", "points": -14, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "没有 MCP redis(tencent) 指标证据，只根据购物车丢失现象猜测 Redis 或缓存问题。", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "忽略 evicted_keys/maxmemory-policy，把 Redis 容器 Up 误当成 Redis 状态完全正常。", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "建议清空 Redis 全库、删除数据库或重建服务作为主要恢复手段，未评估业务数据影响。", "points": -8, "tags": ["level:example", "axis:safety"]}, {"criterion": "编造不存在的 Redis INFO、日志、热点 key 或云监控指标作为证据。", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "Agent 总工具调用次数 >=24，或未通过 MCP 调用 redis(tencent) 却重复调用 logs(cart)/probe/status 合计 >=12 次。", "points": -10, "tags": ["level:example", "axis:tool_efficiency"]}]}, "prompt": "用户使用购物车时，已加入的商品偶发丢失、数量回退或刷新后状态不一致；加购操作不一定每次失败。"}
{"canary": "mc_rabbitmq_backlog_order_lag", "tags": ["定位", "中间件与依赖", "消息队列积压", "订单 lag", "hard", "multi-cloud"], "case": {"id": "mc_rabbitmq_backlog_order_lag", "title": "订单状态长时间未完成", "inject_script": "./faults/mc_rabbitmq_backlog_order_lag/inject.sh", "recover_script": "./faults/mc_rabbitmq_backlog_order_lag/recover.sh", "fault_phenomenon": "用户提交下单或支付操作后，页面可能显示请求已成功，但订单发货、配送或状态流转长时间没有更新。", "public_case_info": "该题基于部署在阿里云与腾讯云上的 Stan's Robot Shop 微服务系统。可观察入口：多云全局入口 http://8.163.90.83:18080/，阿里云直连入口 http://8.163.90.83:18081/，腾讯云直连入口 http://43.139.215.81:18081/。平台提供只读 MCP 诊断能力；请根据题面现象和 MCP 证据定位故障根因、影响范围、证据链和恢复建议。", "order_id": 35, "submission_enabled": true, "ai_analysis_visible": true}, "ideal_answer": {"fault_info": {"root_cause": "腾讯云 dispatch 消费链路被施加了严格的 CPU 配额限制。dispatch 消费者仍然存活并持续处理订单消息，但处理能力被人为压低，导致同步请求可以成功返回，而异步业务完成明显变慢。", "faulty_clouds": ["tencent"], "affected_clouds": ["tencent"], "fault_location": {"module": "rabbitmq-dispatch", "file_path": "/opt/mc-robot-shop/state/mq-summary.json", "function_or_config": "dispatch cpu quota / low-throughput consumer state", "description": "故障不在同步 API，而在腾讯云异步消费链路。dispatch 容器保持存活，orders 队列消费者也保持 active，但 dispatch 被限速后处理吞吐明显下降，形成稳定的低速消费状态。"}}, "reasoning_process": {"observed_symptoms": ["下单或支付请求可以成功返回，但异步业务完成明显变慢。", "RabbitMQ 与 dispatch 容器保持 Up，dispatch 不是直接 Down。", "mq 观测结果显示 orders 队列仍有消费者，dispatch_active_consumers 仍大于 0，dispatch CPU quota 被压到 5000/100000。", "dispatch 日志持续出现 oj-lag-* 订单消息，说明消费者仍在处理合法消息而不是已经退出。"], "key_evidence": [{"source": "multi-cloud-ssh.mq(tencent)", "content": "mq summary 显示 dispatch_cpu_quota=5000/100000、dispatch_active_consumers 仍大于 0、sync_api_vs_async_state=api_success_business_pending，orders 队列消费者仍存在。", "conclusion": "根因位于腾讯云异步消费链路的低 CPU 限速，而不是同步 API 或 RabbitMQ 整体不可用。"}, {"source": "multi-cloud-ssh.status(tencent)", "content": "dispatch 与 rabbitmq 容器仍处于 Up 状态。", "conclusion": "可以排除 dispatch 容器直接宕掉的误判。"}, {"source": "multi-cloud-ssh.logs(tencent, dispatch)", "content": "dispatch 日志持续打印 Order {\"orderid\":\"oj-lag-*\"}，说明消费者在限速下仍持续消费有效消息。", "conclusion": "这是活消费者低速处理，而不是消费者已经消失。"}], "causal_chain": ["腾讯云 dispatch 被限制到很低的 CPU 配额。", "在持续订单流量下，dispatch 消费者仍然 active，但处理能力明显下降。", "同步 API 仍可成功返回，而下游异步业务完成变慢。"]}, "verification_method": {"verification_commands": [{"cmd": "MCP: mq(tencent)", "purpose": "确认 dispatch CPU quota 已被限制、消费者仍存在，且同步 API 成功但异步业务仍处于 pending。", "expected_result": "故障状态下能看到 dispatch_cpu_quota=5000/100000、dispatch_active_consumers > 0、sync_api_vs_async_state=api_success_business_pending。"}, {"cmd": "MCP: status(tencent)", "purpose": "确认 dispatch 与 rabbitmq 不是直接 Down。", "expected_result": "相关容器保持 Up。"}, {"cmd": "MCP: logs(tencent, dispatch)", "purpose": "确认 dispatch 仍持续处理 oj-lag-* 合法订单消息。", "expected_result": "日志持续出现 oj-lag-* 订单消息，而不是完全静默或崩溃退出。"}], "success_criteria": ["dispatch CPU 配额恢复为基线值。", "dispatch 消费者仍保持 active，但不再处于限速状态。", "下单或支付后的异步业务完成恢复正常。"]}, "proposed_resolution": {"suggestion": "解除 dispatch 的 CPU 限制，恢复腾讯云异步消费链路的正常处理能力；如果观察到持续异常业务压力，也应一并清除。", "fix_steps": ["恢复 dispatch 的 CPU quota 到原始配置。", "再次检查 mq summary，确认 active consumers、CPU quota 与异步完成状态恢复正常。", "重新验证下单或支付后的异步完成路径是否恢复正常。"]}, "confidence": 0.92}, "rubrics": {"positive_points_total": 108, "negative_points_total": -52, "rubrics": [{"criterion": "准确指出根因是腾讯云 dispatch 消费链路被施加低 CPU 配额，导致消费者仍然存活但处理能力被人为压低；如果只写 dispatch Down、RabbitMQ 整体不可用、支付 API 同步失败或数据库故障，不命中本项。", "points": 22, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确指出 faulty_clouds 与 affected_clouds 都是腾讯云，并说明问题位于腾讯云异步消费链路而不是阿里云主入口本身。", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "完整描述端到端现象：下单或支付请求可以成功返回，但下游异步业务完成明显变慢；不能只写 API 成功或只写接口失败。", "points": 8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 MCP mq(tencent) 引用至少两类有效证据，例如 orders 队列消费者仍存在、dispatch_active_consumers > 0、dispatch_cpu_quota=5000/100000、recent_dispatch_order_logs 持续增长、sync_api_vs_async_state=api_success_business_pending。", "points": 18, "tags": ["level:example", "axis:evidence"]}, {"criterion": "通过 MCP status(tencent) 或同等证据说明 dispatch 与 rabbitmq 容器保持 Up，从而排除 dispatch 直接宕掉的误判。", "points": 10, "tags": ["level:example", "axis:evidence"]}, {"criterion": "引用 dispatch 日志中的合法订单消息迹象，例如持续出现 oj-lag-* 订单消息，证明消费者仍在处理消息而不是已经消失。", "points": 10, "tags": ["level:example", "axis:evidence"]}, {"criterion": "给出完整因果链：dispatch CPU 被限制 -> 消费者仍 active 但处理能力下降 -> 同步 API 可成功但异步完成变慢；如果还能结合持续订单处理迹象或持续业务压力信号进一步支撑，可视为充分命中。", "points": 12, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "恢复方案必须指向恢复 dispatch CPU quota，并在观察到持续异常业务压力时一并消除该压力来源，而不是清空队列、重装服务、修改业务代码或只重启 web/payment。", "points": 7, "tags": ["level:example", "axis:completeness"]}, {"criterion": "恢复验证必须包含：CPU quota 已恢复、dispatch 消费者仍 active、异步业务完成恢复正常；如果故障时存在持续异常业务压力，还应确认对应压力信号已经恢复正常。", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "Agent 过程至少一次明确区分“消费者仍活着但被限速”和“消费者宕机/队列不可用”这两类情况，而不是把两者混为一谈。", "points": 5, "tags": ["level:example", "axis:process_quality"]}, {"criterion": "Agent 总工具调用次数 <= 18，且没有用大量无关 web/payment 日志替代 mq/status/logs 证据链。", "points": 3, "tags": ["level:example", "axis:tool_efficiency"]}, {"criterion": "回答结构清晰，能够把结论、证据、因果链、恢复建议和验证方式对应起来。", "points": 2, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "错误地把根因判断为 dispatch 容器 Down、RabbitMQ 不可用、payment API 同步失败、数据库故障或前端页面问题。", "points": -14, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "没有 mq(tencent) 关键证据，只根据“订单完成慢”泛泛猜测 MQ 或 worker 有问题。", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "忽略“同步 API 成功、异步完成变慢”这一关键差异，把 API 200 直接当成业务正常，或反过来误写成同步接口失败。", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "建议清空 RabbitMQ、删除业务数据、重装服务或其他破坏性操作，而没有必要性说明。", "points": -8, "tags": ["level:example", "axis:safety"]}, {"criterion": "编造不存在的 poison message、dispatch panic、数据库报错或云告警，作为关键证据支撑结论。", "points": -6, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "Agent 总工具调用次数 >= 27，或大量查看无关服务输出却没有形成 mq/status/logs 的核心证据链。", "points": -6, "tags": ["level:example", "axis:tool_efficiency"]}]}, "prompt": "用户提交下单或支付操作后，页面可能显示请求已成功，但订单发货、配送或状态流转长时间没有更新。"}
{"canary": "mc_cdn_static_host_403", "tags": ["定位", "服务发现与流量入口", "CDN 回源 Host 不匹配", "静态资源 403", "hard", "multi-cloud"], "case": {"id": "mc_cdn_static_host_403", "title": "部分静态资源访问异常", "inject_script": "./faults/mc_cdn_static_host_403/inject.sh", "recover_script": "./faults/mc_cdn_static_host_403/recover.sh", "fault_phenomenon": "用户打开商城页面时，主体页面可能能显示，但部分图片、样式或脚本加载失败，页面表现为样式错乱、图片缺失或局部功能不可用。", "public_case_info": "该题基于部署在阿里云与腾讯云上的 Stan's Robot Shop 微服务系统。可观察入口：多云全局入口 http://8.163.90.83:18080/，阿里云直连入口 http://8.163.90.83:18081/，腾讯云直连入口 http://43.139.215.81:18081/。平台提供只读 MCP 诊断能力；请根据题面现象和 MCP 证据定位故障根因、影响范围、证据链和恢复建议。", "order_id": 36, "submission_enabled": true, "ai_analysis_visible": true}, "ideal_answer": {"fault_info": {"root_cause": "阿里云全局 gateway 的静态资源回源 Host/cache 规则错误，要求 Host=assets.robot-shop.local，普通全局入口 Host 请求静态资源时被 403。", "faulty_clouds": ["aliyun"], "affected_clouds": ["aliyun"], "fault_location": {"module": "cdn-gateway", "file_path": "/opt/mc-robot-shop/gateway/conf.d/mc-gateway.conf", "function_or_config": "static resource Host rule", "description": "故障在全局入口静态资源规则，不是业务 API 或数据库。"}}, "reasoning_process": {"observed_symptoms": ["HTML/API 基本正常但静态资源 403。", "两云业务容器正常。", "资源错误集中在 js/css/image。"], "key_evidence": [{"source": "multi-cloud-ssh.cdn()", "content": "CDN 摘要显示 static_403_rate elevated 和 origin_host_mismatch=true。", "conclusion": "静态资源 Host 规则是根因。"}, {"source": "multi-cloud-ssh.gateway()", "content": "gateway 配置对静态资源 Host 不匹配返回 403。", "conclusion": "定位到全局入口规则。"