Traces from Vercel and Next.js: from the app or with a Trace Drain

Two ways to get traces from an app on Vercel into Belvador: let your app export them (any plan), or let Vercel forward its own traces with a Trace Drain (Pro and Enterprise). Logs need the first way — Vercel's Log Drains do not speak OTLP.

Not tested yet: this setup follows the vendor's documentation; Belvador has not run it end to end.

From your app (Next.js)#

Next.js loads instrumentation.ts at start-up. Export with the standard OpenTelemetry Node SDK in the Node.js runtime only — it does not run in the Edge runtime:

// instrumentation.tsexport async function register() {  if (process.env.NEXT_RUNTIME === "nodejs") {    await import("./instrumentation.node");  }}

// instrumentation.node.tsimport { getNodeAutoInstrumentations } from "@opentelemetry/auto-instrumentations-node";import { NodeSDK } from "@opentelemetry/sdk-node";new NodeSDK({ instrumentations: [getNodeAutoInstrumentations()] }).start();

Install @opentelemetry/sdk-node and @opentelemetry/auto-instrumentations-node, and set the standard variables in Project Settings → Environment Variables — the key as a sensitive variable, never a NEXT_PUBLIC_ one. The SDK reads the endpoint, header, service name and environment from them, and sends logs from Pino, Bunyan or Winston as on Node.js.

@vercel/otel (registerOTel) also works if you prefer it; give it an explicit OTLP exporter so it does not pick a Vercel-configured one.

Serverless functions freeze when the response is sent, so a batch can wait until the next request or be lost. Belvador has not measured how much on Vercel.

With a Trace Drain (Pro, Enterprise)#

Vercel forwards the traces it records for your deployments — no code. In Team Settings → Drains → Add Drain:

Use an ingest key with Traces. Vercel also adds its own x-vercel-signature header; Belvador authenticates with the bearer key and ignores it. Sampling rules run top to bottom and drop requests no rule matches; with no rule, every trace is forwarded. Resource attributes include vercel.projectId and vercel.deploymentId.

Logs from Vercel's Log Drains#

Log Drains send Vercel's JSON or NDJSON, not OTLP, so Belvador's logs endpoint cannot read them directly. Send logs from the app as above, or put a Collector in between with a receiver that parses Vercel's format.

Status#

Both ways follow Vercel's and OpenTelemetry's documentation; Belvador has not run them on Vercel yet, so they are marked untested. After setup, run the connection check and look for your service under Services seen.