fix: keep legacy alert triggers unattributed

docs/alerts.md

@@ -440,7 +440,7 @@ worker 会把 `triggered`、`skipped`、`degraded`、`failed` 写入 `alert_trig
 - 触发归属使用 `rule_id + rule_lifecycle_id`：每次创建规则都会生成不可复用的 lifecycle ID，worker 在加载规则时固定该 ID，并随触发记录一起写入。即使旧 worker 已经开始计算、期间规则被删除且 SQLite 复用了整数 ID，迟到的旧触发也不会归到新规则。`rule_id` 为空的记录计入 `unattributed_trigger_count`；引用已不存在规则或 lifecycle 不匹配的记录计入 `orphaned_trigger_count`。
 - 摘要的总数、状态分组、规则分组和孤儿计数在同一个 SQLite 读事务快照内完成，后台 worker 同时写入触发记录时不会返回“总数小于状态分组之和”的自相矛盾结果。
 
-该概览是 Issue #2281 的基础阶段，只解决全局统计与可靠归属。它不引入“论文失效”规则类型，也不把普通告警命名为失效信号；`impact`、`affected_entities` 和新的 thesis invalidation 规则语义留待后续独立 PR。启动时会为既有 SQLite 的 `alert_rules` / `alert_triggers` 增加 lifecycle 列，并只把能按原创建时间规则可靠匹配的既有触发回填到当前 lifecycle；无法可靠匹配的历史保留为 orphan。回滚代码不会自动删除新增列或历史数据，旧版本会忽略这些附加列。
+该概览是 Issue #2281 的基础阶段，只解决全局统计与可靠归属。它不引入“论文失效”规则类型，也不把普通告警命名为失效信号；`impact`、`affected_entities` 和新的 thesis invalidation 规则语义留待后续独立 PR。启动时会为既有 SQLite 的 `alert_rules` / `alert_triggers` 增加 lifecycle 列；既有规则会获得 lifecycle，但迁移前的触发记录无法证明属于哪一代规则，因此不按时间戳猜测回填，统一保留为 orphan。多实例同时首次升级时，重复加列会安全收敛。回滚代码不会自动删除新增列或历史数据，旧版本会忽略这些附加列。
 
 ## Phase 边界
 

src/storage.py

@@ -1487,27 +1487,36 @@ class DatabaseManager(metaclass=_DatabaseManagerMeta):
         trigger_columns = {
             column["name"] for column in inspector.get_columns(AlertTriggerRecord.__tablename__)
         }
+        for table, column, column_type, existing in (
+            ("alert_rules", "lifecycle_id", "VARCHAR(36)", rule_columns),
+            ("alert_triggers", "rule_lifecycle_id", "VARCHAR(36)", trigger_columns),
+        ):
+            if column in existing:
+                continue
+            for attempt in range(self._sqlite_write_retry_max + 1):
+                try:
+                    with self._engine.begin() as connection:
+                        connection.exec_driver_sql(
+                            f"ALTER TABLE {table} ADD COLUMN {column} {column_type}"
+                        )
+                    existing.add(column)
+                    break
+                except OperationalError as exc:
+                    if self._is_sqlite_duplicate_column_error(exc, column):
+                        existing.add(column)
+                        break
+                    if self._is_sqlite_locked_error(exc) and attempt < self._sqlite_write_retry_max:
+                        delay = self._sqlite_write_retry_base_delay * (2 ** attempt)
+                        if delay > 0:
+                            time.sleep(delay)
+                        continue
+                    raise
+
         with self._engine.begin() as connection:
-            if "lifecycle_id" not in rule_columns:
-                connection.exec_driver_sql(
-                    "ALTER TABLE alert_rules ADD COLUMN lifecycle_id VARCHAR(36)"
-                )
-            if "rule_lifecycle_id" not in trigger_columns:
-                connection.exec_driver_sql(
-                    "ALTER TABLE alert_triggers ADD COLUMN rule_lifecycle_id VARCHAR(36)"
-                )
             connection.exec_driver_sql(
                 "UPDATE alert_rules SET lifecycle_id = lower(hex(randomblob(16))) "
                 "WHERE lifecycle_id IS NULL OR lifecycle_id = ''"
             )
-            connection.exec_driver_sql(
-                "UPDATE alert_triggers SET rule_lifecycle_id = ("
-                "SELECT lifecycle_id FROM alert_rules "
-                "WHERE alert_rules.id = alert_triggers.rule_id "
-                "AND alert_triggers.triggered_at IS NOT NULL "
-                "AND alert_triggers.triggered_at >= alert_rules.created_at"
-                ") WHERE rule_lifecycle_id IS NULL"
-            )
             connection.exec_driver_sql(
                 "CREATE INDEX IF NOT EXISTS ix_alert_rules_lifecycle_id "
                 "ON alert_rules (lifecycle_id)"

tests/test_alert_api.py

@@ -932,6 +932,25 @@ class AlertApiTestCase(unittest.TestCase):
         self.assertEqual(by_rule_id[replacement_rule["id"]]["trigger_count"], 0)
         self.assertEqual(payload["orphaned_trigger_count"], 1)
 
+    def test_monitor_summary_keeps_lifecycle_less_legacy_trigger_orphaned(self) -> None:
+        rule = self._create_rule({"name": "current", "target": "600519"})
+        with self.db.get_session() as session:
+            session.add(
+                AlertTriggerRecord(
+                    rule_id=rule["id"],
+                    rule_lifecycle_id=None,
+                    target="600519",
+                    status="triggered",
+                )
+            )
+            session.commit()
+
+        payload = self.client.get("/api/v1/alerts/summary").json()
+
+        by_rule_id = {item["rule_id"]: item for item in payload["rules"]}
+        self.assertEqual(by_rule_id[rule["id"]]["trigger_count"], 0)
+        self.assertEqual(payload["orphaned_trigger_count"], 1)
+
     def test_monitor_summary_uses_one_read_snapshot_during_concurrent_trigger_write(self) -> None:
         rule = self._create_rule({"name": "snapshot", "target": "600519"})
         AlertRepository(self.db).create_trigger(