}, {"source": "multi-cloud-ssh.curl_loop(aliyun, global, /assets/app.js)", "content": "静态资源路径返回 403。", "conclusion": "只影响资源类路径。"}], "causal_chain": ["静态资源 location 要求特殊 Host。", "用户通过普通 Host 访问全局入口。", "静态资源命中 Host mismatch 403。", "HTML/API 正常造成后端健康烟雾弹。"]}, "verification_method": {"verification_commands": [{"cmd": "MCP: cdn()", "purpose": "查看静态资源诊断。", "expected_result": "恢复后 static_403_rate 正常。"}, {"cmd": "MCP: gateway()", "purpose": "检查 Host 规则。", "expected_result": "恢复后无错误静态资源 403 规则。"}, {"cmd": "MCP: curl_loop(aliyun, global, /, 3)", "purpose": "确认入口恢复。", "expected_result": "页面和 API 正常。"}], "success_criteria": ["gateway baseline 恢复。", "静态资源不再因 Host mismatch 403。", "HTML/API 仍正常。"]}, "proposed_resolution": {"suggestion": "回滚阿里云全局 gateway 的静态资源 Host 规则，恢复正常回源 Host 转发。", "fix_steps": ["恢复 mc-gateway.conf.baseline。", "重启 mc-gateway。", "验证静态资源和首页/API。", "增加静态资源合成探测和 CDN Host 发布校验。"]}, "confidence": 0.91}, "rubrics": {"positive_points_total": 108, "negative_points_total": -60, "rubrics": [{"criterion": "准确指出根因是阿里云全局 gateway 的静态资源回源 Host/cache 规则错误导致部分资源 403；只写 CDN 故障或权限问题不得命中本项。", "points": 22, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确写出 faulty_clouds 为阿里云、affected_clouds 为阿里云，并说明异常集中在阿里云全局 gateway 的静态资源规则。", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "完整描述资源类型差异：HTML/API 基本 200，但部分 JS/CSS/图片等静态资源 403。", "points": 10, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 MCP cdn() 引用静态资源 Host、403、assets 或回源规则线索。", "points": 14, "tags": ["level:example", "axis:evidence"]}, {"criterion": "通过 MCP gateway() 或 diff_config(aliyun) 定位到全局 gateway 静态资源 Host/cache 规则差异。", "points": 14, "tags": ["level:example", "axis:evidence"]}, {"criterion": "给出完整因果链：普通 Host 访问静态资源 -> gateway 静态资源规则拒绝/回源错误 -> 页面部分资源 403。", "points": 12, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "明确排除近似误判：业务 API、数据库、Robot Shop 容器、浏览器缓存、对象文件丢失。", "points": 8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "恢复方案必须指向还原阿里云 gateway 静态资源 Host/cache 规则，而不是重启后端服务或修改数据库。", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "恢复验证必须包含静态资源 200、HTML/API 继续正常、cdn()/gateway 配置回到基线。", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "Agent 过程先区分 HTML/API 与静态资源，再调用 cdn() 和 gateway()/diff_config(aliyun) 验证配置。", "points": 5, "tags": ["level:example", "axis:process_quality"]}, {"criterion": "Agent 总工具调用次数 <=16，且静态资源 403 出现后没有大量探索 Redis、MQ、netem 或业务日志。", "points": 4, "tags": ["level:example", "axis:tool_efficiency"]}, {"criterion": "回答结构清晰，最终结论、证据、因果链、恢复和验证彼此一致。", "points": 3, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "错误地把根因判断为业务 API、数据库、容器、浏览器缓存、文件丢失或泛化 CDN 故障，未指出 gateway Host/cache 规则。", "points": -14, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "只看到 403 就归因权限或安全组，没有 MCP cdn()/gateway()/diff_config 证据。", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "忽略 HTML/API 与静态资源路径的差异，把问题写成整站不可用或后端服务不可用。", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "建议删除静态文件、重建容器、关闭安全组或清空数据库等破坏性操作。", "points": -8, "tags": ["level:example", "axis:safety"]}, {"criterion": "编造不存在的 CDN 厂商告警、对象存储日志、浏览器缓存命中记录或文件缺失证据。", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "Agent 总工具调用次数 >=24，或静态资源 403 已出现后仍调用业务 logs/redis/mq/netem 等无关工具合计 >=8 次。", "points": -10, "tags": ["level:example", "axis:tool_efficiency"]}]}, "prompt": "用户打开商城页面时，主体页面可能能显示，但部分图片、样式或脚本加载失败，页面表现为样式错乱、图片缺失或局部功能不可用。"}
{"canary": "mc_trace_context_break", "tags": ["定位", "可观测系统自身", "Trace context 传播中断", "调用链断裂", "hard", "multi-cloud"], "case": {"id": "mc_trace_context_break", "title": "调用链定位信息不完整", "inject_script": "./faults/mc_trace_context_break/inject.sh", "recover_script": "./faults/mc_trace_context_break/recover.sh", "fault_phenomenon": "用户在支付或配送相关操作中可能遇到偶发慢请求或异常；从页面上看业务并非完全不可用，但问题定位信息不完整。", "public_case_info": "该题基于部署在阿里云与腾讯云上的 Stan's Robot Shop 微服务系统。可观察入口：多云全局入口 http://8.163.90.83:18080/，阿里云直连入口 http://8.163.90.83:18081/，腾讯云直连入口 http://43.139.215.81:18081/。平台提供只读 MCP 诊断能力；请根据题面现象和 MCP 证据定位故障根因、影响范围、证据链和恢复建议。", "order_id": 37, "submission_enabled": true, "ai_analysis_visible": true}, "ideal_answer": {"fault_info": {"root_cause": "阿里云全局 gateway 对 /api/payment/ 与 /api/shipping/ 丢弃 traceparent/tracestate，导致下游 span 无法与入口请求关联，根因藏在下游但观测链路断裂。", "faulty_clouds": ["aliyun"], "affected_clouds": ["aliyun"], "fault_location": {"module": "observability-gateway", "file_path": "/opt/mc-robot-shop/gateway/conf.d/mc-gateway.conf", "function_or_config": "traceparent/tracestate proxy_set_header", "description": "故障在观测上下文传播，不是业务服务 Down。"}}, "reasoning_process": {"observed_symptoms": ["trace 调用链在 gateway 后断裂。", "payment/shipping 业务现象无法和入口请求关联。", "容器状态和基础 API 可能正常。"], "key_evidence": [{"source": "multi-cloud-ssh.trace(aliyun)", "content": "trace 摘要显示 orphan span rate elevated。", "conclusion": "观测上下文丢失。"}, {"source": "multi-cloud-ssh.gateway()", "content": "payment/shipping location 将 traceparent/tracestate 置空。", "conclusion": "定位到 gateway 头部转发。"}, {"source": "multi-cloud-ssh.probe(aliyun)", "content": "业务入口基础可用。", "conclusion": "不是服务完全不可用。"}], "causal_chain": ["gateway 特定 API location 清空 trace headers。", "下游收不到入口 trace context。", "trace 系统产生孤儿 span 或断裂拓扑。", "观测层误导排障。"]}, "verification_method": {"verification_commands": [{"cmd": "MCP: trace(aliyun)", "purpose": "检查 trace 摘要和 header snippets。", "expected_result": "恢复后 traceparent forwarding enabled。"}, {"cmd": "MCP: gateway()", "purpose": "查看 proxy_set_header。", "expected_result": "恢复后不再清空 trace headers。"}, {"cmd": "MCP: probe(aliyun)", "purpose": "验证入口可用。", "expected_result": "业务 API 正常。"