File size: 15,506 Bytes
2794920
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
557930c
 
 
 
 
 
 
 
 
 
 
 
 
 
2794920
 
 
eb895b1
804f70e
2794920
557930c
2794920
557930c
c8f3b98
 
557930c
 
 
 
 
 
 
 
eb895b1
557930c
 
 
 
eb895b1
 
 
557930c
 
 
 
eb895b1
 
 
 
 
 
 
 
 
c8f3b98
eb895b1
c8f3b98
 
 
 
 
 
 
 
 
 
557930c
eb895b1
557930c
 
eb895b1
557930c
 
 
 
 
 
 
 
c8f3b98
 
557930c
 
 
2794920
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
557930c
 
 
 
2794920
557930c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
eb895b1
 
 
 
557930c
eb895b1
557930c
 
 
eb895b1
557930c
 
eb895b1
 
557930c
eb895b1
557930c
 
 
eb895b1
557930c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2794920
 
 
557930c
 
2794920
 
 
 
557930c
2794920
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
557930c
 
2794920
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
557930c
2794920
 
804f70e
2794920
557930c
 
 
2794920
 
 
 
 
 
557930c
 
 
2794920
557930c
2794920
 
 
 
 
 
 
 
557930c
 
 
2794920
 
 
557930c
 
 
 
2794920
557930c
 
2794920
557930c
 
85b7ac8
 
 
557930c
2794920
 
 
557930c
 
 
19ed2d4
2794920
557930c
2794920
eb895b1
557930c
 
 
 
 
 
 
 
 
 
 
85b7ac8
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
"""
Inference Script for Cloud-Native Debug Environment
===================================
MANDATORY
- Before submitting, ensure the following variables are defined in your environment configuration:
    API_BASE_URL   The API endpoint for the LLM.
    MODEL_NAME     The model identifier to use for inference.
    HF_TOKEN       Your Hugging Face / API key.
    LOCAL_IMAGE_NAME The name of the local image to use for the environment if you are using from_docker_image()
                     method

- Defaults are set only for API_BASE_URL and MODEL_NAME
    (and should reflect your active inference setup):
    API_BASE_URL = os.getenv("API_BASE_URL", "<your-active-endpoint>")
    MODEL_NAME = os.getenv("MODEL_NAME", "<your-active-model>")

- The inference script must be named `inference.py` and placed in the root directory of the project
- Participants must use OpenAI Client for all LLM calls using above variables

STDOUT FORMAT
- The script must emit exactly three line types to stdout, in this order:

    [START] task=<task_name> env=<benchmark> model=<model_name>
    [STEP]  step=<n> action=<action_str> reward=<0.00> done=<true|false> error=<msg|null>
    [END]   success=<true|false> steps=<n> score=<score> rewards=<r1,r2,...,rn>

  Rules:
    - One [START] line at episode begin.
    - One [STEP] line per step, immediately after env.step() returns.
    - One [END] line after the episode completes, always emitted (even on exception).
    - reward and rewards are formatted to 2 decimal places.
    - done and success are lowercase booleans: true or false.
    - error is the raw error string, or null if none.
    - All fields on a single line with no newlines within a line.
    - Each tasks should return score in [0, 1]

  Example:
    [START] task=dockerfile_syntax env=cloud_native_devops model=meta-llama/Llama-3.1-70B-Instruct
    [STEP] step=1 action=edit_file reward=0.30 done=false error=null
    [STEP] step=2 action=submit reward=0.00 done=true error=null
    [END] success=true steps=2 score=0.850 rewards=0.30,0.00
"""


import json
import os
import re
import sys
import time
from typing import Any, Dict, List, Optional

import requests
from openai import OpenAI


API_BASE_URL = os.getenv("API_BASE_URL") or "https://router.huggingface.co/v1"
MODEL_NAME = os.getenv("MODEL_NAME") or "meta-llama/Llama-3.1-70B-Instruct"
API_KEY = os.getenv("HF_TOKEN") or os.getenv("API_KEY")
ENV_URL = os.getenv("ENV_URL", "http://localhost:7860")
LOCAL_IMAGE_NAME = os.getenv("LOCAL_IMAGE_NAME")
BENCHMARK = "cloud_native_devops"
MAX_STEPS = 8  # leave 2 steps buffer before env hard-limit of 10
SUCCESS_SCORE_THRESHOLD = 0.1  # normalized score in [0, 1]

SYSTEM_PROMPT = """You are an expert DevOps engineer debugging cloud-native deployment pipelines.
You will receive broken Dockerfile, GitHub Actions workflow, and/or Kubernetes manifest files along with error messages.

