test(azure): TestStreamer_Stop is flaky under CI load

[jongio]

TestStreamer_Stop in cli/src/internal/azure/realtime_test.go fails intermittently on CI. It failed once during PR #553 and passed on a plain re-run of the same job with no code changes.

Why it's flaky

The test sleeps a fixed 10ms to let the streamer goroutine start, then fails hard on a 1 second timeout (realtime_test.go:582 and :590). On a loaded runner 10ms isn't always enough for the goroutine to reach its receive, so Stop() can fire before anything is listening. The observed failure took 1.01s, right at the fatal timeout.

Evidence it's environmental, not a real bug

PR #553 touched zero files under cli/src/internal/azure/. Re-running the same job on the same commit passed.

Suggested fix

Replace the fixed sleep with real synchronization: have the streamer signal readiness on a channel and block on it, or use require.Eventually. Raising the timeout would only make the flake rarer, not remove it.

[achirothmane]

I’m testing a read-only GitHub Action called CI Retry Gate that analyzes the exact failed workflow attempt before deciding whether a rerun is justified.

Your TestStreamer_Stop case is a useful real-world example because the test failed under CI load and then passed on a plain rerun of the same commit with no code changes.

The historical failure can be inspected directly with:

• run-id: 30404127674
• run-attempt: 1

In that attempt, Test (ubuntu-latest) failed at the Run tests step, while attempt 2 completed successfully.

Minimal read-only trial:
https://github.com/achirothmane/workflow-failure-lab/blob/main/examples/historical-run-trial.yml

It only uses actions: read + contents: read, and historical-attempt mode cannot trigger a rerun.

I’m testing whether the evidence would help distinguish a genuine transient/timing failure from a regression before someone manually retries it.