search

gridcf / gct

Grid Community Toolkit
Apache License 2.0
48 stars 30 forks source link

Enable additional CI build targets #188

Closed fscheiner closed 2 years ago

fscheiner commented 2 years ago
fscheiner commented 2 years ago

New target OSes are limited to GHA and the x86_64 arch. Depending on availability in the future this can be extended to the ppc64le (and maybe also to the arm64, but not used yet!) arch(es) used with Travis CI.

fscheiner commented 2 years ago

The Travis-CI build for this PR failed because it was automatically cancelled after a runtime of more than 50 minutes. But the corresponding Travis-CI build in my fork for the pushed branch that is the base for this PR succeeded.

Hence the checks for this PR can be considered successful.

fscheiner commented 2 years ago

@ellert, @msalle:

I requested a re-run of the failed test, and it succeeded in 49m 11s (Which is less than 50 minutes, but not by much.)

Indeed it's a very tight margin to failure here. I'll look into further splitting up the build jobs for Travis CI to see if this makes a difference.

fscheiner commented 2 years ago

@ellert, @msalle: I just noticed two interesting constellations at Travis CI:

  1. https://app.travis-ci.com/github/gridcf/gct/builds/249689107
  2. https://app.travis-ci.com/github/fscheiner/gct/builds/249689898

...where for 1) the two different build jobs started at different times. The second one about ~10~ 8 mins later than the first one. Still the total runtime equals to the runtime of the second job, so is lower (!) than the runtime of the first job.

For 2) also both build jobs started at different times, the second one about 7 minutes later than the first one. But here, the total runtime equals to the the runtime of the first job for an unknown reason. So if the second job runs as long as the first one of 1) which had a runtime of about 45 minutes), the whole build will most likely fail due to the 50 minute time limit, because the second job is already 7 minutes late.