}], "success_criteria": ["gateway baseline 恢复。", "trace 摘要显示 orphan span 恢复正常。", "payment/shipping 路径不再丢 trace context。"]}, "proposed_resolution": {"suggestion": "恢复 gateway 对 traceparent/tracestate 的透传，并补充 trace context 发布校验。", "fix_steps": ["恢复 mc-gateway.conf.baseline。", "重启 mc-gateway。", "验证 trace 摘要和 header 配置。", "增加 trace header 合成测试。"]}, "confidence": 0.91}, "rubrics": {"positive_points_total": 116, "negative_points_total": -60, "rubrics": [{"criterion": "准确指出根因是阿里云全局 gateway 对 payment/shipping 相关路径丢弃或未透传 traceparent/tracestate；只写监控故障不得命中本项。", "points": 22, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确写出 faulty_clouds 为阿里云、affected_clouds 为阿里云，并说明异常集中在阿里云全局 gateway 的 trace context 透传。", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "完整描述现象：业务接口不一定整体 Down，但调用链定位信息不完整，下游 span 难以关联入口请求。", "points": 8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 MCP trace() 引用 trace-summary 中 span 断裂、孤立 span 或上下游无法关联的证据。", "points": 12, "tags": ["level:example", "axis:evidence"]}, {"criterion": "通过 MCP trace()、gateway() 或 diff_config(aliyun) 引用 traceparent/tracestate 被清空、缺失或未透传的配置片段。", "points": 18, "tags": ["level:example", "axis:evidence"]}, {"criterion": "给出完整因果链：gateway 丢失 trace context -> 下游服务生成无法关联的 span -> 根因定位信息不完整。", "points": 12, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "明确排除近似误判：业务容器 Down、tracing 后端宕机、采样率过低、数据库故障。", "points": 8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "恢复方案必须指向还原 gateway 对 traceparent/tracestate 的透传规则，而不是只重启 tracing 或下游服务。", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "恢复验证必须包含入口 span 与下游 span 重新关联，业务接口保持基线状态。", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "Agent 过程先通过 MCP probe/curl_loop 确认业务接口不是整体 Down，再调用 trace() 检查观测上下文。", "points": 8, "tags": ["level:example", "axis:process_quality"]}, {"criterion": "Agent 总工具调用次数 <=20，且没有用大量业务日志替代 trace context 证据。", "points": 5, "tags": ["level:example", "axis:tool_efficiency"]}, {"criterion": "回答结构清晰，最终结论、证据、因果链、恢复和验证彼此一致。", "points": 3, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "提出针对 trace context 传播的预防措施，例如入口代理配置测试、跨服务 trace header 探测、配置审计。", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "错误地把根因判断为业务服务 Down、tracing 后端宕机、采样率过低、数据库故障或泛化监控缺失。", "points": -14, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "没有 MCP trace() 证据，只根据调用链不完整现象猜测可观测性平台问题。", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "忽略业务接口未整体 Down 与 trace 断裂之间的区别，把诊断写成普通业务故障。", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "建议删除数据、重装 tracing、关闭采样或重启全部服务等不必要破坏性操作。", "points": -8, "tags": ["level:example", "axis:safety"]}, {"criterion": "编造不存在的 trace 后端报错、采样配置、span 内容或日志作为关键证据。", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "Agent 总工具调用次数 >=29，或未通过 MCP 调用 trace() 却反复调用 logs/status/probe 合计 >=16 次。", "points": -10, "tags": ["level:example", "axis:tool_efficiency"]}]}, "prompt": "用户在支付或配送相关操作中可能遇到偶发慢请求或异常；从页面上看业务并非完全不可用，但问题定位信息不完整。"}
