-
Notifications
You must be signed in to change notification settings - Fork 7.3k
Intermittent failures on test-regress-GH-14576.js #25619
Comments
The goal of this test is to test a bug when, sometimes, node would stall. So it could be that the bug was actually not fixed, just that it happens less often. If that is the case, we would need to reopen the issue too. |
Adding as P-1 since the longer we wait to investigate this, the less context we'll have around what had been done initially to fix the problem and write the test. Adding to 0.12.7 milestone since a v0.12.6 release should be coming up very soon. Also, /cc @tunniclm @dmelikyan |
@misterdjules can you add the original issue number ? |
Yes thanks |
My initial thought when looking through this a bit, is that on Windows SuspendThread() and ResumeThread() are used instead of signals/SignalHandler and so the fix here is to code that is not used on Windows. It is possible that the suspended thread holds some important locks (such as the process_wide_mutex_) at the time it is suspended to take a sample, however, so far I have not found an obvious code that takes locks in the sampler. It would be good to recreate this hang so we can capture some stack traces to see where the threads are stuck. |
Reproduced this on 32-bit windows and took a memory dump. Key thread stacks:
|
So Thread |
Thread I took a second memory dump some minutes after the first and Thread So it looks like the suspended thread holds a CriticalSection that is required to resume it. |
The |
Two failures in recent CI history
http://jenkins.nodejs.org/job/node-test-commit-windows/DESTCPU=ia32,label=windows-2k8r2/121/
http://jenkins.nodejs.org/job/node-test-commit-windows/DESTCPU=ia32,label=windows-2k8r2/110/
Need to either investigate/fix or mark as flaky test until we do.
The text was updated successfully, but these errors were encountered: