大模型速度增加提示

This commit is contained in:
吉鑫升 committed 2026-05-14 19:12:14 +08:00
1 parent 4e764350a4
commit 3ea33e1dee
52 files changed
+4673 -176

No files matched your search

@@ -0,0 +1,25 @@
"""Persistence layer for per-LLM-call metrics.
Each completed LLM invocation (success or failure) writes a single row so
administrators can audit usage and speed. Designed to be cheap to write —
the middleware fires once per provider call and never blocks streaming.
"""
from deerflow.persistence.llm_metrics.base import LlmMetricRecord, LlmMetricsStore
from deerflow.persistence.llm_metrics.model import LlmCallMetricRow
from deerflow.persistence.llm_metrics.sql import SqlLlmMetricsStore
__all__ = [
"LlmCallMetricRow",
"LlmMetricRecord",
"LlmMetricsStore",
"SqlLlmMetricsStore",
"make_llm_metrics_store",
]
def make_llm_metrics_store(session_factory) -> LlmMetricsStore | None:
"""Return a SQL-backed metrics store, or ``None`` when no DB is configured."""
if session_factory is None:
return None
return SqlLlmMetricsStore(session_factory)
@@ -0,0 +1,66 @@
"""Abstract repository contract for LLM call metrics."""
from __future__ import annotations
from abc import ABC, abstractmethod
from dataclasses import dataclass
from datetime import datetime
from typing import Any
@dataclass
class LlmMetricRecord:
"""In-memory shape of a metric row before persistence."""
id: str
created_at: datetime
user_id: str | None = None
thread_id: str | None = None
run_id: str | None = None
agent_name: str | None = None
model_name: str | None = None
duration_ms: int = 0
input_tokens: int | None = None
output_tokens: int | None = None
total_tokens: int | None = None
tokens_per_sec: float | None = None
status: str = "success"
error_type: str | None = None
error_message: str | None = None
@dataclass
class LlmMetricsQuery:
"""Filter set for the admin list / export endpoints."""
user_id: str | None = None
model_name: str | None = None
status: str | None = None
since: datetime | None = None
until: datetime | None = None
limit: int = 100
offset: int = 0
class LlmMetricsStore(ABC):
"""Persistence contract for the metrics middleware + admin API."""
@abstractmethod
async def record(self, metric: LlmMetricRecord) -> None:
"""Persist one metric row. Must never raise on best-effort writes."""
raise NotImplementedError
@abstractmethod
async def list(self, query: LlmMetricsQuery) -> list[dict[str, Any]]:
"""Return the matching rows (newest first), paginated by ``query``."""
raise NotImplementedError
@abstractmethod
async def count(self, query: LlmMetricsQuery) -> int:
"""Return the unpaginated row count for ``query``."""
raise NotImplementedError
@abstractmethod
async def iter_all(self, query: LlmMetricsQuery):
"""Yield rows matching ``query`` without pagination — used by export."""
raise NotImplementedError
@@ -0,0 +1,62 @@
"""ORM model for per-LLM-call metrics."""
from __future__ import annotations
from datetime import UTC, datetime
from sqlalchemy import DateTime, Float, Integer, String, Text
from sqlalchemy.orm import Mapped, mapped_column
from deerflow.persistence.base import Base
class LlmCallMetricRow(Base):
"""One row per completed (or failed) LLM provider invocation."""
__tablename__ = "llm_call_metrics"
id: Mapped[str] = mapped_column(String(64), primary_key=True)
# Wall-clock time the LLM call ended. Indexed because admin views always
# sort/filter by recency.
created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), default=lambda: datetime.now(UTC), index=True)
user_id: Mapped[str | None] = mapped_column(String(64), index=True)
thread_id: Mapped[str | None] = mapped_column(String(64), index=True)
run_id: Mapped[str | None] = mapped_column(String(64), index=True)
agent_name: Mapped[str | None] = mapped_column(String(128), index=True)
model_name: Mapped[str | None] = mapped_column(String(128), index=True)
