fix: improve SSH service stability and backup.py error handling

scripts/hf-entrypoint.sh

@@ -94,12 +94,26 @@ fi
 # ============================================
 echo "[$(date -u +%Y-%m-%dT%H:%M:%SZ)] hf-entrypoint: starting supervisord..."
 mkdir -p /var/run /var/log/supervisor /var/log/hf-entrypoint
+touch /var/log/hf-entrypoint/supervisord-stdout.log /var/log/hf-entrypoint/supervisord-stderr.log
 /usr/bin/supervisord -c /etc/supervisor/supervisord.conf \
-    >> /var/log/hf-entrypoint/supervisord-stdout.log \
-    2>> /var/log/hf-entrypoint/supervisord-stderr.log &
+    > >(tee -a /var/log/hf-entrypoint/supervisord-stdout.log) \
+    2> >(tee -a /var/log/hf-entrypoint/supervisord-stderr.log >&2) &
 SUPERVISORD_PID=$!
 echo "[$(date -u +%Y-%m-%dT%H:%M:%SZ)] hf-entrypoint: supervisord started (pid=$SUPERVISORD_PID)"
 
+# 启动后台 tail -f 进程，把 openclaw-gateway 的日志实时转发到容器 stdout/stderr
+# 这样用户在 docker logs 中可以看到 restore 的所有输出
+touch /var/log/hf-entrypoint/openclaw-gateway-stdout.log /var/log/hf-entrypoint/openclaw-gateway-stderr.log
+if command -v tail >/dev/null 2>&1; then
+    # stdout: prefix with [gateway-out so users can distinguish
+    tail -f /var/log/hf-entrypoint/openclaw-gateway-stdout.log 2>/dev/null &
+    TAIL_STDOUT_PID=$!
+    # stderr: prefix with [gateway-err]
+    tail -f /var/log/hf-entrypoint/openclaw-gateway-stderr.log 2>/dev/null >&2 &
+    TAIL_STDERR_PID=$!
+    echo "[$(date -u +%Y-%m-%dT%H:%M:%SZ)] hf-entrypoint: log forwarders started (stdout_pid=$TAIL_STDOUT_PID, stderr_pid=$TAIL_STDERR_PID)"
+fi
+
 while [[ ! -f /var/run/supervisord.pid ]]; do
     sleep 0.5
 done
@@ -410,6 +424,7 @@ SLOW_WARN_THRESHOLD=900   # 15分钟超时预警
 MAX_WAIT_TIMEOUT=28800     # 最大等待28800秒（8小时），超时后强制继续启动
 IDLE_WARN_THRESHOLD=120   # 日志无新内容120秒则告警
 TIME_NO_NEW_LOG=0
+gateway_was_seen_alive=false
 
 mkdir -p "$(dirname "$RESTORE_LOG_FILE")"
 
