fix(awooop): make learning receipt backfill idempotent

apps/api/src/services/awooop_ansible_check_mode_service.py

@@ -528,7 +528,7 @@ def _claim_from_apply_operation_row(row: dict[str, Any]) -> tuple[AnsibleCheckMo
     input_payload = _json_loads(row.get("input"))
     output_payload = _json_loads(row.get("output"))
     dry_run_result = _json_loads(row.get("dry_run_result"))
-    catalog_id = str(input_payload.get("catalog_id") or "")
+    catalog_id = str(input_payload.get("catalog_id") or row.get("catalog_id") or "")
     if not catalog_id:
         return None
     catalog_item = get_ansible_catalog_item(catalog_id) or {}
@@ -541,12 +541,14 @@ def _claim_from_apply_operation_row(row: dict[str, Any]) -> tuple[AnsibleCheckMo
     apply_playbook_path = str(
         input_payload.get("apply_playbook_path")
         or input_payload.get("playbook_path")
+        or row.get("playbook_path")
         or catalog_item.get("playbook_path")
         or ""
     )
     check_mode_playbook_path = str(
         input_payload.get("check_mode_playbook_path")
         or input_payload.get("playbook_path")
+        or row.get("playbook_path")
         or apply_playbook_path
     )
     if not apply_playbook_path:
@@ -563,7 +565,12 @@ def _claim_from_apply_operation_row(row: dict[str, Any]) -> tuple[AnsibleCheckMo
         playbook_path=check_mode_playbook_path,
         apply_playbook_path=apply_playbook_path,
         inventory_hosts=tuple(str(host) for host in inventory_hosts),
-        risk_level=str(input_payload.get("risk_level") or catalog_item.get("risk_level") or ""),
+        risk_level=str(
+            input_payload.get("risk_level")
+            or row.get("risk_level")
+            or catalog_item.get("risk_level")
+            or ""
+        ),
         input_payload=input_payload,
     )
     result = AnsibleRunResult(
@@ -866,37 +873,54 @@ async def _record_post_apply_verifier_and_learning(
         from src.repositories.knowledge_repository import KnowledgeDBRepository
 
         async with get_db_context(project_id) as db:
-            repo = KnowledgeDBRepository(db)
-            await repo.create(
-                KnowledgeEntryCreate(
-                    title=f"AI 自動修復沉澱：{claim.incident_id}",
-                    content=(
-                        "AI Agent 已完成 Ansible controlled apply 並寫入驗證摘要。\n\n"
-                        f"- Incident: {claim.incident_id}\n"
-                        f"- Catalog: {claim.catalog_id}\n"
-                        f"- PlayBook: {claim.apply_playbook_path}\n"
-                        f"- Apply operation: {apply_op_id}\n"
-                        f"- Verification result: {verification_result}\n"
-                        f"- Return code: {result.returncode}\n"
-                        f"- Next step: {post_state['next_required_step']}\n"
-                    ),
-                    entry_type=EntryType.INCIDENT_CASE,
-                    category="AI自動化/Ansible受控修復",
-                    tags=[
-                        "ai_auto_repair",
-                        "ansible_controlled_apply",
-                        verification_result,
-                        str(claim.catalog_id or ""),
-                    ],
-                    source=EntrySource.AI_EXTRACTED,
-                    status=EntryStatus.REVIEW,
-                    related_incident_id=claim.incident_id,
-                    related_playbook_id=str(claim.catalog_id or "")[:36] or None,
-                    path_type=_post_apply_km_path_type(apply_op_id),
-                    created_by="ai_agent_ansible_worker",
-                )
+            path_type = _post_apply_km_path_type(apply_op_id)
+            existing = await db.execute(
+                text("""
+                    SELECT id
+                    FROM knowledge_entries
+                    WHERE related_incident_id = :incident_id
+                      AND path_type = :path_type
+                    LIMIT 1
+                """),
+                {
+                    "incident_id": claim.incident_id,
+                    "path_type": path_type,
+                },
             )
-            status["learning"] = True
+            if existing.scalar() is not None:
+                status["learning"] = True
+            else:
+                repo = KnowledgeDBRepository(db)
+                await repo.create(
+                    KnowledgeEntryCreate(
+                        title=f"AI 自動修復沉澱：{claim.incident_id}",
+                        content=(
+                            "AI Agent 已完成 Ansible controlled apply 並寫入驗證摘要。\n\n"
+                            f"- Incident: {claim.incident_id}\n"
+                            f"- Catalog: {claim.catalog_id}\n"
+                            f"- PlayBook: {claim.apply_playbook_path}\n"
+                            f"- Apply operation: {apply_op_id}\n"
+                            f"- Verification result: {verification_result}\n"
+                            f"- Return code: {result.returncode}\n"
+                            f"- Next step: {post_state['next_required_step']}\n"
+                        ),
+                        entry_type=EntryType.INCIDENT_CASE,
+                        category="AI自動化/Ansible受控修復",
+                        tags=[
+                            "ai_auto_repair",
+                            "ansible_controlled_apply",
+                            verification_result,
+                            str(claim.catalog_id or ""),
+                        ],
+                        source=EntrySource.AI_EXTRACTED,
+                        status=EntryStatus.REVIEW,
+                        related_incident_id=claim.incident_id,
+                        related_playbook_id=str(claim.catalog_id or "")[:36] or None,
+                        path_type=path_type,
+                        created_by="ai_agent_ansible_worker",
+                    )
+                )
+                status["learning"] = True
     except Exception as exc:
         logger.warning(
             "ansible_post_apply_learning_writeback_failed",
@@ -983,6 +1007,9 @@ async def backfill_missing_auto_repair_execution_receipts_once(
                         apply.error,
                         apply.duration_ms,
                         apply.status,
+                        apply.catalog_id,
+                        apply.playbook_path,
+                        apply.risk_level,
                         apply.created_at
                     FROM automation_operation_log apply
                     WHERE apply.operation_type = 'ansible_apply_executed'

apps/api/tests/test_awooop_truth_chain_service.py

@@ -1590,12 +1590,42 @@ def test_ansible_apply_operation_row_can_backfill_auto_repair_receipt() -> None:
     assert result.duration_ms == 456
 
 
+def test_ansible_apply_operation_row_reconstructs_from_columns_when_input_is_sparse() -> None:
+    reconstructed = _claim_from_apply_operation_row({
+        "op_id": "apply-op-2",
+        "parent_op_id": "check-op-2",
+        "incident_id": "INC-20260629-231F8E",
+        "status": "success",
+        "catalog_id": "ansible:188-momo-backup-user",
+        "playbook_path": "infra/ansible/playbooks/188-momo-backup-user.yml",
+        "risk_level": "low",
+        "input": {
+            "incident_id": "INC-20260629-231F8E",
+            "source_candidate_op_id": "candidate-op-2",
+            "check_mode_op_id": "check-op-2",
+        },
+        "output": {"returncode": 0, "stdout_tail": "ok"},
+        "dry_run_result": {"apply_executed": True},
+        "duration_ms": 7727,
+    })
+
+    assert reconstructed is not None
+    claim, result = reconstructed
+    assert claim.catalog_id == "ansible:188-momo-backup-user"
+    assert claim.inventory_hosts == ("host_188",)
+    assert claim.apply_playbook_path == "infra/ansible/playbooks/188-momo-backup-user.yml"
+    assert claim.risk_level == "low"
+    assert result.returncode == 0
+
+
 def test_ansible_apply_receipt_backfill_queries_existing_apply_rows() -> None:
     source = inspect.getsource(backfill_missing_auto_repair_execution_receipts_once)
 
     assert "operation_type = 'ansible_apply_executed'" in source
     assert "auto_repair_executions existing" in source
     assert "executed_steps::text LIKE" in source
+    assert "apply.catalog_id" in source
+    assert "apply.playbook_path" in source
 
 
 def test_ansible_auto_repair_receipt_insert_casts_asyncpg_parameters() -> None:
@@ -1628,6 +1658,19 @@ def test_ansible_learning_writeback_receipt_records_learning_service_call() -> N
     assert "stores_secret_values" in source
 
 
+def test_ansible_post_apply_km_writeback_is_idempotent_for_learning_backfill() -> None:
+    from src.services.awooop_ansible_check_mode_service import (
+        _record_post_apply_verifier_and_learning,
+    )
+
+    source = inspect.getsource(_record_post_apply_verifier_and_learning)
+
+    assert "SELECT id" in source
+    assert "FROM knowledge_entries" in source
+    assert "path_type = :path_type" in source
+    assert "KnowledgeDBRepository" in source
+
+
 def test_ansible_live_controlled_apply_sends_telegram_receipt_but_backfill_does_not() -> None:
     live_source = inspect.getsource(run_controlled_apply_for_claim)
     backfill_source = inspect.getsource(backfill_missing_auto_repair_execution_receipts_once)