Propagate CancelledError in gather_from_workers #8089
Conversation
distributed/utils_comm.py
Outdated
| # Note: CancelledError and asyncio.TimeoutError are rare conditions | ||
| # that can be raised by the network stack. | ||
| # See https://github.com/dask/distributed/issues/8006 | ||
| if isinstance(r, (OSError, asyncio.CancelledError, asyncio.TimeoutError)): |
Unit Test ResultsSee test report for an extended history of previous test failures. This is useful for diagnosing flaky tests. 20 files ±0 20 suites ±0 10h 52m 48s ⏱️ - 14m 11s For more details on these failures, see this check. Results for commit a6c2e94. ± Comparison against base commit 163165b. ♻️ This comment has been updated with latest results. |
fjetter
left a comment
fjetter
left a comment
There was a problem hiding this comment.
I'm -1 on catching the CancelledError just for good measure without us knowing how this can actually be provoked.
The impact of an exception here is very different and depending on where it is actually raised. I don't consider this catch safe.
In fact, with return_exceptions=True we should actually makes sure that BaseExceptions generally are reraised
|
@fjetter check now |
2 participants
Follow-up from #7997 (comment)
Propagate task cancellation downwards from
gather_from_workersto theget-data-from-*tasks. This should hopefully get rid of some of our "task was never awaited" errors.