# Wall-clock duration of the provider invocation in milliseconds. Always set.
duration_ms: Mapped[int] = mapped_column(Integer, default=0)
input_tokens: Mapped[int | None] = mapped_column(Integer)
output_tokens: Mapped[int | None] = mapped_column(Integer)
total_tokens: Mapped[int | None] = mapped_column(Integer)
# Output tokens / seconds; null when there is no completion (errors).
tokens_per_sec: Mapped[float | None] = mapped_column(Float)
# "success" or "error". Errors store the classification and original message
# so admins can scan for transient vs auth/quota issues.
status: Mapped[str] = mapped_column(String(16), default="success", index=True)
error_type: Mapped[str | None] = mapped_column(String(64))
error_message: Mapped[str | None] = mapped_column(Text)
def to_dict(self) -> dict:
return {
"id": self.id,
"created_at": self.created_at.isoformat() if isinstance(self.created_at, datetime) else self.created_at,
"user_id": self.user_id,
"thread_id": self.thread_id,
"run_id": self.run_id,
"agent_name": self.agent_name,
"model_name": self.model_name,
"duration_ms": self.duration_ms,
"input_tokens": self.input_tokens,
"output_tokens": self.output_tokens,
"total_tokens": self.total_tokens,
"tokens_per_sec": self.tokens_per_sec,
"status": self.status,
"error_type": self.error_type,
"error_message": self.error_message,
}
@@ -0,0 +1,109 @@
"""SQLAlchemy-backed LLM metrics repository."""
from __future__ import annotations
import logging
from collections.abc import AsyncIterator
from datetime import UTC, datetime
from typing import Any
from sqlalchemy import desc, func, select
from sqlalchemy.ext.asyncio import AsyncSession, async_sessionmaker
from deerflow.persistence.llm_metrics.base import LlmMetricRecord, LlmMetricsQuery, LlmMetricsStore
from deerflow.persistence.llm_metrics.model import LlmCallMetricRow
logger = logging.getLogger(__name__)
def _row_to_dict(row: LlmCallMetricRow) -> dict[str, Any]:
return row.to_dict()
def _normalize_created_at(value: datetime) -> datetime:
return value if value.tzinfo else value.replace(tzinfo=UTC)
class SqlLlmMetricsStore(LlmMetricsStore):
"""SQLite/Postgres-backed implementation; writes are intentionally fire-and-forget."""
def __init__(self, session_factory: async_sessionmaker[AsyncSession]) -> None:
self._sf = session_factory
async def record(self, metric: LlmMetricRecord) -> None:
row = LlmCallMetricRow(
id=metric.id,
created_at=_normalize_created_at(metric.created_at),
user_id=metric.user_id,
thread_id=metric.thread_id,
run_id=metric.run_id,
agent_name=metric.agent_name,
model_name=metric.model_name,
duration_ms=metric.duration_ms,
input_tokens=metric.input_tokens,
output_tokens=metric.output_tokens,
total_tokens=metric.total_tokens,
tokens_per_sec=metric.tokens_per_sec,
status=metric.status,
error_type=metric.error_type,
error_message=metric.error_message,
)
try:
async with self._sf() as session:
session.add(row)
await session.commit()
except Exception:
# Metrics are best-effort. Never let a persistence failure surface
# to the agent runtime — log and move on.
logger.warning("Failed to persist llm_call_metrics row %s", metric.id, exc_info=True)
def _apply_filters(self, stmt, query: LlmMetricsQuery):
if query.user_id:
stmt = stmt.where(LlmCallMetricRow.user_id == query.user_id)
if query.model_name:
stmt = stmt.where(LlmCallMetricRow.model_name == query.model_name)
if query.status:
stmt = stmt.where(LlmCallMetricRow.status == query.status)
if query.since is not None:
stmt = stmt.where(LlmCallMetricRow.created_at >= _normalize_created_at(query.since))
if query.until is not None:
stmt = stmt.where(LlmCallMetricRow.created_at <= _normalize_created_at(query.until))
return stmt
async def list(self, query: LlmMetricsQuery) -> list[dict[str, Any]]:
stmt = select(LlmCallMetricRow)
stmt = self._apply_filters(stmt, query)
stmt = stmt.order_by(desc(LlmCallMetricRow.created_at)).limit(max(1, min(query.limit, 1000))).offset(max(0, query.offset))
async with self._sf() as session:
result = await session.execute(stmt)
return [_row_to_dict(row) for row in result.scalars()]
async def count(self, query: LlmMetricsQuery) -> int:
stmt = select(func.count()).select_from(LlmCallMetricRow)
stmt = self._apply_filters(stmt, query)
async with self._sf() as session:
return (await session.scalar(stmt)) or 0
async def iter_all(self, query: LlmMetricsQuery) -> AsyncIterator[dict[str, Any]]:
# Page through the result so the Excel export stays bounded in memory
# for large date windows.
page_size = 500
offset = 0
while True:
page_query = LlmMetricsQuery(
user_id=query.user_id,
model_name=query.model_name,
status=query.status,
since=query.since,
until=query.until,
limit=page_size,
offset=offset,
)
page = await self.list(page_query)
if not page:
return
for item in page:
yield item
if len(page) < page_size:
return
offset += page_size
@@ -0,0 +1,59 @@
"""Create llm_call_metrics table.
Revision ID: 20260514_02
Revises: 20260514_01
Create Date: 2026-05-14
"""
from __future__ import annotations
import sqlalchemy as sa
from alembic import op
from sqlalchemy import inspect
revision = "20260514_02"
down_revision = "20260514_01"
branch_labels = None
depends_on = None
def _has_table(table_name: str) -> bool:
inspector = inspect(op.get_bind())
return table_name in inspector.get_table_names()
def upgrade() -> None:
if _has_table("llm_call_metrics"):
return
op.create_table(
"llm_call_metrics",
sa.Column("id", sa.String(64), primary_key=True),
sa.Column("created_at", sa.DateTime(timezone=True), nullable=False),
sa.Column("user_id", sa.String(64), nullable=True),
sa.Column("thread_id", sa.String(64), nullable=True),
sa.Column("run_id", sa.String(64), nullable=True),
sa.Column("agent_name", sa.String(128), nullable=True),
sa.Column("model_name", sa.String(128), nullable=True),
sa.Column("duration_ms", sa.Integer(), nullable=False, server_default="0"),
sa.Column("input_tokens", sa.Integer(), nullable=True),
sa.Column("output_tokens", sa.Integer(), nullable=True),
sa.Column("total_tokens", sa.Integer(), nullable=True),
sa.Column("tokens_per_sec", sa.Float(), nullable=True),
sa.Column("status", sa.String(16), nullable=False, server_default="success"),
sa.Column("error_type", sa.String(64), nullable=True),
sa.Column("error_message", sa.Text(), nullable=True),
)
op.create_index("ix_llm_call_metrics_created_at", "llm_call_metrics", ["created_at"])
op.create_index("ix_llm_call_metrics_user_id", "llm_call_metrics", ["user_id"])
op.create_index("ix_llm_call_metrics_model_name", "llm_call_metrics", ["model_name"])
op.create_index("ix_llm_call_metrics_status", "llm_call_metrics", ["status"])
op.create_index("ix_llm_call_metrics_thread_id", "llm_call_metrics", ["thread_id"])
op.create_index("ix_llm_call_metrics_run_id", "llm_call_metrics", ["run_id"])
op.create_index("ix_llm_call_metrics_agent_name", "llm_call_metrics", ["agent_name"])
def downgrade() -> None:
if _has_table("llm_call_metrics"):
op.drop_table("llm_call_metrics")
@@ -17,7 +17,9 @@ there is no matching entity directory.
from deerflow.persistence.agents.model import AgentRow
from deerflow.persistence.feedback.model import FeedbackRow
from deerflow.persistence.llm_metrics.model import LlmCallMetricRow
from deerflow.persistence.models.run_event import RunEventRow
from deerflow.persistence.recommended_questions.model import RecommendedQuestionRow
from deerflow.persistence.run.model import RunRow
from deerflow.persistence.scheduled_tasks.model import (
ScheduledTaskDeliveryProfileRow,
@@ -32,6 +34,8 @@ from deerflow.persistence.user.model import UserRow
__all__ = [
"FeedbackRow",
"AgentRow",
"LlmCallMetricRow",
"RecommendedQuestionRow",
"RunEventRow",
"RunRow",
"ScheduledTaskDeliveryProfileRow",