@@ -423,9 +438,10 @@ show_restore_progress() {
 
     if [[ -n "$current_lines" ]] && [[ "$current_lines" -gt "$LAST_RESTORE_LINE" ]]; then
         local new_lines=$((current_lines - LAST_RESTORE_LINE))
-        echo "[$(date -u +"%Y-%m-%dT%H:%M:%SZ")] hf-entrypoint: --- Restore log ($new_lines new lines) ---"
+        echo "[$(date -u +%Y-%m-%dT%H:%M:%SZ)] hf-entrypoint: --- Restore log ($new_lines new lines) ---"
         # FAST: use awk to prefix lines in a single call (no bash while-read loop)
-        tail -n "$new_lines" "$RESTORE_LOG_FILE" | awk -v ts="[$(date -u +"%Y-%m-%dT%H:%M:%SZ")]" '{print ts, $0}'
+        # 用 || true 保护管道，避免 set -o pipefail 触发脚本立即退出
+        tail -n "$new_lines" "$RESTORE_LOG_FILE" | awk -v ts="[$(date -u +%Y-%m-%dT%H:%M:%SZ)]" '{print ts, $0}' || true
         LAST_RESTORE_LINE="$current_lines"
         TIME_NO_NEW_LOG=0
     fi
@@ -494,27 +510,35 @@ while true; do
     # 使用 supervisorctl status 作为权威检查，同时检查实际进程名作为兜底
     # 注意: supervisord 中命令是 openclaw-entrypoint.sh gateway，不包含字符串 "openclaw-gateway"
     #       因此 pgrep -f "openclaw-gateway" 永远不匹配，必须检查实际的进程名
+    # 关键: 所有检查命令必须用 || true 保护，避免 set -e 误触发脚本立即退出
     gateway_alive=false
     if [[ -f "/var/run/supervisord.pid" ]]; then
         # 方法1: 用 supervisorctl 检查 program 状态（最权威，匹配 program:openclaw-gateway）
+        # supervisorctl status 在 program 不存在或状态不是 RUNNING 时返回非0，必须用 || true
         if command -v supervisorctl >/dev/null 2>&1; then
-            svc_status=$(supervisorctl status openclaw-gateway 2>/dev/null)
+            svc_status=$(supervisorctl status openclaw-gateway 2>/dev/null || true)
             if echo "$svc_status" | grep -qE "(RUNNING|STARTING)"; then
                 gateway_alive=true
             fi
         fi
         # 方法2: 兜底检查 - 检查实际运行的入口脚本或恢复进程
+        # pgrep 在无匹配时返回 1，在 if 条件中不会触发 set -e，但为了安全用 || true
         if [[ "$gateway_alive" != true ]]; then
-            if pgrep -f "openclaw-entrypoint" >/dev/null 2>&1 \
-               || pgrep -f "backup\.py restore" >/dev/null 2>&1 \
-               || pgrep -f "openclaw-restore" >/dev/null 2>&1; then
+            if { pgrep -f "openclaw-entrypoint" >/dev/null 2>&1 || true; } \
+               || { pgrep -f "backup\.py restore" >/dev/null 2>&1 || true; } \
+               || { pgrep -f "openclaw-restore" >/dev/null 2>&1 || true; }; then
                 gateway_alive=true
             fi
         fi
     fi
+    if [[ "$gateway_alive" == true ]]; then
+        gateway_was_seen_alive=true
+    fi
     if [[ "$gateway_alive" != true ]]; then
         # 如果 gateway 进程已经不存在，且未看到完成标志，视为恢复失败
-        if [[ ! -f "$RESTORE_COMPLETED_FILE" ]] && [[ $WAITED -gt 10 ]]; then
+        # 规则: 1) 如果曾经看到 gateway alive（gateway_was_seen_alive=true），立即检测为崩溃
+        #       2) 如果���未看到 alive，等待 10 秒后再判定（确保 supervisord 有时间启动它）
+        if [[ ! -f "$RESTORE_COMPLETED_FILE" ]] && { [[ "$gateway_was_seen_alive" == true ]] || [[ $WAITED -gt 10 ]]; }; then
             echo "[$(date -u +%Y-%m-%dT%H:%M:%SZ)] hf-entrypoint: ✗ openclaw-gateway process exited before restore completed, marking as failed"
             # 在退出前转储完整的恢复日志，让用户看到崩溃前发生了什么
             _dump_restore_logs "gateway process exited during restore (waited ${WAITED}s)"
@@ -558,6 +582,13 @@ while true; do
         if [[ -f "$OPENCLAW_STDERR_LOG" ]]; then
             _dump_log_file "GATEWAY STDERR (hidden Python errors/warnings check)" "$OPENCLAW_STDERR_LOG" 100
         fi
+        # 清理 tail 转发进程（避免继续写入 stdout 干扰后续日志）
+        if [[ -n "${TAIL_STDOUT_PID:-}" ]]; then
+            kill "$TAIL_STDOUT_PID" 2>/dev/null || true
+        fi
+        if [[ -n "${TAIL_STDERR_PID:-}" ]]; then
+            kill "$TAIL_STDERR_PID" 2>/dev/null || true
+        fi
         break
     fi
 
@@ -605,7 +636,7 @@ while true; do
             gw_stderr_size=$(stat -c%s "$OPENCLAW_STDERR_LOG" 2>/dev/null || echo "0")
             if [[ "$gw_stderr_size" -gt "0" ]]; then
                 echo "[$(date -u +"%Y-%m-%dT%H:%M:%SZ")] hf-entrypoint: gateway-stderr (last 30 lines, ${gw_stderr_size} bytes total) --"
-                tail -n 30 "$OPENCLAW_STDERR_LOG"
+                tail -n 30 "$OPENCLAW_STDERR_LOG" 2>/dev/null || true
                 echo "[$(date -u +"%Y-%m-%dT%H:%M:%SZ")] hf-entrypoint: -- end gateway-stderr"
             fi
         fi