Implement distributed tracing

Sometimes, you need to trace a request across multiple services.

LangSmith supports distributed tracing out of the box, linking runs within a trace across services using context propagation headers (langsmith-trace and optional baggage for metadata/tags).

Example client-server setup:

• Trace starts on client
• Continues on server

Distributed tracing in Python

# client.py
from langsmith.run_helpers import get_current_run_tree, traceable
import httpx

@traceable
async def my_client_function():
    headers = {}
    async with httpx.AsyncClient(base_url="...") as client:
        if run_tree := get_current_run_tree():
            # add langsmith-id to headers
            headers.update(run_tree.to_headers())
        return await client.post("/my-route", headers=headers)

Then the server (or other service) can continue the trace by handling the headers appropriately. If you are using an asgi app Starlette or FastAPI, you can connect the distributed trace using LangSmith’s TracingMiddleware.

The TracingMiddleware class was added in langsmith==0.1.133.

Example using FastAPI:

from langsmith import traceable
from langsmith.middleware import TracingMiddleware
from fastapi import FastAPI, Request

app = FastAPI()  # Or Flask, Django, or any other framework
app.add_middleware(TracingMiddleware)

@traceable
async def some_function():
    ...

@app.post("/my-route")
async def fake_route(request: Request):
    return await some_function()

Or in Starlette:

from starlette.applications import Starlette
from starlette.middleware import Middleware
from langsmith.middleware import TracingMiddleware

routes = ...
middleware = [
    Middleware(TracingMiddleware),
]
app = Starlette(..., middleware=middleware)

If you are using other server frameworks, you can always “receive” the distributed trace by passing the headers in through langsmith_extra:

# server.py
import langsmith as ls
from fastapi import FastAPI, Request

@ls.traceable
async def my_application():
    ...

app = FastAPI()  # Or Flask, Django, or any other framework

@app.post("/my-route")
async def fake_route(request: Request):
    # request.headers:  {"langsmith-trace": "..."}
    # as well as optional metadata/tags in `baggage`
    with ls.tracing_context(parent=request.headers):
        return await my_application()

The example above uses the tracing_context context manager. You can also directly specify the parent run context in the langsmith_extra parameter of a method wrapped with @traceable.

# ... same as above

@app.post("/my-route")
async def fake_route(request: Request):
    # request.headers:  {"langsmith-trace": "..."}
    my_application(langsmith_extra={"parent": request.headers})

Distributed tracing in TypeScript

Distributed tracing in TypeScript requires langsmith version >=0.1.31

First, we obtain the current run tree from the client and convert it to langsmith-trace and baggage header values, which we can pass to the server:

// client.mts
import { getCurrentRunTree, traceable } from "langsmith/traceable";

const client = traceable(
    async () => {
        const runTree = getCurrentRunTree();
        return await fetch("...", {
            method: "POST",
            headers: runTree.toHeaders(),
        }).then((a) => a.text());
    },
    { name: "client" }
);

await client();

Then, the server converts the headers back to a run tree, which it uses to further continue the tracing.

To pass the newly created run tree to a traceable function, we can use the withRunTree helper, which will ensure the run tree is propagated within traceable invocations.

Express.JS

// server.mts
import { RunTree } from "langsmith";
import { traceable, withRunTree } from "langsmith/traceable";
import express from "express";
import bodyParser from "body-parser";

    const server = traceable(
        (text: string) => `Hello from the server! Received "${text}"`,
        { name: "server" }
    );

    const app = express();
    app.use(bodyParser.text());

app.post("/", async (req, res) => {
    const runTree = RunTree.fromHeaders(req.headers);
    const result = await withRunTree(runTree, () => server(req.body));
    res.send(result);
});

// server.mts
import { RunTree } from "langsmith";
import { traceable, withRunTree } from "langsmith/traceable";
import { Hono } from "hono";

    const server = traceable(
        (text: string) => `Hello from the server! Received "${text}"`,
        { name: "server" }
    );

    const app = new Hono();

app.post("/", async (c) => {
    const body = await c.req.text();
    const runTree = RunTree.fromHeaders(c.req.raw.headers);
    const result = await withRunTree(runTree, () => server(body));
    return c.body(result);
});

Distributed tracing in Java

Java supports distributed tracing using OpenTelemetry context propagation with the W3C Trace Context standard (traceparent header).

