Skip to content

Conversation

@msimberg
Copy link
Contributor

@msimberg msimberg commented Nov 11, 2025

This is essentially the same as #437, but applies the change to a few more tests. This also abstracts the setting of the appropriate flag into a mixin.

Tested with

UENV=prgenv-gnu/25.11:rc1,prgenv-gnu-openmpi/25.11:2132109439 \
reframe --config config/cscs.py -c checks/prgenv/helloworld.py \
-c checks/microbenchmarks/mpi/osu/osu_run.py \
-c checks/microbenchmarks/xccl/ \
--prgenv '(?!PrgEnv-ce)' --run

on daint. The latter uenv is in the build:: namespace from eth-cscs/alps-uenv#263. The same tests run on clariden, but prgenv-gnu-openmpi isn't deployed on santis, so has to be chosen directly using the squashfs path.

@msimberg
Copy link
Contributor Author

cscs-ci run alps-daint-uenv;MY_UENV=prgenv-gnu/25.11:rc1,build::prgenv-gnu-openmpi/25.11:2132109439

@msimberg
Copy link
Contributor Author

cscs-ci run alps-clariden-uenv;MY_UENV=prgenv-gnu/25.11:rc1

@msimberg
Copy link
Contributor Author

cscs-ci run alps-daint-uenv;MY_UENV=prgenv-gnu/25.11:rc1

@msimberg
Copy link
Contributor Author

NB. some of the osu tests sometimes fail the threshold even with this PR. This PR is only to get the tests actually running correctly to start with. Additionally, not all of the tests listed above are production tests. The hello world tests are, and were failing here for clariden: https://git.cscs.ch/alps/vcluster-validation/-/jobs/4210270.

@jgphpc jgphpc self-requested a review November 12, 2025 07:03
@msimberg msimberg marked this pull request as draft November 18, 2025 20:45
@msimberg
Copy link
Contributor Author

cscs-ci run alps-daint-uenv;MY_UENV=prgenv-gnu/25.11:rc1

@msimberg
Copy link
Contributor Author

cscs-ci run alps-eiger-uenv;MY_UENV=prgenv-gnu/25.11:rc2

@msimberg msimberg marked this pull request as ready for review November 19, 2025 16:43
@msimberg
Copy link
Contributor Author

I've moved the --mpi=pmix for a few CE test into a separate mixin. I don't want to put it into the generic ContainerEngineMixin because 1. not all containers necessarily need --mpi=pmix and with container we don't have a similar system of features as we have for uenvs, 2. many container engine tests don't use MPI so they don't need the flag at all, and 3. I'm unsure if some tests will misbehave if we set --mpi=pmix for all jobs. With all this, I've restricted my changes to the tests that I know relatively well and that are relevant for production.

For uenv we can safely set the --mpi flag because we know that they're using Cray MPICH if there's no feature, and if there is an openmpi/cray-mpich feature we can rely on that.

@msimberg
Copy link
Contributor Author

Also note that on daint some of the CE tests are likely to fail because of the recent upgrade which broke PMIx support. I've run the tests on clariden successfully.

@msimberg
Copy link
Contributor Author

cscs-ci run alps-daint-uenv;MY_UENV=prgenv-gnu/25.11:rc2

@msimberg
Copy link
Contributor Author

cscs-ci run alps-eiger-uenv;MY_UENV=prgenv-gnu/25.11:rc2

@msimberg
Copy link
Contributor Author

cscs-ci run alps-eiger-uenv;MY_UENV=prgenv-gnu/25.11:rc2

@msimberg
Copy link
Contributor Author

cscs-ci run alps-santis-uenv;MY_UENV=prgenv-gnu/25.11:rc2

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant