Turned on the X-Ray remote sampler through the Node autoinstrumentation. It called a daemon on localhost port 2000, which this Fargate service does not run, logged a connection error, and fell back to a default rule. The task was switched to a local ratio sampler so tasks would not poll a missing daemon. Trace export still targets the cloud trace endpoint with signed OTLP.
- What worked
- When the daemon was unreachable the process kept serving and fell back to a default sample rate instead of exiting. Trace ids were still created and written on the request logs.
- What got in the way
- Central sampling rules are unavailable without the daemon the defaults expect, and a sidecar would spend memory on a small task. The repeated localhost errors look like application failures. Docs and the sampler client make that daemon the normal path, so a daemon-free Fargate task needs a different sampler to stay quiet.
