diff --git a/zhenxun/builtin_plugins/hooks/auth_event_selector.py b/zhenxun/builtin_plugins/hooks/auth_event_selector.py index 5ecaefd1..0b74f8a6 100644 --- a/zhenxun/builtin_plugins/hooks/auth_event_selector.py +++ b/zhenxun/builtin_plugins/hooks/auth_event_selector.py @@ -39,14 +39,8 @@ class HandleEventSelectorDependencies: _HANDLE_EVENT_PATCHED = False _ORIGINAL_HANDLE_EVENT: Callable[..., Awaitable[None]] | None = None _ORIGINAL_ADAPTER_HANDLE_EVENTS: dict[object, object] = {} -_MATCHER_DEADLINE_BY_LANE = { - "system": 15.0, - "command": 12.0, - "temp": 15.0, - "passive_light": 3.0, - "fallback_ai": 5.0, -} -_DEFAULT_MATCHER_DEADLINE = 5.0 +_DEFAULT_MATCHER_DEADLINE = 21600.0 +_MATCHER_DEADLINE_BY_LANE: dict[str, float] = {} def _matcher_deadline_for_lane(lane: str) -> float: @@ -76,7 +70,6 @@ async def _run_matcher_with_deadline( with anyio_mod.fail_after(timeout): await coro except TimeoutError: - signal_overload(20.0) logger.warning( "matcher dispatch timeout: " f"matcher={_matcher_name(matcher)}, lane={lane}, timeout={timeout:.1f}s", diff --git a/zhenxun/builtin_plugins/hooks/auth_runtime_config.py b/zhenxun/builtin_plugins/hooks/auth_runtime_config.py index 383a8d07..535aa76c 100644 --- a/zhenxun/builtin_plugins/hooks/auth_runtime_config.py +++ b/zhenxun/builtin_plugins/hooks/auth_runtime_config.py @@ -2,6 +2,7 @@ from __future__ import annotations from dataclasses import dataclass, fields import os +from urllib.parse import urlparse @dataclass(frozen=True, slots=True) @@ -101,6 +102,22 @@ def _warn_invalid_env(env_name: str, raw: str, exc: Exception) -> None: return +def _default_passive_db_limit() -> int: + try: + from zhenxun.configs.config import BotConfig + + scheme = urlparse(BotConfig.db_url or "").scheme.lower() + except Exception: + scheme = "" + if scheme == "sqlite": + return 1 + if scheme.startswith("postgres"): + return 6 + if scheme == "mysql": + return 4 + return 2 + + def _load_config(cls: type, prefix: str): values = {} default_obj = cls() @@ -115,6 +132,9 @@ def _load_config(cls: type, prefix: str): raw = candidate_value break if raw is None or not raw.strip(): + if cls is AuthDispatchRuntimeConfig and item.name == "passive_db_limit": + values[item.name] = _default_passive_db_limit() + continue values[item.name] = default continue try: diff --git a/zhenxun/services/db_context/config.py b/zhenxun/services/db_context/config.py index 38a19b41..62dc3bd3 100644 --- a/zhenxun/services/db_context/config.py +++ b/zhenxun/services/db_context/config.py @@ -27,9 +27,9 @@ MYSQL_CONFIG = { SQLITE_CONFIG = { "journal_mode": "WAL", # 提高并发写入性能 - # SQLite 的底层锁等待应接近业务超时，避免上层放弃后 worker 仍长时间占线。 + # SQLite 底层锁等待不应超过上层 DB_TIMEOUT_SECONDS=3.0。 # Windows bind mount / Docker 场景下 SQLite 不适合高写并发。 - "busy_timeout": 5000, + "busy_timeout": 3000, "foreign_keys": "ON", } diff --git a/zhenxun/services/db_context/utils.py b/zhenxun/services/db_context/utils.py index 8873f188..a0799442 100644 --- a/zhenxun/services/db_context/utils.py +++ b/zhenxun/services/db_context/utils.py @@ -15,6 +15,12 @@ _SQLITE_STALL_UNTIL = 0.0 _SQLITE_STALL_REASON = "" _DB_UNHEALTHY_TIMEOUT_SECONDS = 30.0 _SQLITE_STALL_TIMEOUT_SECONDS = 60.0 +_SQLITE_LOCK_PATTERNS = ( + "database is locked", + "database is busy", + "database table is locked", + "database table is busy", +) def _is_sqlite_connection() -> bool: @@ -46,6 +52,24 @@ def sqlite_stall_reason() -> str: return _SQLITE_STALL_REASON +def _is_sqlite_lock_error(exc: BaseException) -> bool: + if not _is_sqlite_connection(): + return False + message = str(exc).casefold() + return any(pattern in message for pattern in _SQLITE_LOCK_PATTERNS) + + +def _mark_sqlite_lock_unhealthy(exc: BaseException, operation: str | None) -> None: + reason = f"{operation or 'database_operation'} sqlite lock: {exc}" + _mark_sqlite_stall(reason, _SQLITE_STALL_TIMEOUT_SECONDS) + signal_db_unhealthy(_SQLITE_STALL_TIMEOUT_SECONDS, reason=reason) + logger.warning( + "SQLite 数据库锁等待失败，已暂停低优先级数据库任务", + LOG_COMMAND, + e=exc if isinstance(exc, Exception) else None, + ) + + async def with_db_timeout( coro, timeout: float = DB_TIMEOUT_SECONDS, @@ -78,3 +102,7 @@ async def with_db_timeout( LOG_COMMAND, ) raise + except Exception as exc: + if _is_sqlite_lock_error(exc): + _mark_sqlite_lock_unhealthy(exc, operation) + raise diff --git a/zhenxun/services/low_priority_writer.py b/zhenxun/services/low_priority_writer.py index 83286168..88dbf38f 100644 --- a/zhenxun/services/low_priority_writer.py +++ b/zhenxun/services/low_priority_writer.py @@ -8,7 +8,7 @@ import time from typing import Any from zhenxun.services.log import logger -from zhenxun.services.message_load import should_pause_tasks, signal_db_unhealthy +from zhenxun.services.message_load import should_pause_db_tasks, signal_db_unhealthy from zhenxun.utils.manager.priority_manager import PriorityLifecycle LOG_COMMAND = "LowPriorityWriter" @@ -159,7 +159,7 @@ async def _worker_loop() -> None: except asyncio.TimeoutError: pass event.clear() - if should_pause_tasks(): + if should_pause_db_tasks(): continue async with _FLUSH_LOCK: for state in list(_WRITERS.values()): @@ -186,7 +186,7 @@ async def _flush_state( force: bool = False, ) -> int: if not force: - if should_pause_tasks(): + if should_pause_db_tasks(): return 0 if time.monotonic() < state.backoff_until: return 0 diff --git a/zhenxun/services/message_load.py b/zhenxun/services/message_load.py index 3a320110..7391f2e0 100644 --- a/zhenxun/services/message_load.py +++ b/zhenxun/services/message_load.py @@ -57,3 +57,7 @@ def db_unhealthy_reason() -> str: def should_pause_tasks() -> bool: return is_overloaded() or is_db_unhealthy() + + +def should_pause_db_tasks() -> bool: + return is_db_unhealthy()