You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
airflow-core/tests/unit/triggers/test_shared_stream.py:1437: in test_ack_mode_queue_full_during_fanout_does_not_break_iteration
await asyncio.wait_for(collect_full(), timeout=2.0)
...
E TimeoutError
This seems to be a flaky or recently introduced timing issue in the shared stream trigger tests and is currently blocking unrelated PRs when the full core test suite runs.
This looks like a flaky timing-related failure rather than something specific to the linked PR, especially since the same test is also failing on main with a TimeoutError.
Since the timeout occurs in test_ack_mode_queue_full_during_fanout_does_not_break_iteration, it may be worth checking whether the test is relying on timing assumptions that are too tight for CI environments. The asyncio.wait_for(..., timeout=2.0) timeout may not be sufficient under slower or loaded CI runners, especially if queue backpressure or fanout scheduling is involved.
It may also be useful to confirm whether the collector coroutine can get blocked waiting for an item that is never yielded after the queue-full condition is triggered.
This seems important to stabilize because it can block unrelated PRs when the full core test suite runs.
Activity
shahar1 commentedon Jun 16, 2026
@Lee-W could be related to #67523?
venkatamandavilli-code commentedon Jun 17, 2026
Thanks for reporting this.
This looks like a flaky timing-related failure rather than something specific to the linked PR, especially since the same test is also failing on main with a
TimeoutError.Since the timeout occurs in
test_ack_mode_queue_full_during_fanout_does_not_break_iteration, it may be worth checking whether the test is relying on timing assumptions that are too tight for CI environments. Theasyncio.wait_for(..., timeout=2.0)timeout may not be sufficient under slower or loaded CI runners, especially if queue backpressure or fanout scheduling is involved.It may also be useful to confirm whether the collector coroutine can get blocked waiting for an item that is never yielded after the queue-full condition is triggered.
This seems important to stabilize because it can block unrelated PRs when the full core test suite runs.
Lee-W commentedon Jun 17, 2026
I think #68651 makes sense. Let's see how it goes
Lee-W commentedon Jun 17, 2026
cc @FrankYang0529