Does Langfuse tracing work in Vercel edge functions?

[holdenmatt]

Has anyone gotten Langfuse working in Vercel "edge" functions?

I have a nextjs app hosted on Vercel using the "ai" sdk. I followed the guide here to setup Langfuse tracing using OpenTelemetry: https://langfuse.com/docs/integrations/vercel-ai-sdk

I see traces from my dev environment but none from my production app. After debugging a while, I disabled the "edge" runtime in my api route and production traces finally appeared.

If I use "edge", I don't see any issues in Vercel build or runtime logs, traces just never appear in Langfuse.

Is this expected?

[hassiebp]

Hi @holdenmatt! Could you please enable debug logs and share whether you see any logs / spans exported in your edge runtime? Also, could you please add a call to the forceFlush method on the LangfuseExporter instance before returning in your handler?

new LangfuseExporter({ debug: true })

[RobertBroersma]

I'm running into the same issue.

I've enabled debug logs in my LangfuseExporter and I can see span-create, generation-create and trace-create in my logs on Vercel.

I'm using streamText from the Vercel AI SDK and I've added await langfuseExporter.forceFlush() to the end of the onFinish callback. I can see the

[LangfuseExporter] Force flushing Langfuse...

in my logs yet nothing appears in my dashboard.

I've also tried using the waitUntil feature from @vercel/functions in the onFinish callback.

Exact same setup works locally.

EDIT: For me it's not working in the node.js runtime either on Vercel :(