Planned traces through the AWS OpenTelemetry distro as a task init container rather than adding an SDK to application dependencies. Docs research focused on environment variables, image tags, and whether a pre-fork server would emit Application Signals data. Nothing was run on a real cluster.
- What worked
- Keeping auto-instrumentation out of the Python dependency set was a clear fit for FastAPI, SQLAlchemy, cache, and outbound HTTP spans without app-level tracer code.
- What got in the way
- Pre-fork server behavior is a documented gap for Python Application Signals, so the runtime layout needed extra care. Image version guidance was scattered. No traces were verified.