Your job is to:
1. Analyze the error message carefully
2. Identify the root cause in the configuration files
3. Provide a precise fix

When you identify a fix, respond with a JSON object in this exact format:
{
  "action_type": "YOUR_CHOSEN_ACTION_TYPE",
  "reasoning": "Brief explanation of the bug and fix",
  "edits": [
    {
      "file_path": "path/to/file",
      "line_number": 5,               // Only needed for replace_line, add_line, delete_line, add_block
      "old_content": "exactly broken", // Only needed for edit_file, delete_block
      "new_content": "corrected block" // Not needed for delete_line, delete_block
    }
  ]
}

Available action_type values for edits:
- "edit_file" (requires old_content and new_content)
- "replace_line" (requires line_number and new_content)
- "add_line" (requires line_number and new_content)
- "delete_line" (requires line_number)
- "add_block" (requires line_number and new_content)
- "delete_block" (requires old_content)

To create a new file (e.g. a missing ConfigMap), use "edit_file" with empty old_content:
{
  "action_type": "edit_file",
  "reasoning": "Create missing ConfigMap manifest",
  "edits": [
    {
      "file_path": "k8s/configmap.yaml",
      "old_content": "",
      "new_content": "apiVersion: v1\\nkind: ConfigMap\\n..."
    }
  ]
}

If you believe all issues are fixed and want to submit, respond with:
{"action_type": "submit"}

If you need a hint, respond with:
{"action_type": "request_hint"}

Rules:
- Match old_content EXACTLY as it appears in the file (whitespace matters)
- Fix one issue at a time for precision
- Focus on the error message — it tells you exactly what's wrong
- Common issues: typos, wrong syntax, missing fields, wrong secret references
- For GitHub Actions: check secret syntax (${{ }} not ${ }), env blocks, permissions
- For Dockerfiles: check instruction syntax, file paths, base image tags
- For Kubernetes: check label selectors, port matching, resource limits, probe configs, ingress rules
- For full-stack pipelines: issues may span multiple files (workflow + Dockerfile + K8s manifests)
- Always respond with valid JSON only, no markdown fences"""


# ---------------------------------------------------------------------------
# Logging helpers (mandatory stdout format)
# ---------------------------------------------------------------------------

def log_start(task: str, env: str, model: str) -> None:
    print(f"[START] task={task} env={env} model={model}", flush=True)


def log_step(step: int, action: str, reward: float, done: bool, error: Optional[str]) -> None:
    error_val = error if error else "null"
    done_val = str(done).lower()
    print(
        f"[STEP] step={step} action={action} reward={reward:.2f} done={done_val} error={error_val}",
        flush=True,
    )


def log_end(success: bool, steps: int, score: float, rewards: List[float]) -> None:
    rewards_str = ",".join(f"{r:.2f}" for r in rewards)
    print(f"[END] success={str(success).lower()} steps={steps} score={score:.3f} rewards={rewards_str}", flush=True)


# ---------------------------------------------------------------------------
# Client / env helpers
# ---------------------------------------------------------------------------

def create_client() -> OpenAI:
    """Create OpenAI-compatible client for HuggingFace router."""
    return OpenAI(
        base_url=API_BASE_URL,
        api_key=API_KEY,
    )


def env_request(method: str, endpoint: str, json_data: Optional[Dict] = None) -> Dict[str, Any]:
    """Make a request to the environment server."""
    url = f"{ENV_URL}{endpoint}"
    if method == "GET":
        resp = requests.get(url, timeout=30)
    else:
        resp = requests.post(url, json=json_data or {}, timeout=30)
    resp.raise_for_status()
    return resp.json()


def format_observation(obs: Dict[str, Any]) -> str:
    """Format observation into a prompt for the LLM."""
    parts = []
    parts.append(f"Task: {obs.get('task_description', 'Unknown')}")
    parts.append(f"Difficulty: {obs.get('difficulty', 'unknown')}")
    parts.append(f"Step: {obs.get('step_number', 0)}/{obs.get('max_steps', 10)}")
    parts.append(f"Issues fixed: {obs.get('issues_fixed', 0)}/{obs.get('total_issues', '?')}")

    error = obs.get("error", {})
    parts.append(f"\n--- ERROR ---")
    parts.append(f"Phase: {error.get('phase', 'unknown')}")
    parts.append(f"Message: {error.get('error_message', 'No error')}")
    if error.get("failed_step"):
        parts.append(f"Failed step: {error['failed_step']}")
    if error.get("line_hint"):
        parts.append(f"Line hint: {error['line_hint']}")

    parts.append(f"\n--- FILES ---")
    for f in obs.get("files", []):
        parts.append(f"\n=== {f['path']} ({f.get('file_type', 'unknown')}) ===")
        content = f.get("content", "")
        lines = content.split("\n")
        for i, line in enumerate(lines, 1):
            parts.append(f"{i:3d} | {line}")

    if obs.get("available_secrets"):
        parts.append(f"\n--- AVAILABLE SECRETS ---")
        parts.append(", ".join(obs["available_secrets"]))

    if obs.get("last_action_feedback"):
        parts.append(f"\n--- LAST ACTION FEEDBACK ---")
        parts.append(obs["last_action_feedback"])

    return "\n".join(parts)


def parse_llm_response(text: str) -> Dict[str, Any]:
    """Parse LLM response into an action dict."""
    text = text.strip()

    # Strip markdown code fences if present
    if text.startswith("```"):
        lines = text.split("\n")
        lines = [l for l in lines if not l.strip().startswith("```")]
        text = "\n".join(lines).strip()

    # Try to find JSON in the response
    json_match = re.search(r'\{[\s\S]*\}', text)
    if json_match:
        try:
            return json.loads(json_match.group())
        except json.JSONDecodeError:
            pass

    # Fallback: treat as submit
    return {"action": "submit"}


def build_action(parsed: Dict[str, Any]) -> Dict[str, Any]:
    """Convert parsed LLM response to environment action format."""
    action_type = parsed.get("action_type")

    # Backwards compatibility and standard aliases
    if parsed.get("action") == "submit" or action_type == "submit":
        return {"action_type": "submit"}
    if parsed.get("action") == "hint" or action_type == "request_hint":
        return {"action_type": "request_hint"}

    edits = parsed.get("edits", [])
    if not edits and not action_type:
        return {"action_type": "submit"}

    action_str = action_type if action_type else "edit_file"

    return {
        "action_type": action_str,
        "edits": [
            {
                "file_path": e.get("file_path", ""),
                "line_number": e.get("line_number"),
                "old_content": e.get("old_content", ""),
                "new_content": e.get("new_content", ""),
            }
            for e in edits
        ],
    }


def run_episode(client: OpenAI, task_id: Optional[str] = None, scenario_id: Optional[str] = None) -> Dict[str, Any]:
    """Run a single episode: reset, loop (observe -> LLM -> act), grade."""
    reset_payload: Dict[str, Any] = {}
    if task_id:
        reset_payload["task_id"] = task_id
    if scenario_id:
        reset_payload["scenario_id"] = scenario_id

    # Best-effort task name for Start
    target_task = task_id or "random_task"
    log_start(task=target_task, env=BENCHMARK, model=MODEL_NAME)

    trajectory = []
    rewards: List[float] = []
    steps_taken = 0
    score = 0.0
    success = False

    try:
        reset_resp = env_request("POST", "/reset", reset_payload)
        obs = reset_resp["observation"]
        info = reset_resp.get("info", {})

        actual_task_id = info.get("task_id", target_task)
        actual_scenario_id = info.get("scenario_id", scenario_id or "unknown")

        messages = [{"role": "system", "content": SYSTEM_PROMPT}]
        for step_num in range(1, MAX_STEPS + 1):
            user_msg = format_observation(obs)
            messages.append({"role": "user", "content": user_msg})