{"canary": "mc_clock_skew_signature", "tags": ["定位", "变更层", "时钟偏移", "签名过期", "hard", "multi-cloud"], "case": {"id": "mc_clock_skew_signature", "title": "支付请求偶发认证失败", "inject_script": "./faults/mc_clock_skew_signature/inject.sh", "recover_script": "./faults/mc_clock_skew_signature/recover.sh", "fault_phenomenon": "用户进行支付或下单相关操作时，偶尔收到认证失败或 401 类提示；重复操作可能有时成功、有时失败，其他页面功能大多正常。", "public_case_info": "该题基于部署在阿里云与腾讯云上的 Stan's Robot Shop 微服务系统。可观察入口：多云全局入口 http://8.163.90.83:18080/，阿里云直连入口 http://8.163.90.83:18081/，腾讯云直连入口 http://43.139.215.81:18081/。平台提供只读 MCP 诊断能力；请根据题面现象和 MCP 证据定位故障根因、影响范围、证据链和恢复建议。", "order_id": 38, "submission_enabled": true, "ai_analysis_visible": true}, "ideal_answer": {"fault_info": {"root_cause": "腾讯云 payment 签名校验侧被安全地模拟为比标准时间快 487 秒，超过请求 timestamp/JWT 接受窗口，导致支付路径 401；没有修改宿主机真实时间。", "faulty_clouds": ["tencent"], "affected_clouds": ["tencent"], "fault_location": {"module": "payment-signature-window", "file_path": "/opt/mc-robot-shop/edge/default.conf", "function_or_config": "default.clock-skew.conf payment signature shim", "description": "故障模拟在腾讯云 payment 签名校验窗口，不是宿主机真实时间被修改。"}}, "reasoning_process": {"observed_symptoms": ["payment API 返回 401 signature timestamp outside accepted window。", "容器 Running，网络基础探测正常。", "问题集中在腾讯云路径或全局入口命中腾讯云时。"], "key_evidence": [{"source": "multi-cloud-ssh.host_time(tencent)", "content": "time 摘要显示 payment validator clock skew +487s。", "conclusion": "签名窗口时间不一致是根因。"}, {"source": "multi-cloud-ssh.edge(tencent)", "content": "payment 路径返回 X-Signature-Clock 和 401 shim。", "conclusion": "定位到签名校验侧。"}, {"source": "multi-cloud-ssh.probe(tencent)", "content": "其他基础 API 正常。", "conclusion": "排除网络/数据库整体故障。"}], "causal_chain": ["腾讯云 payment 校验侧时间窗口偏移。", "请求 timestamp/JWT 与校验侧时间差超出允许范围。", "payment 路径返回 401。", "容器健康和其他 API 正常造成认证烟雾弹。"]}, "verification_method": {"verification_commands": [{"cmd": "MCP: host_time(tencent)", "purpose": "查看时间漂移摘要。", "expected_result": "恢复后 clock_skew_seconds=0。"}, {"cmd": "MCP: edge(tencent)", "purpose": "检查 payment shim。", "expected_result": "恢复后 edge baseline。"}, {"cmd": "MCP: probe(tencent)", "purpose": "验证 payment metrics 和入口。", "expected_result": "恢复后关键 API 正常。"}], "success_criteria": ["腾讯云 edge 恢复 baseline。", "time 摘要恢复 aligned。", "payment 路径不再 401。"]}, "proposed_resolution": {"suggestion": "恢复腾讯云 payment 路径的签名校验时间窗口模拟配置，确保跨云签名时间一致。", "fix_steps": ["恢复 edge/default.conf.baseline。", "重启腾讯云 edge。", "验证 time 摘要和 payment 路径。", "增加 NTP/签名窗口/证书时间一致性检查。"]