Package Name

llmobs

Package Version(s)

No response

Describe the goal of the feature

Currently, LLMObs writers are silently killed on SIGTERM without flushing buffered spans. Relevant logs look like this:

Shutting down tracer with 48 spans. These spans will not be sent to Datadog: trace_id=...

The expected behavior is that LLMObs writers should handle SIGTERM gracefully and attempt to flush buffered spans before exit, and the shutdown timeout, i.e., the time to wait for writers to flush, should be configurable.

Is your feature request related to a problem?

Unflushed spans get dropped when pods restart or during deployments resulting in broken traces.

Describe alternatives you've considered

No response

Additional context

No response