同步 incidents 相容欄位寫入

database/autoheal_models.py

@@ -111,16 +111,19 @@ class Incident(Base):
     task_name = Column(String(100), nullable=False)
     error_type = Column(String(50), nullable=False)
     error_message = Column(Text, nullable=False)
+    error_traceback = Column(Text)
     traceback_str = Column(Text)
     severity = Column(String(20), default='medium')
     status = Column(String(20), default='open')  # open/healing/closed/escalated
     retry_count = Column(Integer, default=0)
+    playbook_id = Column(Integer, ForeignKey('playbooks.id'), nullable=True)
     matched_playbook_id = Column(Integer, ForeignKey('playbooks.id'), nullable=True)
+    resolved_at = Column(DateTime, nullable=True)
     created_at = Column(DateTime, default=datetime.now)
     updated_at = Column(DateTime, default=datetime.now)
 
     # Relationship
-    playbook = relationship("Playbook", backref="incidents")
+    playbook = relationship("Playbook", foreign_keys=[matched_playbook_id], backref="incidents")
 
 
 class Playbook(Base):

migrations/036_normalize_incidents_dual_columns.sql

@@ -0,0 +1,85 @@
+-- =============================================================================
+-- Migration 036: normalize incidents legacy/current columns
+-- 日期: 2026-05-12 台北
+-- =============================================================================
+-- 背景:
+--   migration 013 建立 incidents.error_traceback / playbook_id / resolved_at；
+--   後續 ORM 改用 traceback_str / matched_playbook_id。migration 031 已加新欄位，
+--   但 create_all-only 舊環境可能缺 legacy 欄位，且新寫入只更新新欄位。
+--
+-- 設計:
+--   1. 保留雙欄做向下相容，不做破壞性 drop/rename。
+--   2. 補齊缺欄，並用 COALESCE 雙向回填，讓分析查詢可安全 coalesce。
+--   3. 補 playbook_id FK 與 matched_playbook_id FK；已存在則略過。
+-- =============================================================================
+
+ALTER TABLE IF EXISTS incidents
+    ADD COLUMN IF NOT EXISTS error_traceback TEXT;
+
+ALTER TABLE IF EXISTS incidents
+    ADD COLUMN IF NOT EXISTS traceback_str TEXT;
+
+ALTER TABLE IF EXISTS incidents
+    ADD COLUMN IF NOT EXISTS playbook_id INTEGER;
+
+ALTER TABLE IF EXISTS incidents
+    ADD COLUMN IF NOT EXISTS matched_playbook_id INTEGER;
+
+ALTER TABLE IF EXISTS incidents
+    ADD COLUMN IF NOT EXISTS resolved_at TIMESTAMP;
+
+ALTER TABLE IF EXISTS incidents
+    ALTER COLUMN severity TYPE VARCHAR(20);
+
+UPDATE incidents
+SET traceback_str = error_traceback
+WHERE traceback_str IS NULL
+  AND error_traceback IS NOT NULL;
+
+UPDATE incidents
+SET error_traceback = traceback_str
+WHERE error_traceback IS NULL
+  AND traceback_str IS NOT NULL;
+
+UPDATE incidents
+SET matched_playbook_id = playbook_id
+WHERE matched_playbook_id IS NULL
+  AND playbook_id IS NOT NULL;
+
+UPDATE incidents
+SET playbook_id = matched_playbook_id
+WHERE playbook_id IS NULL
+  AND matched_playbook_id IS NOT NULL;
+
+DO $$
+BEGIN
+    IF NOT EXISTS (
+        SELECT 1
+        FROM pg_constraint
+        WHERE conname = 'incidents_playbook_id_fkey'
+    ) THEN
+        ALTER TABLE incidents
+            ADD CONSTRAINT incidents_playbook_id_fkey
+            FOREIGN KEY (playbook_id)
+            REFERENCES playbooks(id);
+    END IF;
+
+    IF NOT EXISTS (
+        SELECT 1
+        FROM pg_constraint
+        WHERE conname = 'incidents_matched_playbook_id_fkey'
+    ) THEN
+        ALTER TABLE incidents
+            ADD CONSTRAINT incidents_matched_playbook_id_fkey
+            FOREIGN KEY (matched_playbook_id)
+            REFERENCES playbooks(id);
+    END IF;
+END $$;
+
+CREATE INDEX IF NOT EXISTS idx_incidents_matched_playbook_id
+    ON incidents(matched_playbook_id);
+
+DO $$
+BEGIN
+    RAISE NOTICE 'Migration 036 done: incidents legacy/current columns normalized';
+END $$;

services/auto_heal_service.py

@@ -232,6 +232,11 @@ class AutoHealService:
             record_autoheal_action(action="NO_PLAYBOOK", error_type=error_type, success=False)
             return AutoHealResult(success=False, action=None, message=msg)
 
+        playbook_id = int(playbook.get("id") or 0)
+        if playbook_id > 0:
+            context["matched_playbook_id"] = playbook_id
+            context["playbook_id"] = playbook_id
+
         if playbook["action_type"] not in _ALLOWED_ACTION_TYPES:
             msg = f"action_type '{playbook['action_type']}' is not allowed"
             self._log.warning("[AutoHeal] %s", msg)
@@ -317,6 +322,7 @@ class AutoHealService:
                 error_type=error_type,
                 error_message=str(context.get("exception") or context.get("error_message") or error_type)[:2000],
                 traceback_str=str(context.get("traceback_str") or "")[:8000] or None,
+                error_traceback=str(context.get("traceback_str") or "")[:8000] or None,
                 severity=str(context.get("severity") or "medium"),
                 status="healing",
                 retry_count=0,
@@ -569,15 +575,28 @@ class AutoHealService:
         session = None
         try:
             session = get_session()
+            status = "closed" if result.success else "escalated"
+            playbook_id = int(
+                context.get("matched_playbook_id")
+                or context.get("playbook_id")
+                or 0
+            ) or None
             session.execute(
                 text("""
                 UPDATE incidents
                    SET status = :status,
+                       matched_playbook_id = COALESCE(:playbook_id, matched_playbook_id),
+                       playbook_id = COALESCE(:playbook_id, playbook_id),
+                       resolved_at = CASE
+                           WHEN :status = 'closed' THEN COALESCE(resolved_at, NOW())
+                           ELSE resolved_at
+                       END,
                        updated_at = NOW()
                  WHERE id = :incident_id
                 """),
                 {
-                    "status": "closed" if result.success else "escalated",
+                    "status": status,
+                    "playbook_id": playbook_id,
                     "incident_id": int(incident_id),
                 },
             )

tests/test_auto_heal_safety.py

@@ -40,3 +40,53 @@ def test_auto_heal_code_fix_writes_audit(monkeypatch):
     assert result.success is True
     assert calls
     assert calls[0][2].action == "CODE_FIX"
+
+
+def test_incident_model_keeps_legacy_and_current_columns():
+    from database.manager import Base
+
+    columns = set(Base.metadata.tables["incidents"].columns.keys())
+
+    assert {
+        "error_traceback",
+        "traceback_str",
+        "playbook_id",
+        "matched_playbook_id",
+        "resolved_at",
+    } <= columns
+
+
+def test_auto_heal_status_update_backfills_dual_playbook_columns(monkeypatch):
+    from services.auto_heal_service import AutoHealResult, AutoHealService
+
+    captured = {}
+
+    class Session:
+        def execute(self, stmt, params):
+            captured["sql"] = str(stmt)
+            captured["params"] = params
+
+        def commit(self):
+            captured["committed"] = True
+
+        def rollback(self):
+            captured["rolled_back"] = True
+
+        def close(self):
+            captured["closed"] = True
+
+    monkeypatch.setattr("services.auto_heal_service.get_session", lambda: Session())
+
+    svc = AutoHealService()
+    svc._update_incident_status(
+        {"incident_id": 123, "matched_playbook_id": 77},
+        AutoHealResult(True, "ALERT_ONLY", "ok"),
+    )
+
+    assert "matched_playbook_id" in captured["sql"]
+    assert "playbook_id" in captured["sql"]
+    assert "resolved_at" in captured["sql"]
+    assert captured["params"]["status"] == "closed"
+    assert captured["params"]["playbook_id"] == 77
+    assert captured["params"]["incident_id"] == 123
+    assert captured["committed"] is True