Source examples/online_serving/opentelemetry.

设置 OpenTelemetry POC

1. 安装 OpenTelemetry 包

   pip install \
     'opentelemetry-sdk>=1.26.0,<1.27.0' \
     'opentelemetry-api>=1.26.0,<1.27.0' \
     'opentelemetry-exporter-otlp>=1.26.0,<1.27.0' \
     'opentelemetry-semantic-conventions-ai>=0.4.1,<0.5.0'
   

2. 在 Docker 容器中启动 Jaeger

   # From: https://jaeger.golang.ac.cn/docs/1.57/getting-started/
   docker run --rm --name jaeger \
       -e COLLECTOR_ZIPKIN_HOST_PORT=:9411 \
       -p 6831:6831/udp \
       -p 6832:6832/udp \
       -p 5778:5778 \
       -p 16686:16686 \
       -p 4317:4317 \
       -p 4318:4318 \
       -p 14250:14250 \
       -p 14268:14268 \
       -p 14269:14269 \
       -p 9411:9411 \
       jaegertracing/all-in-one:1.57
   

3. 在新 shell 中，导出 Jaeger IP

   export JAEGER_IP=$(docker inspect   --format '{{ .NetworkSettings.IPAddress }}' jaeger)
   export OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=grpc://$JAEGER_IP:4317
   

   然后为 OpenTelemetry 设置 vLLM 的服务名称，启用与 Jaeger 的不安全连接并运行 vLLM

   export OTEL_SERVICE_NAME="vllm-server"
   export OTEL_EXPORTER_OTLP_TRACES_INSECURE=true
   vllm serve facebook/opt-125m --otlp-traces-endpoint="$OTEL_EXPORTER_OTLP_TRACES_ENDPOINT"
   

4. 在新 shell 中，使用来自虚拟客户端的跟踪上下文发送请求

   export JAEGER_IP=$(docker inspect --format '{{ .NetworkSettings.IPAddress }}' jaeger)
   export OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=grpc://$JAEGER_IP:4317
   export OTEL_EXPORTER_OTLP_TRACES_INSECURE=true
   export OTEL_SERVICE_NAME="client-service"
   python dummy_client.py
   

5. 打开 Jaeger Web UI： https://127.0.0.1:16686/

   在搜索窗格中，选择 vllm-server 服务并点击 Find Traces。您应该获得一个跟踪列表，每个请求对应一个跟踪。

6. 点击跟踪将显示其 span 及其标签。在此演示中，每个跟踪有 2 个 span。一个来自包含提示文本的虚拟客户端，另一个来自包含有关请求元数据的 vLLM。

Exporter Protocol

OpenTelemetry 支持 grpc 或 http/protobuf 作为导出器中跟踪数据的传输协议。默认情况下，使用 grpc。要将 http/protobuf 设置为协议，请按如下方式配置 OTEL_EXPORTER_OTLP_TRACES_PROTOCOL 环境变量

export OTEL_EXPORTER_OTLP_TRACES_PROTOCOL=http/protobuf
export OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=http://$JAEGER_IP:4318/v1/traces
vllm serve facebook/opt-125m --otlp-traces-endpoint="$OTEL_EXPORTER_OTLP_TRACES_ENDPOINT"

FastAPI 的 Instrumentation

OpenTelemetry 允许自动 instrumentation FastAPI。

1. 安装 instrumentation 库

   pip install opentelemetry-instrumentation-fastapi
   

2. 使用 opentelemetry-instrument 运行 vLLM

   opentelemetry-instrument vllm serve facebook/opt-125m
   

3. 向 vLLM 发送请求并在 Jaeger 中查找其跟踪。它应该包含来自 FastAPI 的 span。

示例材料

dummy_client.py

# SPDX-License-Identifier: Apache-2.0

import requests
from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import (
    OTLPSpanExporter)
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import (BatchSpanProcessor,
                                            ConsoleSpanExporter)
from opentelemetry.trace import SpanKind, set_tracer_provider
from opentelemetry.trace.propagation.tracecontext import (
    TraceContextTextMapPropagator)

trace_provider = TracerProvider()
set_tracer_provider(trace_provider)

trace_provider.add_span_processor(BatchSpanProcessor(OTLPSpanExporter()))
trace_provider.add_span_processor(BatchSpanProcessor(ConsoleSpanExporter()))

tracer = trace_provider.get_tracer("dummy-client")

url = "https://127.0.0.1:8000/v1/completions"
with tracer.start_as_current_span("client-span", kind=SpanKind.CLIENT) as span:
    prompt = "San Francisco is a"
    span.set_attribute("prompt", prompt)
    headers = {}
    TraceContextTextMapPropagator().inject(headers)
    payload = {
        "model": "facebook/opt-125m",
        "prompt": prompt,
        "max_tokens": 10,
        "best_of": 20,
        "n": 3,
        "use_beam_search": "true",
        "temperature": 0.0,
        # "stream": True,
    }
    response = requests.post(url, headers=headers, json=payload)