            error_msg: Optional[str] = None

            try:
                completion = client.chat.completions.create(
                    model=MODEL_NAME,
                    messages=messages,
                    temperature=0.1,
                    max_tokens=1024,
                )
                llm_text = completion.choices[0].message.content or '{"action": "submit"}'
            except Exception as e:
                error_msg = str(e)
                print(f"[DEBUG] Model request failed: {e}", flush=True)
                llm_text = '{"action": "submit"}'

            messages.append({"role": "assistant", "content": llm_text})

            parsed = parse_llm_response(llm_text)
            action = build_action(parsed)

            step_resp = env_request("POST", "/step", {"action": action})
            obs = step_resp["observation"]
            reward = step_resp.get("reward", 0.0)
            done = step_resp.get("done", False)
            step_info = step_resp.get("info", {})
            steps_taken = step_num

            rewards.append(reward)

            log_step(
                step=step_num,
                action=action["action_type"],
                reward=reward,
                done=done,
                error=error_msg,
            )

            trajectory.append({
                "step": step_num,
                "action": action,
                "reward": reward,
                "done": done,
                "info": step_info,
            })

            if done:
                break

        # Grade the trajectory
        grade_resp = env_request("POST", "/grader", {
            "task_id": actual_task_id,
            "trajectory": trajectory,
        })
        result = grade_resp.get("result", {})
        score = result.get("score", 0.0)
        score = min(max(score, 0.0), 1.0)  # clamp to [0, 1]
        success = score >= SUCCESS_SCORE_THRESHOLD

    finally:
        log_end(success=success, steps=steps_taken, score=score, rewards=rewards)

    return {"score": score, "success": success, "steps": steps_taken, "rewards": rewards}


def run_all_tasks(client: OpenAI) -> Dict[str, float]:
    """Run baseline on all tasks (and ALL their scenarios) and report scores."""
    try:
        from server.tasks.task_registry import TASK_REGISTRY
    except ImportError as e:
        print(f"[DEBUG] Could not import TASK_REGISTRY: {e}", flush=True)
        return {}

    scores: Dict[str, List[float]] = {}

    for task_id, task_cls in TASK_REGISTRY.items():
        task_scores = []
        
        # Iterate over all exact scenarios for this task
        scenarios = task_cls.SCENARIOS
        for scenario in scenarios:
            scenario_id = scenario["id"]
            result = run_episode(client, task_id=task_id, scenario_id=scenario_id)
            task_scores.append(result.get("score", 0.0))
            
        scores[task_id] = task_scores

    # Summary
    print(f"\n[DEBUG] {'='*60}", flush=True)
    print("[DEBUG] BASELINE RESULTS SUMMARY", flush=True)
    print(f"[DEBUG] {'='*60}", flush=True)
    avg_scores = {}
    for task_id, task_scores in scores.items():
        avg = sum(task_scores) / len(task_scores) if task_scores else 0.0
        avg_scores[task_id] = avg
        print(f"[DEBUG]   {task_id:40s} {avg:.3f}", flush=True)

    overall = sum(avg_scores.values()) / len(avg_scores) if avg_scores else 0.0
    print(f"[DEBUG]   {'OVERALL':40s} {overall:.3f}", flush=True)

    return avg_scores


def main():
    """Entry point for baseline inference."""
    if not API_KEY:
        print("[DEBUG] WARNING: HF_TOKEN not set. Set it via: export HF_TOKEN=your_token_here", flush=True)
        print("[DEBUG] Continuing anyway (will fail if auth is required)...", flush=True)

    # Verify environment is running
    try:
        health = env_request("GET", "/health")
        print(f"[DEBUG] Environment status: {health.get('status', 'unknown')}", flush=True)
    except Exception as e:
        print(f"[DEBUG] Cannot connect to environment at {ENV_URL}: {e}", flush=True)
        print("[DEBUG] Start the server first: python -m uvicorn server.app:app --host 0.0.0.0 --port 7860", flush=True)
        sys.exit(1)

    client = create_client()

    # If a specific task is requested via CLI arg
    if len(sys.argv) > 1:
        task_id = sys.argv[1]
        scenario_id = sys.argv[2] if len(sys.argv) > 2 else None
        run_episode(client, task_id=task_id, scenario_id=scenario_id)
    else:
        run_all_tasks(client)


if __name__ == "__main__":
    main()