概述

生产环境的 Agent 需要完善的监控体系。

结构化日志

import structlog

structlog.configure(
    processors=[structlog.processors.JSONRenderer()]
)
logger = structlog.get_logger()

logger.info("tool_called",
    tool="get_weather",
    args={"location": "Beijing"},
    agent_id="agent_001",
    trace_id="trace_xxx"
)

追踪

from opentelemetry import trace

tracer = trace.get_tracer(__name__)

with tracer.start_as_current_span("agent_run") as span:
    span.set_attribute("agent.id", agent_id)
    result = agent.run(task)
    span.set_attribute("task.completed", True)

指标

from prometheus_client import Counter, Histogram

tool_calls = Counter('agent_tool_calls', 'Tool calls', ['tool_name'])
agent_duration = Histogram('agent_run_duration', 'Agent run duration')