QA (public) » openQA Project (public)

Not sure how we could run opensuse there - github only provides ubuntu, mac and windows images: https://github.com/actions/runner-images

dheidler wrote in #note-9:

Not sure how we could run opensuse there - github only provides ubuntu, mac and windows images: https://github.com/actions/runner-images

We can use container-in-container like in os-autoinst, see
https://github.com/os-autoinst/os-autoinst/blob/master/.github/workflows/ci.yml#L14

All three merged. Awaiting results in master branch

https://github.com/openSUSE/Mojo-IOLoop-ReadWriteProcess/pull/73 to fix the status branch. Preparing a release as well as discussed in the daily https://github.com/openSUSE/Mojo-IOLoop-ReadWriteProcess/pull/67

Trying to reproduce the failure locally with coverage enabled also no joy with runs=2000 env PERL5OPT="-MDevel::Cover" count-fail-ratio prove -l --timer t/01_run.t

## count-fail-ratio: Run: 170. Fails: 0. Fail ratio 0±0%. No fails, computed failure probability < 1.76%
## mean runtime: 4453±332.37 ms

Running in CI now https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14217779071

okurz wrote in #note-19:

Now first we created a release 1.0.0 and uploaded to https://metacpan.org/pod/Mojo::IOLoop::ReadWriteProcess with great help from tinita but I am missing "authorized releasers" page from https://metacpan.org/dist/Mojo-IOLoop-ReadWriteProcess/permissions , asked szarate

Asked Ettore via Telegram to add okurz as co-mantainer or transfer ownership

Was running in CI now with 1000 runs each based on https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/tree/refs/heads/fix/instability with relevant changes in
https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/blob/refs/heads/fix/instability/.github/workflows/ci-tests.yaml

  build:
    runs-on: ${{ matrix.os }}
    strategy:
…
      fail-fast: false
    name: 🐪 Perl ${{ matrix.perl }} on ${{ matrix.os }}
    steps:
…
        run: |
          curl https://raw.githubusercontent.com/okurz/retry/refs/heads/main/count-fail-ratio > count-fail-ratio
          chmod +x count-fail-ratio
          TEST_SHARED=1 TEST_SUBREAPER=1 runs=1000 ./count-fail-ratio cover -test -report codecovbash

https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14220387974 shows that all linux builds had and macos builds fail due to how unix time in seconds is calculated apparently.

maybe surprisingly Perl 5.16 based tests are super stable according to https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14220387974/job/39846683543#step:6:27012

## count-fail-ratio: Run: 1000. Fails: 0. Fail ratio 0±0%. No fails, computed failure probability < .30%

other runs show problem. First problem https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14220387974/job/39846683540#step:6:886

Errors in testing.  Cannot continue.
t/01_run.t (Wstat: 139 Tests: 12 Failed: 0)
  Non-zero wait status: 139

so probably unclean shutdown of the complete test module so that we don't end up with 0/1.

https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14220387974/job/39846683540#step:6:23820 for 5.20 shows

## count-fail-ratio: Run: 876. Fails: 26. Fail ratio 2.96±1.12%

all same failure as above.

5.26 also stable https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14220387974/job/39846683574#step:6:27012

## count-fail-ratio: Run: 1000. Fails: 0. Fail ratio 0±0%. No fails, computed failure probability < .30%

same for 5.30 https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14220387974/job/39846683578#step:6:27012

5.34 showed "## count-fail-ratio: Run: 1000. Fails: 2. Fail ratio .20±.27%", same for 5.38, same failures as above. 5.40 stable, "latest" stable.

So it's surprising that 5.20 shows such significant problems. Retriggering all runs again to crosscheck: https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14220387974

Surprisingly again 5.16 stable https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14220387974/job/39864871863#step:6:27012, 5.20 24 fails with "(Wstat: 139 Tests: 12 Failed: 0)" https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14220387974/job/39864871854#step:6:27156, 5.26 stable https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14220387974/job/39864871866, 5.30 stable https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14220387974/job/39864871865#step:6:27012, 5.34 3 fails with "(Wstat: 139 Tests: 12 Failed: 0)" https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14220387974/job/39864871868, 5.38 one fail with "(Wstat: 139 (Signal: SEGV, dumped core) Tests: 12 Failed: 0)" https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14220387974/job/39864871873, 5.40 stable https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14220387974/job/39864871882#step:6:27012, latest stable https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14220387974/job/39864871886#step:6:27012 so quite different to above so the Perl version does not seem to have an impact but likely the load on the node running tests, e.g. less load likely producing no failures and higher load more likely to cause problems.

Code "139" means segmentation fault so I assume either in this module or any other module we use unclean memory handling during termination.

Now running again rebased after tinita's latest changes as well as with while (my $pid = waitpid(-1, WNOHANG) > 0) { … as well as w/o coverage collection: https://github.com/okurz/Mojo-IOLoop-ReadWriteProcess/actions/runs/14252642847

Moving segfault fix to #180026 . For current ticket we should be ok to live with a workaround of retrying: https://github.com/openSUSE/Mojo-IOLoop-ReadWriteProcess/pull/79