test_reschedule_concurrent_requests_deadlock added in #5379 may be flaky. In https://github.com/dask/distributed/runs/4099410548?check_suite_focus=true (from #5208) it failed on all Windows machines, but no others.

The failures all seem to be gen_cluster timing out after 30s, so I'm not even sure at which step in the test it's deadlocking. I'd guess either

Perhaps #5370 is not fully closed?

cc @fjetter