fix(aiops): 打通 AI 自主學習鏈 — verifier 改 await + aol 動作回灌

統帥 2026-04-19 全景審計發現:
  - automation_operation_log: 22 筆 (全部 drift_narrator),33 件/7d approval 動作 0 筆回灌
  - incident_evidence.verification_result: 1212 筆 100% NULL,verifier 從未寫入
  - 根因: _run_post_execution_verify 用 asyncio.create_task fire-and-forget,
          Pod recycle 時 task 被殺,verification_result 永遠寫不進去

修復 (打通 verifier→learning→Playbook EWMA→finetune 全鏈):

approval_execution.py:
  + _log_aol_started: 主流程開始時 INSERT aol(playbook_executed, pending)
  + _log_aol_completed: 4 個 return 點 UPDATE aol 為 success/failed + duration + stderr
    └ NO_ACTION / parse_fail / K8s 成功 / K8s 失敗 全部留痕
  ~ _run_post_execution_verify 兩處 (成功+失敗 path) 從 create_task 改 await + 60s timeout
  + 失敗時 stderr_feed_back 寫入 result.error → 解開 E6 stderr 回灌閉環

declarative_remediation.py:
  ~ _log_remediation_event task 加 named + add_done_callback,task 失敗時有 log
    (原 fire-and-forget 0 筆寫入,現在可診斷為何 task 死掉)

預期效果:
  - aol playbook_executed 即時可見 (33 件/7d 立刻有資料)
  - incident_evidence.verification_result 開始累積 → finetune_exporter 7d cron 終於有料
  - Playbook EWMA trust_score 開始動態變化
  - stderr_feed_back 接通 → 失敗訊號回灌 retry/Playbook 負向強化

不影響:
  - background_task 跑在背景,+60s 延遲不阻塞 API
  - aol 寫入失敗只 logger.warning,不阻塞執行主流程

Refs: MASTER §3.1 L6×D1 (ADR-081 PostExecutionVerifier),
      MASTER §3.4 D4 (ADR-083 學習閉環),
      ADR-090 監控盲區治理 (2026-04-18 全景審計)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

apps/api/src/services/approval_execution.py

@@ -24,6 +24,7 @@ Approval Execution Service - Phase 16 R4.2 瘦身 Router 抽取
 """
 
 import asyncio
+import time
 from typing import TYPE_CHECKING
 
 import structlog
@@ -39,6 +40,11 @@ if TYPE_CHECKING:
 
 logger = structlog.get_logger(__name__)
 
+# ADR-090 § 自動化動作回灌 (2026-04-19 ogt + Claude Opus 4.7 亞太):
+# PostExecutionVerifier 從 fire-and-forget 改 await,確保 verification_result 必寫入 incident_evidence.
+# 上限 60s 涵蓋 verifier warmup(10s) + collect(30s) + 緩衝 20s.
+_VERIFIER_AWAIT_TIMEOUT_SEC = 60.0
+
 
 class ApprovalExecutionService:
     """
@@ -134,6 +140,11 @@ class ApprovalExecutionService:
             action=approval.action,
         )
 
+        # ADR-090 § 自動化動作回灌 (2026-04-19): 主流程開始即在 aol 留痕,
+        # 結束時 update。不依賴 fire-and-forget,確保 33 件/7d approval 全部可觀測。
+        _aol_op_id = await self._log_aol_started(approval)
+        _aol_started_ms = time.time()
+
         service = get_approval_service()
         timeline = get_timeline_service()
 
@@ -181,6 +192,13 @@ class ApprovalExecutionService:
                         approval, success=True, error=None,
                     )
                 )
+                # ADR-090 § aol completed (NO_ACTION 視為成功)
+                await self._log_aol_completed(
+                    op_id=_aol_op_id,
+                    status="success",
+                    duration_ms=int((time.time() - _aol_started_ms) * 1000),
+                    output={"reason": "NO_ACTION", "action": approval.action[:200]},
+                )
                 return True  # NO_ACTION 視為成功完成
 
             # 真解析失敗 (非 NO_ACTION)
@@ -215,6 +233,13 @@ class ApprovalExecutionService:
                     error_message="Could not parse operation type",
                 )
             )
+            # ADR-090 § aol completed (parse 失敗)
+            await self._log_aol_completed(
+                op_id=_aol_op_id,
+                status="failed",
+                duration_ms=int((time.time() - _aol_started_ms) * 1000),
+                error=f"parse_fail: {approval.action[:300]}",
+            )
             return False  # 解析失敗 → 執行未發生
 
         # ADR-076 Task 3: 執行失敗重試機制
@@ -337,16 +362,25 @@ class ApprovalExecutionService:
                     timeout_sec=30.0,
                 )
 
-            # ADR-081 Phase 1: 執行後驗證 (fire-and-forget)
-            # PostExecutionVerifier 等待 K8s 收斂後抓取後狀態，補填 EvidenceSnapshot
+            # ADR-081 Phase 1 + ADR-090 修復 (2026-04-19 ogt + Claude Opus 4.7):
+            # PostExecutionVerifier 改 await + 60s timeout,確保 verification_result 必寫入。
+            # 之前 fire-and-forget 在 Pod recycle 時 task 被殺,導致 1212 筆 evidence 全 NULL.
             from src.core.feature_flags import aiops_flags
             if aiops_flags.is_sub_flag_enabled("AIOPS_P1_POST_EXECUTION_VERIFIER"):
-                asyncio.create_task(
-                    self._run_post_execution_verify(
-                        approval=approval,
-                        action_taken=f"{operation_type.value}:{resource_name}",
+                try:
+                    await asyncio.wait_for(
+                        self._run_post_execution_verify(
+                            approval=approval,
+                            action_taken=f"{operation_type.value}:{resource_name}",
+                        ),
+                        timeout=_VERIFIER_AWAIT_TIMEOUT_SEC,
+                    )
+                except asyncio.TimeoutError:
+                    logger.warning(
+                        "post_verify_timeout_exceeded",
+                        approval_id=str(approval.id),
+                        timeout_sec=_VERIFIER_AWAIT_TIMEOUT_SEC,
                     )
-                )
 
             # 2026-04-07 Claude Code: Sprint 4 B3 — 記錄人工批准處置類型
             try:
@@ -373,6 +407,19 @@ class ApprovalExecutionService:
                 except Exception as _resolve_e:
                     logger.warning("incident_resolve_after_execution_failed", error=str(_resolve_e))
 
+            # ADR-090 § aol completed (執行成功)
+            await self._log_aol_completed(
+                op_id=_aol_op_id,
+                status="success",
+                duration_ms=int((time.time() - _aol_started_ms) * 1000),
+                output={
+                    "operation_type": operation_type.value,
+                    "resource_name": resource_name,
+                    "namespace": namespace,
+                    "executor_duration_ms": result.duration_ms,
+                    "total_attempts": total_attempts,
+                },
+            )
             return True  # K8s 執行成功
 
         else:
@@ -438,18 +485,41 @@ class ApprovalExecutionService:
                     timeout_sec=30.0,
                 )
 
-            # 2026-04-18 ogt + Claude Opus 4.7: ADR-090 L6 斷鏈修復 — P0.3
-            # 失敗時也跑 verifier,把 verification_result='failed' 回寫 evidence
-            # 之前 988 筆 evidence 的 verification_result 全 NULL,因 verifier 只在 success 時跑
+            # ADR-090 修復 (2026-04-19 ogt + Claude Opus 4.7):
+            # 失敗時也跑 verifier,把 verification_result='failed' 回寫 evidence。
+            # 改 await + 60s timeout (原為 fire-and-forget,task 在 Pod recycle 時被殺)。
             from src.core.feature_flags import aiops_flags
             if aiops_flags.is_sub_flag_enabled("AIOPS_P1_POST_EXECUTION_VERIFIER"):
-                asyncio.create_task(
-                    self._run_post_execution_verify(
-                        approval=approval,
-                        action_taken=f"{operation_type.value}:{resource_name}:FAILED",
+                try:
+                    await asyncio.wait_for(
+                        self._run_post_execution_verify(
+                            approval=approval,
+                            action_taken=f"{operation_type.value}:{resource_name}:FAILED",
+                        ),
+                        timeout=_VERIFIER_AWAIT_TIMEOUT_SEC,
+                    )
+                except asyncio.TimeoutError:
+                    logger.warning(
+                        "post_verify_timeout_exceeded_failed_path",
+                        approval_id=str(approval.id),
+                        timeout_sec=_VERIFIER_AWAIT_TIMEOUT_SEC,
                     )
-                )
 
+            # ADR-090 § aol completed (執行失敗)
+            await self._log_aol_completed(
+                op_id=_aol_op_id,
+                status="failed",
+                duration_ms=int((time.time() - _aol_started_ms) * 1000),
+                output={
+                    "operation_type": operation_type.value,
+                    "resource_name": resource_name,
+                    "namespace": namespace,
+                    "executor_duration_ms": result.duration_ms,
+                    "total_attempts": total_attempts,
+                },
+                error=result.error,
+                stderr=result.error,  # E6 stderr 回灌 — 給 retry/Playbook 負向強化用
+            )
             return False  # K8s 執行失敗
 
     async def _push_execution_result_to_alert(
@@ -1014,6 +1084,107 @@ class ApprovalExecutionService:
         return None
 
 
+    # =========================================================================
+    # ADR-090 § AOL Writer (2026-04-19 ogt + Claude Opus 4.7 亞太)
+    # 把 approval execution 的生命週期回灌 automation_operation_log.
+    # 之前 33 件/7d approval 動作完全沒寫入 aol,只有 drift_narrator 的
+    # 22 筆 notification_formatted。修復後每次執行都留痕。
+    # =========================================================================
+
+    async def _log_aol_started(self, approval: ApprovalRequest) -> str | None:
+        """
+        在 automation_operation_log 寫一筆 'pending' 紀錄,回傳 op_id 供 _log_aol_completed 更新。
+
+        失敗時 (DB 異常) 回 None,主流程繼續 — aol 寫入永不阻塞執行。
+        """
+        try:
+            from sqlalchemy import text as _sql
+            from src.db.base import get_db_context
+            import json as _json
+
+            input_payload = {
+                "approval_id": str(approval.id),
+                "incident_id": approval.incident_id or "",
+                "action": (approval.action or "")[:500],
+                "risk_level": getattr(approval, "risk_level", None) or "",
+                "requested_by": getattr(approval, "requested_by", "") or "",
+            }
+
+            async with get_db_context() as db:
+                row = await db.execute(
+                    _sql("""
+                        INSERT INTO automation_operation_log (
+                            operation_type, actor, status,
+                            input, output, tags
+                        ) VALUES (
+                            'playbook_executed',
+                            'approval_execution',
+                            'pending',
+                            CAST(:input AS jsonb),
+                            '{}'::jsonb,
+                            :tags
+                        )
+                        RETURNING op_id
+                    """),
+                    {
+                        "input": _json.dumps(input_payload, ensure_ascii=False),
+                        "tags": ["approval", "execution", "playbook"],
+                    },
+                )
+                op_id = row.scalar()
+                return str(op_id) if op_id else None
+        except Exception as e:
+            logger.warning("aol_started_write_failed", approval_id=str(approval.id), error=str(e))
+            return None
+
+    async def _log_aol_completed(
+        self,
+        op_id: str | None,
+        status: str,
+        duration_ms: int,
+        output: dict | None = None,
+        error: str | None = None,
+        stderr: str | None = None,
+    ) -> None:
+        """
+        UPDATE automation_operation_log 為 success/failed 並寫入結果摘要 + stderr。
+
+        status 必須是 aol constraint 允許的值:
+        pending | success | failed | dry_run | rolled_back
+
+        op_id 為 None 時靜默跳過 (started 寫入失敗時不應觸發 update 例外)。
+        """
+        if not op_id:
+            return
+        try:
+            from sqlalchemy import text as _sql
+            from src.db.base import get_db_context
+            import json as _json
+
+            async with get_db_context() as db:
+                await db.execute(
+                    _sql("""
+                        UPDATE automation_operation_log
+                        SET status = :status,
+                            duration_ms = :duration_ms,
+                            output = CAST(:output AS jsonb),
+                            error = :error,
+                            stderr_feed_back = :stderr
+                        WHERE op_id = CAST(:op_id AS uuid)
+                    """),
+                    {
+                        "status": status,
+                        "duration_ms": duration_ms,
+                        "output": _json.dumps(output or {}, ensure_ascii=False),
+                        "error": (error or "")[:2000] if error else None,
+                        "stderr": (stderr or "")[:8000] if stderr else None,
+                        "op_id": op_id,
+                    },
+                )
+        except Exception as e:
+            logger.warning("aol_completed_write_failed", op_id=op_id, error=str(e))
+
+
 # =============================================================================
 # Singleton Instance
 # =============================================================================

apps/api/src/services/declarative_remediation.py

@@ -168,13 +168,30 @@ class DeclarativeRemediation:
         )
 
         # 2026-04-18 ADR-090-D: 寫入 remediation_events 表(MASTER §7.1 #6 KPI 資料源)
-        # fire-and-forget,不阻塞主流程
+        # 2026-04-19 ogt + Claude Opus 4.7 修復: 原 fire-and-forget 0 筆寫入。
+        # evaluate() 是同步函式,無法直接 await — 改用 named task + done_callback,
+        # 確保 task 失敗時有 log,後續可診斷為何 0 筆。
         try:
             import asyncio as _a
-            _a.create_task(_log_remediation_event(spec, action, target, namespace))
+            _task = _a.create_task(
+                _log_remediation_event(spec, action, target, namespace),
+                name=f"log_remediation:{action[:30]}",
+            )
+
+            def _on_done(t: _a.Task) -> None:
+                if t.cancelled():
+                    logger.warning("log_remediation_event_cancelled", action=action[:80])
+                elif t.exception():
+                    logger.warning(
+                        "log_remediation_event_failed",
+                        action=action[:80],
+                        error=str(t.exception()),
+                    )
+
+            _task.add_done_callback(_on_done)
         except RuntimeError:
             # 非 async context (正規呼叫都是 async),靜默跳過
-            pass
+            logger.debug("log_remediation_event_no_event_loop", action=action[:80])
 
         return spec