test_magic_folder timeout on windows #2914

New Issue

warner · 2018-03-29T20:50:46Z

warner commented

2018-03-29 20:50:46 +00:00

Our Appveyor CI system shows test_magic_folder.RealTest.test_batched_process hitting a timeout error ("test still running after 120 seconds") on windows, on maybe one out of every 5 builds.

I've reconfigured .appveyor.yml to include the trial test.log as an artifact, so we can see if it's throwing an error, or if it's just taking too long and sometimes exceeds the limit. Then we can either be lazy and increase the timeout, or actually make it run faster, or if it's an intermittent deadlock then fix it for real.

Our Appveyor CI system shows `test_magic_folder.RealTest.test_batched_process` hitting a timeout error ("test still running after 120 seconds") on windows, on maybe one out of every 5 builds. I've reconfigured `.appveyor.yml` to include the trial test.log as an artifact, so we can see if it's throwing an error, or if it's just taking too long and sometimes exceeds the limit. Then we can either be lazy and increase the timeout, or actually make it run faster, or if it's an intermittent deadlock then fix it for real.

warner added the

labels 2018-03-29 20:50:46 +00:00

warner added this to the 1.13.0 milestone 2018-03-29 20:50:46 +00:00

warner commented

2018-03-29 21:05:19 +00:00

This test takes 0.441 seconds on our linux buildbot ("xenial"), and the longest magic-folder test takes 2.3 seconds (test_magic_folder.RealTestAliceBob.test_alice_bob), so I'm guessing this is an intermittent deadlock, rather than a very slow test that would complete if we just increased the timeout somewhat. I'm not sure why we've only seen it on appveyor, though; it suggests that there's something windows-specific about the bug.

This test takes 0.441 seconds on our linux buildbot ("xenial"), and the longest magic-folder test takes 2.3 seconds (`test_magic_folder.RealTestAliceBob.test_alice_bob`), so I'm guessing this is an intermittent deadlock, rather than a very slow test that would complete if we just increased the timeout somewhat. I'm not sure why we've only seen it on appveyor, though; it suggests that there's something windows-specific about the bug.

warner commented

2018-03-30 02:25:04 +00:00

I got some logs: when it passes on appveyor, it only takes about 400ms, just like on linux. I'll wait until we get a failing build to see what's happening there.