test_workspace_concurrency failing regularly

[mrocklin]

We've started to see failures in test_diskutils.py::test_workspace_concurrency that I'm unable to understand

_________________________ test_workspace_concurrency __________________________
Traceback (most recent call last):
  File "C:\projects\distributed\distributed\tests\test_diskutils.py", line 260, in test_workspace_concurrency
    _test_workspace_concurrency(tmpdir, 2.0, 6)
  File "C:\projects\distributed\distributed\tests\test_diskutils.py", line 255, in _test_workspace_concurrency
    assert n_purged >= 0.5 * n_created > 0
AssertionError: assert 0 >= (0.5 * 1200)

cc @pitrou , author of that test, in case he's able to explain what is going on

[mrocklin]

Oh, I should point out that this is really only regularly failing on windows

[pitrou]

Can you pinpoint to a specific commit?

[mrocklin]

Not yet, no. "Regularly failing" is a probably too strong to use here. "Fails intermittently, but relatively frequently" would be better.

…

On Thu, Aug 2, 2018 at 4:10 PM, Antoine Pitrou ***@***.***> wrote: Can you pinpoint to a specific commit? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#2155 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AASszOAuHb40JN6fdHJlasT1PEiWnySaks5uM1zIgaJpZM4Vsy3W> .

[mrocklin]

I agree that running this in a for loop on a Windows VM and doing git bisect would be a good approach to isolate things. I was hoping that you might be able to provide some information before doing that, especially because I'm unlikely to get to this near term. I've xfailed this short term in #2152 On Thu, Aug 2, 2018 at 4:15 PM, Matthew Rocklin <mrocklin@anaconda.com> wrote:

…

Not yet, no. "Regularly failing" is a probably too strong to use here. "Fails intermittently, but relatively frequently" would be better. On Thu, Aug 2, 2018 at 4:10 PM, Antoine Pitrou ***@***.***> wrote: > Can you pinpoint to a specific commit? > > — > You are receiving this because you authored the thread. > Reply to this email directly, view it on GitHub > <#2155 (comment)>, > or mute the thread > <https://github.com/notifications/unsubscribe-auth/AASszOAuHb40JN6fdHJlasT1PEiWnySaks5uM1zIgaJpZM4Vsy3W> > . >

[fjetter]

We've seen this popping up occasionally again on OSX