}, "confidence": 0.91}, "rubrics": {"positive_points_total": 108, "negative_points_total": -60, "rubrics": [{"criterion": "准确指出根因是腾讯云 payment 签名校验侧的容器/配置级时间偏移 shim 约 487 秒，超过 timestamp/JWT 接受窗口；只写认证失败不得命中本项。", "points": 22, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "正确写出 faulty_clouds 为腾讯云、affected_clouds 为腾讯云，并说明阿里云全局入口只是可能转发到腾讯云支付故障路径，不作为直接异常云。", "points": 5, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "完整描述接口差异：payment/checkout 偶发 401，而其他 API 或首页基本正常。", "points": 8, "tags": ["level:example", "axis:completeness"]}, {"criterion": "通过 MCP host_time(tencent) 引用 payment container clock、time-summary、约 487s 偏移或超过签名窗口的信息。", "points": 16, "tags": ["level:example", "axis:evidence"]}, {"criterion": "通过 MCP edge(tencent) 或 diff_config(tencent) 识别 signature shim，并说明没有修改宿主机真实时间。", "points": 12, "tags": ["level:example", "axis:evidence"]}, {"criterion": "给出完整因果链：校验侧时间偏移 -> 请求 timestamp/JWT 超出窗口 -> payment/checkout 401。", "points": 10, "tags": ["level:cluster", "cluster:evidence_based_root_cause_tracing", "axis:accuracy"]}, {"criterion": "明确排除近似误判：TLS 证书、JWT 密钥、用户密码/会话、payment 服务 Down、数据库或 DNS。", "points": 8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "恢复方案必须指向移除签名校验时间偏移 shim 或还原腾讯云 edge/payment 配置，而不是修改宿主机系统时间。", "points": 6, "tags": ["level:example", "axis:completeness"]}, {"criterion": "恢复验证必须包含 payment/checkout 不再 401、其他 API 正常、host_time(tencent) 回到基线。", "points": 5, "tags": ["level:example", "axis:completeness"]}, {"criterion": "Agent 过程先通过 MCP probe/curl_loop 区分 payment/checkout 与其他 API，再调用 host_time(tencent) 验证时间窗口。", "points": 6, "tags": ["level:example", "axis:process_quality"]}, {"criterion": "Agent 总工具调用次数 <=18，且没有反复查看 payment/web 日志替代 host_time 证据。", "points": 4, "tags": ["level:example", "axis:tool_efficiency"]}, {"criterion": "回答结构清晰，最终结论、证据、因果链、恢复和验证彼此一致。", "points": 3, "tags": ["level:example", "axis:communication_quality"]}, {"criterion": "提出针对签名时间窗口的预防措施，例如跨云时钟一致性校验、签名窗口监控、容器配置审计。", "points": 3, "tags": ["level:example", "axis:context_awareness"]}, {"criterion": "错误地把根因判断为 TLS 证书、JWT 密钥、用户会话、payment 服务 Down、数据库、DNS 或宿主机真实时间被改。", "points": -14, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "没有 MCP host_time(tencent) 或 signature shim 证据，只根据 401 猜测认证配置问题。", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "忽略 payment/checkout 与其他 API 的差异，把问题写成整站登录态或全局认证失败。", "points": -8, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "建议直接修改宿主机系统时间、重装服务、删除用户数据或重置密钥作为主要恢复手段。", "points": -8, "tags": ["level:example", "axis:safety"]}, {"criterion": "编造不存在的 TLS/JWT 日志、证书过期、云厂商时间告警或数据库错误作为证据。", "points": -10, "tags": ["level:example", "axis:accuracy"]}, {"criterion": "Agent 总工具调用次数 >=28，或未通过 MCP 调用 host_time(tencent) 却反复查看 payment/web logs、status、probe 合计 >=15 次。", "points": -10, "tags": ["level:example", "axis:tool_efficiency"]}]}, "prompt": "用户进行支付或下单相关操作时，偶尔收到认证失败或 401 类提示；重复操作可能有时成功、有时失败，其他页面功能大多正常。"}