test_observations_fit_convert_predict_pipeline intermittently hangs on macOS and is killed by the 300 s pytest timeout, taking the whole leg red. Seen on build (macos-latest, 3.13) in https://github.com/Smithsonian/layup/actions/runs/33888892047 (1 failed, 552 passed); the traceback ends in RuntimeError: cannot wait on un-acquired lock.
It is a hang rather than slowness: the same test passes locally in 7.9 s, and the immediately preceding CI run passed on code differing only by whitespace. Only one of the four legs failed, and re-running it on the next commit passed.
Distinct from #523, which is about orphaned subprocesses after Ctrl+C on Linux. There is no interrupt here. Both live in process_data_by_id, so a fix for one may not cure the other.
Effect: any correct PR can show red at random.
test_observations_fit_convert_predict_pipelineintermittently hangs on macOS and is killed by the 300 s pytest timeout, taking the whole leg red. Seen onbuild (macos-latest, 3.13)in https://github.com/Smithsonian/layup/actions/runs/33888892047 (1 failed, 552 passed); the traceback ends inRuntimeError: cannot wait on un-acquired lock.It is a hang rather than slowness: the same test passes locally in 7.9 s, and the immediately preceding CI run passed on code differing only by whitespace. Only one of the four legs failed, and re-running it on the next commit passed.
Distinct from #523, which is about orphaned subprocesses after Ctrl+C on Linux. There is no interrupt here. Both live in
process_data_by_id, so a fix for one may not cure the other.Effect: any correct PR can show red at random.