[kurok]

Fourth CI flake this session. It turned PR #291 red on macos-latest 3.13.

Root cause

test_the_master_streams_the_cluster_during_the_run asserted:

self.assertTrue(all(e["workers"] == "2" for e in intervals), master_exports)

The first interval snapshot can legitimately land before a worker has sent its first progress message. macos-latest 3.13 produced exactly that:

workers '1' (total 60), then '2' (126), '2' (126), '2' (290), final '2' (418)

That is a correct cluster view during ramp-up — a worker that has not reported yet is not reporting — and the assertion called it a failure.

The fix

Assert what is actually true: the view converges.

counts = [int(e["workers"]) for e in intervals]
self.assertLessEqual(max(counts), 2, master_exports)   # never invents a third
self.assertEqual(counts[-1], 2, master_exports)        # both counted by the end
self.assertEqual(int(master_exports[-1]["workers"]), 2, master_exports)  # final is complete

The point of the original assertion — that the master merges both workers rather than one — is preserved by the last two checks. Ramp-up is not a defect.

Unrelated to #291's change: the workers_reported tags added there go on the end-of-run report config, while this count comes from the progress aggregator. #291 only shifted the timing that exposed it.

Checks

ruff format/check clean, 1920 passed. Ran the test twice locally.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

📢 Thoughts on this report? Let us know!