Client

import com.langchain.smith.wrappers.openai.OpenTelemetryConfig;
import com.langchain.smith.wrappers.openai.WrappedOpenAIClient;
import com.openai.models.ChatModel;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import io.opentelemetry.api.GlobalOpenTelemetry;
import io.opentelemetry.api.trace.Span;
import io.opentelemetry.api.trace.Tracer;
import io.opentelemetry.context.Context;
import io.opentelemetry.context.Scope;
import io.opentelemetry.context.propagation.TextMapGetter;
import io.opentelemetry.context.propagation.TextMapSetter;
import java.net.URI;
import java.net.http.HttpClient;
import java.net.http.HttpRequest;
import java.net.http.HttpResponse;
import java.util.HashMap;
import java.util.Map;

public class DistributedTracingClient {
    public static void main(String[] args) throws Exception {
        // Configure OpenTelemetry
        OpenTelemetryConfig.builder()
            .fromEnv()
            .build();

        Tracer tracer = GlobalOpenTelemetry.get().getTracer("client-app");

        // Create parent span
        Span clientSpan = tracer.spanBuilder("client_request")
            .setAttribute("langsmith.span.kind", "chain")
            .startSpan();

        try (Scope scope = clientSpan.makeCurrent()) {
            // Prepare headers for propagation
            Map<String, String> headers = new HashMap<>();

            // Inject trace context into headers
            GlobalOpenTelemetry.getPropagators()
                .getTextMapPropagator()
                .inject(Context.current(), headers, new TextMapSetter<Map<String, String>>() {
                    @Override
                    public void set(Map<String, String> carrier, String key, String value) {
                        carrier.put(key, value);
                    }
                });

            // Make HTTP request to server with propagated context
            HttpClient client = HttpClient.newHttpClient();
            HttpRequest.Builder requestBuilder = HttpRequest.newBuilder()
                .uri(URI.create("http://localhost:8080/process"))
                .POST(HttpRequest.BodyPublishers.ofString("Hello from client"));

            // Add propagated headers
            for (Map.Entry<String, String> entry : headers.entrySet()) {
                requestBuilder.header(entry.getKey(), entry.getValue());
            }

            HttpResponse<String> response = client.send(
                requestBuilder.build(),
                HttpResponse.BodyHandlers.ofString()
            );

            System.out.println("Response: " + response.body());
        } finally {
            clientSpan.end();
        }

        OpenTelemetryConfig.flush(5, java.util.concurrent.TimeUnit.SECONDS);
    }
}

import com.langchain.smith.wrappers.openai.OpenTelemetryConfig;
import com.langchain.smith.wrappers.openai.WrappedOpenAIClient;
import com.openai.models.ChatModel;
import com.openai.models.chat.completions.ChatCompletion;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import io.opentelemetry.api.GlobalOpenTelemetry;
import io.opentelemetry.api.trace.Span;
import io.opentelemetry.api.trace.Tracer;
import io.opentelemetry.context.Context;
import io.opentelemetry.context.Scope;
import io.opentelemetry.context.propagation.TextMapGetter;
import org.springframework.boot.SpringApplication;
import org.springframework.boot.autoconfigure.SpringBootApplication;
import org.springframework.web.bind.annotation.*;
import jakarta.servlet.http.HttpServletRequest;
import java.util.Enumeration;
import java.util.HashMap;
import java.util.Map;

@SpringBootApplication
@RestController
public class DistributedTracingServer {

    static {
        // Configure OpenTelemetry
        OpenTelemetryConfig.builder()
            .fromEnv()
            .build();
    }

    private final Tracer tracer = GlobalOpenTelemetry.get().getTracer("server-app");
    private final WrappedOpenAIClient openaiClient = WrappedOpenAIClient.fromEnv();

    @PostMapping("/process")
    public String processRequest(@RequestBody String message, HttpServletRequest request) {
        // Extract trace context from incoming headers
        Map<String, String> headers = new HashMap<>();
        Enumeration<String> headerNames = request.getHeaderNames();
        while (headerNames.hasMoreElements()) {
            String headerName = headerNames.nextElement();
            headers.put(headerName, request.getHeader(headerName));
        }

        // Extract parent context
        Context extractedContext = GlobalOpenTelemetry.getPropagators()
            .getTextMapPropagator()
            .extract(Context.current(), headers, new TextMapGetter<Map<String, String>>() {
                @Override
                public Iterable<String> keys(Map<String, String> carrier) {
                    return carrier.keySet();
                }

                @Override
                public String get(Map<String, String> carrier, String key) {
                    return carrier.get(key);
                }
            });

        // Create server span as child of extracted context
        Span serverSpan = tracer.spanBuilder("server_process")
            .setParent(extractedContext)
            .setAttribute("langsmith.span.kind", "chain")
            .startSpan();

        try (Scope scope = serverSpan.makeCurrent()) {
            // Process with OpenAI (will be a child span)
            ChatCompletion completion = openaiClient.chat().completions().create(
                ChatCompletionCreateParams.builder()
                    .model(ChatModel.GPT_4O_MINI)
                    .addSystemMessage("Respond to the user's message.")
                    .addUserMessage(message)
                    .build()
            );

            String response = completion.choices().get(0).message().content().orElse("");
            serverSpan.setAttribute("response", response);

            return response;
        } finally {
            serverSpan.end();
        }
    }

    public static void main(String[] args) {
        SpringApplication.run(DistributedTracingServer.class, args);
    }
}

Java distributed tracing uses the W3C Trace Context standard (traceparent and tracestate headers) via OpenTelemetry’s context propagation API. The trace context is automatically propagated across service boundaries when you inject and extract the context using GlobalOpenTelemetry.getPropagators().

---

Edit the source of this page on GitHub.

Connect these docs programmatically to Claude, VSCode, and more via MCP for real-time answers.