sync update from LAN repo (source master @ c21a2a4a)

This commit is contained in:
gozeros committed 2026-11-10 14:03:50 +08:00
1 parent ea9dcff9c0
commit 66a443e915
140 files changed
+13777 -1040

No files matched your search

@@ -136,13 +136,18 @@ def _build_runtime_middlewares(
# wrap_tool_call chain: its handler is the raw tool, so it observes the
# unconverted exception (before ToolErrorHandlingMiddleware turns it into
# an error ToolMessage) and the raw return value.
# Omit the recorder entirely while paused: no record allocation, skill-merge
# lookup, or fire-and-forget DB task is created by any tool invocation.
# Keep tool execution/error handling and LLM metrics independent of this flag.
if app_config.runtime_performance.tool_metrics_enabled:
from deerflow.agents.middlewares.tool_metrics_middleware import ToolMetricsMiddleware
middlewares.append(ToolMetricsMiddleware())
# Full auditing can stay paused under high concurrency while the low-volume
# skill-only recorder remains active. In skill-only mode non-skill calls are
# passed through without a merge query or fire-and-forget DB write.
performance = app_config.runtime_performance
if performance.tool_metrics_enabled or performance.skill_metrics_enabled:
from deerflow.agents.middlewares.tool_metrics_middleware import ToolMetricsMiddleware
middlewares.append(
ToolMetricsMiddleware(
skill_only=not performance.tool_metrics_enabled,
)
)
return middlewares
@@ -179,20 +184,20 @@ def build_subagent_runtime_middlewares(
model_name = app_config.models[0].name
model_config = app_config.get_model_config(model_name) if model_name else None
if model_config is not None and model_config.supports_vision:
from deerflow.agents.middlewares.view_image_middleware import ViewImageMiddleware
middlewares.append(ViewImageMiddleware())
# Must remain the innermost model-call wrapper: subagent-specific context,
# vision payloads, system prompts and tool schemas all need to be included
# in the final input-size decision.
from deerflow.agents.middlewares.context_window_guard_middleware import ContextWindowGuardMiddleware
middlewares.append(
ContextWindowGuardMiddleware(
max_input_tokens=model_config.max_input_tokens if model_config is not None else None,
)
)
return middlewares
if model_config is not None and model_config.supports_vision:
from deerflow.agents.middlewares.view_image_middleware import ViewImageMiddleware
middlewares.append(ViewImageMiddleware())
# Must remain the innermost model-call wrapper: subagent-specific context,
# vision payloads, system prompts and tool schemas all need to be included
# in the final input-size decision.
from deerflow.agents.middlewares.context_window_guard_middleware import ContextWindowGuardMiddleware
middlewares.append(
ContextWindowGuardMiddleware(
max_input_tokens=model_config.max_input_tokens if model_config is not None else None,
)
)
return middlewares
@@ -1,6 +1,7 @@
"""Middleware that records per-tool-call metrics for the admin audit page.
Every tool / skill invocation ends with one row in ``tool_call_metrics``:
While the administrator-controlled call-metrics switch is enabled, every tool
/ skill invocation ends with one row in ``tool_call_metrics``:
- Successes record the wall-clock duration.
- Failures record a coarse ``error_category`` plus a one-line message.
@@ -40,6 +41,7 @@ from langgraph.errors import GraphBubbleUp
from langgraph.prebuilt.tool_node import ToolCallRequest
from langgraph.types import Command
from deerflow.config.system_settings import get_call_metrics_settings
from deerflow.persistence.engine import fire_and_forget_db_write, get_session_factory
from deerflow.persistence.tool_metrics import ToolMetricRecord, make_tool_metrics_store
from deerflow.persistence.tool_metrics.base import ToolMetricsStore
@@ -188,11 +190,18 @@ def _evaluate_result(result: Any, tool_name: str | None) -> tuple[str, str | Non
class ToolMetricsMiddleware(AgentMiddleware[AgentState]):
"""Wrap each tool call to time it and persist a metric row."""
"""Time tool calls, optionally persisting only attributable skill calls."""
def __init__(self, *, store: ToolMetricsStore | None = None, **kwargs: Any) -> None:
def __init__(
self,
*,
store: ToolMetricsStore | None = None,
skill_only: bool = False,
**kwargs: Any,
) -> None:
super().__init__(**kwargs)
self._store_override = store
self._skill_only = skill_only
# -- store / context ----------------------------------------------------
@@ -205,6 +214,12 @@ class ToolMetricsMiddleware(AgentMiddleware[AgentState]):
return None
return make_tool_metrics_store(sf)
@staticmethod
def _metrics_enabled() -> bool:
"""Read the administrator switch at call time so no restart is needed."""
return bool(get_call_metrics_settings().enabled)
@staticmethod
def _runtime_lookup(request: ToolCallRequest, keys: tuple[str, ...]) -> str | None:
runtime = getattr(request, "runtime", None)
@@ -280,7 +295,11 @@ class ToolMetricsMiddleware(AgentMiddleware[AgentState]):
handler: Callable[[ToolCallRequest], ToolMessage | Command],
) -> ToolMessage | Command:
record = self._build_base_record(request)
if is_internal_tool_name(record.tool_name):
if (
not self._metrics_enabled()
or is_internal_tool_name(record.tool_name)
or (self._skill_only and not record.skill_name)
):
return handler(request)
started = time.monotonic()
try:
@@ -303,7 +322,11 @@ class ToolMetricsMiddleware(AgentMiddleware[AgentState]):
handler: Callable[[ToolCallRequest], Awaitable[ToolMessage | Command]],
) -> ToolMessage | Command:
record = self._build_base_record(request)
if is_internal_tool_name(record.tool_name):
if (
not self._metrics_enabled()
or is_internal_tool_name(record.tool_name)
or (self._skill_only and not record.skill_name)
):
return await handler(request)
started = time.monotonic()
try: