feat: 添加内存诊断工具

pyproject.toml

@@ -29,6 +29,7 @@ dependencies = [
     "pandas>=2.3.1",
     "pillow>=11.3.0",
     "playwright>=1.54.0",
+    "psutil>=6.0.0",
     "pyarrow>=20.0.0",
     "pydantic>=2.11.7",
     "pypinyin>=0.54.0",

requirements.txt

@@ -18,6 +18,7 @@ openai>=1.95.0
 pandas>=2.3.1
 pillow>=11.3.0
 playwright>=1.54.0
+psutil>=6.0.0
 pyarrow>=20.0.0
 pydantic>=2.11.7
 pypinyin>=0.54.0

src/config/config.py

@@ -59,7 +59,7 @@
 LEGACY_ENV_PATH: Path = (PROJECT_ROOT / ".env").resolve().absolute()
 A_MEMORIX_LEGACY_CONFIG_PATH: Path = (CONFIG_DIR / "a_memorix.toml").resolve().absolute()
 MMC_VERSION: str = "1.0.0-pre.21"
-CONFIG_VERSION: str = "8.12.10"
+CONFIG_VERSION: str = "8.12.11"
 MODEL_CONFIG_VERSION: str = "1.17.0"
 
 logger = get_logger("config")
@@ -645,7 +645,8 @@ def write_config_to_file(
 
     if isinstance(config, Config):
         try:
-            a_memorix_web = full_config_data["a_memorix"]["web"]
+            a_memorix_table = cast(Any, full_config_data["a_memorix"])
+            a_memorix_web = cast(Any, a_memorix_table["web"])
             if "import_config" in a_memorix_web and "import" not in a_memorix_web:
                 a_memorix_web["import"] = a_memorix_web.pop("import_config")
         except Exception:

src/config/official_configs.py

@@ -3022,6 +3022,124 @@ class DebugConfig(ConfigBase):
     )
     """是否记录 LLM prompt cache 调试统计，默认关闭"""
 
+    enable_memory_diagnostics: bool = Field(
+        default=False,
+        json_schema_extra={
+            "x-widget": "switch",
+            "x-icon": "memory-stick",
+            "advanced": True,
+        },
+    )
+    """是否启用长时间运行内存诊断，默认关闭"""
+
+    memory_diagnostics_interval_seconds: int = Field(
+        default=300,
+        ge=10,
+        json_schema_extra={
+            "x-widget": "number",
+            "x-icon": "timer",
+            "advanced": True,
+        },
+    )
+    """内存诊断采样间隔，单位秒"""
+
+    memory_diagnostics_top_sessions: int = Field(
+        default=20,
+        ge=1,
+        json_schema_extra={
+            "x-widget": "number",
+            "x-icon": "list-ordered",
+            "advanced": True,
+        },
+    )
+    """内存诊断输出的高占用会话数量"""
+
+    memory_diagnostics_jsonl_path: str = Field(
+        default="logs/memory_diagnostics/memory_diagnostics.jsonl",
+        json_schema_extra={
+            "x-widget": "input",
+            "x-icon": "file-json",
+            "advanced": True,
+        },
+    )
+    """内存诊断 JSONL 输出路径"""
+
+    memory_diagnostics_enable_tracemalloc: bool = Field(
+        default=False,
+        json_schema_extra={
+            "x-widget": "switch",
+            "x-icon": "search-code",
+            "advanced": True,
+        },
+    )
+    """是否启用 tracemalloc Python 分配快照诊断"""
+
+    memory_diagnostics_snapshot_growth_mb: int = Field(
+        default=100,
+        ge=1,
+        json_schema_extra={
+            "x-widget": "number",
+            "x-icon": "trending-up",
+            "advanced": True,
+        },
+    )
+    """RSS 增长超过该阈值时输出 tracemalloc 差异，单位 MB"""
+
+    memory_diagnostics_binary_scan_message_limit: int = Field(
+        default=5000,
+        ge=0,
+        json_schema_extra={
+            "x-widget": "number",
+            "x-icon": "scan-search",
+            "advanced": True,
+        },
+    )
+    """每轮最多扫描多少条缓存消息来估算二进制滞留；此为全量 session 共享预算，非单 session 上限，0 表示不扫描"""
+
+    memory_diagnostics_jsonl_max_total_size_mb: int = Field(
+        default=50,
+        ge=1,
+        json_schema_extra={
+            "x-widget": "number",
+            "x-icon": "hard-drive",
+            "advanced": True,
+        },
+    )
+    """内存诊断 JSONL 当前文件和历史轮转文件的总大小上限，单位 MB"""
+
+    memory_diagnostics_warn_runtime_count: int = Field(
+        default=0,
+        ge=0,
+        json_schema_extra={
+            "x-widget": "number",
+            "x-icon": "message-circle-warning",
+            "advanced": True,
+        },
+    )
+    """runtime 数量超过该阈值时输出 WARNING，0 表示不告警"""
+
+    memory_diagnostics_warn_message_cache_count: int = Field(
+        default=0,
+        ge=0,
+        json_schema_extra={
+            "x-widget": "number",
+            "x-icon": "message-circle-warning",
+            "advanced": True,
+        },
+    )
+    """message_cache 总量超过该阈值时输出 WARNING，0 表示不告警"""
+
+    memory_diagnostics_warn_voice_binary_mb: int = Field(
+        default=0,
+        ge=0,
+        json_schema_extra={
+            "x-widget": "number",
+            "x-icon": "message-circle-warning",
+            "advanced": True,
+        },
+    )
+    """语音二进制估算超过该阈值时输出 WARNING，单位 MB，0 表示不告警"""
+
 
 class ExtraPromptItem(ConfigBase):
     platform: str = Field(

src/main.py

@@ -18,6 +18,7 @@
 from src.manager.async_task_manager import async_task_manager
 from src.plugin_runtime.integration import get_plugin_runtime_manager
 from src.prompt.prompt_manager import prompt_manager
+from src.services.memory_diagnostics_service import MemoryDiagnosticsTask, is_memory_diagnostics_enabled
 from src.services.memory_flow_service import memory_automation_service
 
 # from src.api.main import start_api_server
@@ -113,6 +114,9 @@ async def _init_components(self) -> None:
 
         logger.info(t("startup.chat_manager_initialized"))
         await memory_automation_service.start()
+        if is_memory_diagnostics_enabled():
+            await async_task_manager.add_task(MemoryDiagnosticsTask())
+            logger.info("内存诊断任务已启用")
 
         # await asyncio.sleep(0.5) #防止logger输出飞了