Merge remote-tracking branch 'origin/main' into codex/p0-obs-002-20260715

apps/api/src/main.py

@@ -64,8 +64,8 @@ from src.api.v1 import incidents as incidents_v1  # Phase 6.4: Decision Proposal
 from src.api.v1 import iwooos as iwooos_v1  # IwoooS security governance API
 from src.api.v1 import knowledge as knowledge_v1  # KB Phase 1: Knowledge Base
 from src.api.v1 import learning as learning_v1  # Phase D-G P0: Learning API
-from src.api.v1 import metrics as metrics_v1  # Phase 7: Gold Metrics (真實血脈)
 from src.api.v1 import mcp_control_plane as mcp_control_plane_v1
+from src.api.v1 import metrics as metrics_v1  # Phase 7: Gold Metrics (真實血脈)
 from src.api.v1 import monitoring as monitoring_v1  # 2026-04-03: 監控工具狀態
 from src.api.v1 import notifications as notifications_v1  # 2026-04-10: 通知頻道狀態
 from src.api.v1 import (
@@ -650,22 +650,24 @@ async def lifespan(_app: FastAPI) -> AsyncGenerator[None, None]:
         logger.warning("approval_timeout_resolver_schedule_failed", error=str(e))
 
     # T73: 已有完成證據但仍卡在 INVESTIGATING 的舊 incident 小批次收斂。
-    # 僅處理 auto-repair success / approval EXECUTION_SUCCESS / approval EXPIRED，
-    # 不自動關閉 manual_required 或單純 APPROVED 事件。
+    # API production pool 只有 serving 預算；實際 loop 由單一 standalone
+    # signal worker 擁有，避免 coroutine 被 budget gate 關閉後仍假性記錄
+    # scheduled。reconciler 本身只處理強證據或已不再 firing 的事件。
     try:
         from src.jobs.incident_lifecycle_reconciler import (
             INTERVAL_SECONDS as INCIDENT_LIFECYCLE_RECONCILER_INTERVAL,
         )
-        from src.jobs.incident_lifecycle_reconciler import (
-            run_incident_lifecycle_reconciler_loop,
-        )
-        schedule_api_background_task(run_incident_lifecycle_reconciler_loop())
         logger.info(
-            "incident_lifecycle_reconciler_scheduled",
+            "incident_lifecycle_reconciler_owner_delegated",
             interval_sec=INCIDENT_LIFECYCLE_RECONCILER_INTERVAL,
+            scheduled_in_api_process=False,
+            production_process_owner="awoooi-worker",
         )
     except Exception as e:
-        logger.warning("incident_lifecycle_reconciler_schedule_failed", error=str(e))
+        logger.warning(
+            "incident_lifecycle_reconciler_owner_resolution_failed",
+            error=str(e),
+        )
 
     # AwoooP Ansible candidate backfill worker.
     # 把近期已命中 allowlisted PlayBook、但缺 durable candidate row 的事故補進

apps/api/src/services/mcp_version_lifecycle_service.py

@@ -42,7 +42,7 @@ _SOURCE_HOSTS = {
 }
 _VERSION_PATTERNS = (
     re.compile(
-        rb"""(?ix)["'](?:version|latestVersion|packageVersion)["']\s*[:=]\s*["']v?"""
+        rb"""(?ix)["'](?:latestVersion|packageVersion|softwareVersion)["']\s*[:=]\s*["']v?"""
         rb"""([0-9]+\.[0-9]+\.[0-9]+(?:[-+][0-9A-Za-z.-]+)?)["']"""
     ),
     re.compile(

apps/api/src/services/telegram_alert_ai_automation_matrix.py

@@ -703,13 +703,14 @@ def _inspect_source_contract(repo_root: Path) -> dict[str, Any]:
 
 
 def _read_source_contract_text(repo_root: Path, relative_path: str) -> str:
-    candidates = [repo_root / relative_path]
+    source_path = repo_root / relative_path
+    candidates = [source_path]
     api_prefix = "apps/api/"
     if relative_path.startswith(api_prefix):
         candidates.append(repo_root / relative_path.removeprefix(api_prefix))
     if (
         relative_path == _AWOOOP_VERIFIED_CONTEXT_SURFACE_SOURCE
-        and not (repo_root / "apps").exists()
+        and not source_path.exists()
     ):
         candidates.append(repo_root / "src/services/telegram_alert_ai_automation_matrix.py")
 

apps/api/src/workers/signal_worker.py

@@ -623,7 +623,35 @@ async def _main() -> None:
     backup_restore_legacy_backfill_task: asyncio.Task[Any] | None = None
     security_maintenance_tasks: list[asyncio.Task[Any]] = []
     agent99_controlled_dispatch_reconciler_task: asyncio.Task[Any] | None = None
+    incident_lifecycle_reconciler_task: asyncio.Task[Any] | None = None
     mcp_version_lifecycle_task: asyncio.Task[Any] | None = None
+
+    # Production API pods reserve their DB pool for serving requests. This
+    # singleton worker therefore owns the existing bounded, fail-closed
+    # lifecycle reconciler. It resolves at most one configured batch per pass
+    # and independently reads current Prometheus firing state before treating
+    # inactive alerts as closure candidates.
+    from src.jobs.incident_lifecycle_reconciler import (
+        BATCH_LIMIT as INCIDENT_LIFECYCLE_RECONCILER_BATCH_LIMIT,
+    )
+    from src.jobs.incident_lifecycle_reconciler import (
+        INTERVAL_SECONDS as INCIDENT_LIFECYCLE_RECONCILER_INTERVAL,
+    )
+    from src.jobs.incident_lifecycle_reconciler import (
+        run_incident_lifecycle_reconciler_loop,
+    )
+
+    incident_lifecycle_reconciler_task = asyncio.create_task(
+        run_incident_lifecycle_reconciler_loop(),
+        name="run_incident_lifecycle_reconciler_loop",
+    )
+    logger.info(
+        "signal_worker_incident_lifecycle_reconciler_started",
+        owner="signal_worker",
+        interval_seconds=INCIDENT_LIFECYCLE_RECONCILER_INTERVAL,
+        batch_limit=INCIDENT_LIFECYCLE_RECONCILER_BATCH_LIMIT,
+        direct_incident_closure_performed=False,
+    )
     if settings.ENABLE_AWOOOP_ANSIBLE_CANDIDATE_BACKFILL_WORKER:
         from src.jobs.awooop_ansible_candidate_backfill_job import (
             run_awooop_ansible_candidate_backfill_loop,
@@ -794,6 +822,12 @@ async def _main() -> None:
             await agent99_controlled_dispatch_reconciler_task
         except asyncio.CancelledError:
             pass
+    if incident_lifecycle_reconciler_task is not None:
+        incident_lifecycle_reconciler_task.cancel()
+        try:
+            await incident_lifecycle_reconciler_task
+        except asyncio.CancelledError:
+            pass
     if mcp_version_lifecycle_task is not None:
         mcp_version_lifecycle_task.cancel()
         try: