eth-cscs / DLA-Future

DLA-Future
https://eth-cscs.github.io/DLA-Future/master/
BSD 3-Clause "New" or "Revised" License
64 stars 14 forks source link

Use pika 0.22.2 in CI #1092

Closed msimberg closed 7 months ago

msimberg commented 8 months ago

I believe I've fixed the issue in the change in 0.22.0 that was reverted in 0.22.1 (https://github.com/pika-org/pika/pull/872). Before running CI with the fix I want to verify that I can still trigger the failures with 0.22.0. I'll be updating this PR along with changes to pika as I get closer to pika 0.22.2. pika 0.22.2 doesn't exist yet.

msimberg commented 8 months ago

cscs-ci run

msimberg commented 8 months ago

cscs-ci run

msimberg commented 8 months ago

cscs-ci run

msimberg commented 8 months ago

This seems to be working well now. The CI failure is again a timeout at job exit: https://gitlab.com/cscs-ci/ci-testing/webhook-ci/mirrors/4700071344751697/7514005670787789/-/jobs/6110897893.

msimberg commented 8 months ago

cscs-ci run

msimberg commented 8 months ago

cscs-ci run

msimberg commented 8 months ago

cscs-ci run

msimberg commented 8 months ago

It seems ironically like https://github.com/spack/spack/pull/42717/commits/0da5f1beba1237a779569bd37b1d19f7da237f96 is the commit that changes how intel-oneapi-mkl is concretized, i.e. it leads to intel-oneapi-mkl threads=tbb with a soft preference in the environment. I've changed it to a hard requirement in the common environment and override it with :: where overriding is required.

I've asked about this on the spack slack and will report back if we have alternative ways of solving this or if this is a bug, but otherwise this is ready for review. Unless there are other suggestions for workarounds I'd stick with this until we know better.

msimberg commented 8 months ago

The behaviour of not following the threads=openmp is apparently somewhat expected. I've opened an issue https://github.com/spack/spack/issues/42764 after some discussion with the spack people: https://spackpm.slack.com/archives/CMHK7MF51/p1708429652825989. A change may be possible but the two notes at the bottom of this section seem to prevent threads=openmp from being used: https://spack.readthedocs.io/en/latest/packages_yaml.html#requirements-syntax.