Tracing Distributed Systems in Next.js

Published April 21, 2024

portrait of Chris Esplin.

by Chris Esplin

Watch on YouTube

Distributed Tracing

Distributed tracing is crucial for debugging and improving system performance. Instead of looking at the performance of each service in isolation, distributed tracing links traces across services to show how a request flows through a system. For instance, a trace might start in a React web client, pass through a Next.js API, and continue across subsequent HTTP calls, keeping all activities under one umbrella. This not only simplifies tracking but also provides a clear, consolidated view of how different parts of your application interact.

Distributed tracing flame graph showing how a request flows across multiple services.

Distributed tracing flame graph showing how a request flows across multiple services.

Approach

In this post, we’ll go over a practical example of how tracing across multiple services work in the context of a modern NextJS application.

Here’s the exact github repo we’ll be pulling the example from:

GitHub: Distributed Tracing Example

We will demonstrate how to use LaunchDarkly to connect a NextJS application with an external service by passing HTTP headers in the following steps:

  • Next Client: The web client initiates an HTTP Fetch request, passing the x-highlight-request header which contains a generated sessionId and requestId.
  • Next Server: A Next.js API function receives that request, associates the sessionId and requestId to it’s local trace, then passes the x-highlight-request header to a third service, this time written in Golang.
  • Go Server: The Golang service receives the x-highlight-request header and associates it to the local trace.

Instrumenting the Next.js web client

Client-side integration requires initializing the LaunchDarkly SDK and its observability plugins in your web app.

  • Initialize the SDK once, preferably in a layout.tsx file or _app.tsx file.
  • The observability plugin accepts configuration options documented in the JavaScript SDK’s options reference.
  • Explore LaunchDarkly’s session replay configuration options, including:
    • Canvas and WebGL capture,
    • Privacy and redaction settings
1// app/layout.tsx
2import { CONSTANTS } from "../constants";
3import { HighlightInit } from "@highlight-run/next/client";
4
5export default function RootLayout({
6 children,
7}: {
8 children: React.ReactNode;
9}) {
10 return (
11 <>
12 <HighlightInit
13 // excludedHostnames={['localhost']}
14 projectId={CONSTANTS.ENV.HIGHLIGHT.PROJECT_ID}
15 serviceName="my-nextjs-frontend"
16 tracingOrigins
17 networkRecording={{
18 enabled: true,
19 recordHeadersAndBody: true,
20 }}
21 debug
22 />
23
24 <html lang="en">
25 <body>{children}</body>
26 </html>
27 </>
28 );
29}

Calling the Next.js API function

We’ll use the browser’s Fetch API to call a Next.js API function. With tracingOrigins configured, the LaunchDarkly observability plugin monkey patches window.fetch and automatically adds a traceparent header to matching requests, so use Fetch as you normally would.

Read about the full list of monkey patched browser APIs in our docs.

1 <button
2 onClick={async () => {
3 await fetch("/propagation-test");
4 }}
5>Trigger propagation</button>

First, create a wrapper function around your API route handler. Use the LaunchDarkly observability plugin to capture incoming request headers and wrap the handler in a custom span.

In this case, we’re using the App Router. For SDK setup and configuration details, read the JavaScript SDK observability reference. For a walkthrough of instrumenting Next.js App Router routes specifically, read How to use OpenTelemetry to monitor Next.js apps.

1// utils/app-router-highlight.config.ts:
2import { AppRouterHighlight } from "@highlight-run/next/server";
3import { CONSTANTS } from "../../constants";
4
5export const withAppRouterHighlight = AppRouterHighlight({
6 projectID: CONSTANTS.ENV.HIGHLIGHT.PROJECT_ID,
7});

Next, we wrap our GET handler function. Note that we wrap our handler function with withAppRouterHighlight, and we forward request.headers along with the next HTTP request.

The x-highlight-request header is only header that matters in this example.

1// app/propagation-test/route.ts
2import { NextRequest } from "next/server";
3import { withAppRouterHighlight } from "../_utils/app-router-highlight.config";
4
5export const GET = withAppRouterHighlight(async (request: NextRequest) => {
6 console.info("Here: app/propagation-test/route.ts");
7
8 const response = await fetch("http://localhost:3010/test", {
9 method: "GET",
10 headers: request.headers, // x-highlight-request is the critical header
11 });
12
13 const data = await response.text();
14
15 return new Response(data);
16});

Configure our Golang service

Configure your Go service

LaunchDarkly’s Go SDK supports observability through a dedicated observability plugin. This example uses that plugin alongside the chi router.

See a working example of LaunchDarkly chi implementation on GitHub.

First, configure the observability plugin, provided by the github.com/launchdarkly/observability-sdk/go package:

1import (
2 "github.com/highlight/highlight/sdk/highlight-go"
3)
4
5func main() {
6 // ...
7 highlight.SetProjectID("<YOUR_PROJECT_ID>")
8 highlight.Start(
9 highlight.WithServiceName("my-app"),
10 highlight.WithServiceVersion("git-sha"),
11 )
12 defer highlight.Stop()
13 // ...
14}

Next we add the Chi middleware.

1import (
2 highlightChi "github.com/highlight/highlight/sdk/highlight-go/middleware/chi"
3)
4
5func main() {
6 // ...
7 r := chi.NewRouter()
8 r.Use(highlightChi.Middleware)
9 // ...
10}

The highlightChi.Middleware function will automatically associate the current request with the incoming x-highlight-request header.

W3C traceparent example code

We used the x-highlight-request header in this example, but we also tested OpenTelemetry’s native JavaScript Propagation example using the W3C traceparent header.

Instead of passing x-highlight-request header, it is possible to pass a traceparent header across HTTP calls. It’s a bit involved, but it works great. See the links below for example code:

Result

All three services, the web client, the Next.js API function, and the Golang backend, now feed their data to the same LaunchDarkly project.

For the purposes of example, we can even visualize this data in our product, LaunchDarkly!

This example shows four separate spans, all of which are nested under a single trace. And of course, these traces are accessible from LaunchDarkly’s Session Replay tab, because the session was generated in the web client and propagated throughout the entire distributed “call stack”.

  • highlight-run-with-headers is generated by the AppRouterHighlight wrapper function in Next.js.
  • fetch GET https://localhost:3010/test is generated by the Next.js Fetch request.
  • highlight.chi is created by LaunchDarkly’s chi middleware.
  • go-custom-span is a custom span that we used in our example code for demonstration purposes. It wraps a call to time.Sleep(1 * time.Second) to mimic slow code execution.

LaunchDarkly trace view showing the highlight-run-with-headers span wrapping the distributed request.

LaunchDarkly trace view showing the highlight-run-with-headers span wrapping the distributed request.

highlight-run-with-headers

LaunchDarkly trace view showing the go-custom-span from the Golang backend service.

LaunchDarkly trace view showing the go-custom-span from the Golang backend service.

go-custom-span