While tests run fine if running in just a single worker of with -n 4, it seems that when boosting this further to 8 cores, the scheduler emitted an extra event of EMPTY.
Looking at the async task the monitors for empty, it is possible there's a race condition causing this to trigger twice. This should be handled properly although it's hard to create a reproducing example. In general, how can we deterministically create race conditions for testing?
While tests run fine if running in just a single worker of with
-n 4
, it seems that when boosting this further to8
cores, the scheduler emitted an extra event ofEMPTY
.Looking at the async task the monitors for empty, it is possible there's a race condition causing this to trigger twice. This should be handled properly although it's hard to create a reproducing example. In general, how can we deterministically create race conditions for testing?
https://github.com/automl/amltk/blob/0f4e1af869ace60fa658ca759ae04ff8fc9f10c8/src/amltk/scheduling/scheduler.py#L1148-L1165
Error, traceback of test