Can't see any traces when app is deployed on AWS ECS

[alexyca]

Describe your question

I have a node server on Docker using vercel ai-sdk with langfuse setup (not self-hosted).

Locally, I can get every traces on langfuse dashboard, but when the docker is deployed on ECS, absolutely no data is sent to langfuse. Even debug mode active doesn't log anything in Cloudwatch (all logs are shown locally). The code is the same and all env variables are correct.

Is anyone had trouble linking langfuse and ECS ?

Here are what I tried and didn't work :

• Forced exportMode "immediate" on the LangfuseSpanProcessor
• Forced flush the spanProcessor before the end of my process
• Added debug mode but nothing on AWS
• Search similar issue on GitHub and Asked to Langfuse AI Chatbot

Langfuse Cloud or Self-Hosted?

Langfuse Cloud

If Self-Hosted

No response

If Langfuse Cloud

No response

SDK and integration versions

No response

Pre-Submission Checklist

• I have checked for existing issues/discussions and consulted Langfuse AI.

[anntnzrb]

ECS containers receive SIGTERM before shutdown, but your app isn't flushing traces before terminating. Add a graceful shutdown handler:

const gracefulShutdown = async () => {
  await langfuseSpanProcessor.forceFlush();
  await sdk.shutdown();
  process.exit(0);
};
process.on("SIGTERM", gracefulShutdown);

Also verify your ECS security group allows outbound HTTPS (port 443) to reach Langfuse Cloud, and check that LANGFUSE_HOST, LANGFUSE_PUBLIC_KEY, LANGFUSE_SECRET_KEY are correctly set.

[alexyca]

@Steffen911 hey, I tested a curl / fetch to https://cloud.langfuse.com/api/public/health returning a 200 ok status in ECS + tested in the code during execution.

All me logs work in cloudwatch (even logged langfuse env var are corrects)

[Steffen911]

@alexyca Can you try to run the auth check command and report the result?
https://langfuse.com/docs/observability/sdk/python/setup#initialize-client

[alexyca]

@Steffen911 Sure, but I can't find a way to do it in TS.

For information, I'm using these TS libs :

import { LangfuseSpanProcessor } from "@langfuse/otel"
import { NodeSDK } from "@opentelemetry/sdk-node";

[hassiebp]

Could you please share how you have enabled debug mode? Did you correctly set LANGFUSE_LOG_LEVEL=debug in your deployed environment?

[alexyca]

Yes, I configured the debug mode correctly. I explained the solution bellow.

[alexyca]

I forced a testSpan manually with the use of trace.getTracer() and context.with() to create a test span with trace and context from "@opentelemetry/api".

It's working well locally, but still nothing when deployed.

[alexyca]

For anyone else experiencing the same issue, I found the cause.

I had imported Sentry somewhere in the project. I wasn't using it, but I had imported it. Locally, everything worked fine, but on ECS, Sentry was overwriting Langfuse's opentelemetry configuration.

That's why I had no logs or errors because Sentry was absorbing everything.

👉 I discovered this by manually creating a scope and console.log that scope. I found that the processor was Sentry's and not Langfuse's, as it was locally.

I just delete Sentry import, and everything was just fine.