search

EESSI / software-layer

Software layer of the EESSI project
https://eessi.github.io/docs/software_layer
GNU General Public License v2.0
24 stars 49 forks source link

Skip netCDF v4.9.2 tests on ARM #425

Open Neves-P opened 11 months ago

Neves-P commented 11 months ago

When trying to build OpenFOAM, which depends on netCDF (https://github.com/EESSI/software-layer/pull/404) the build of netCDF v4.9.2 hangs on two tests, resulting in a timeout and test failures:

163 - nc_test4_run_par_test (Timeout)
190 - h5_test_run_par_tests (Timeout)

The issue seen here is similar to other test hangs on prior versions of netCDF (see https://github.com/easybuilders/easybuild-easyconfigs/issues/15959). This lead to offending tests being skipped in the easyconfigs through patches and preconfigopts here https://github.com/easybuilders/easybuild-easyconfigs/pull/17107#issuecomment-1448071918 and here https://github.com/jfgrimm/easybuild-easyconfigs/commit/96598a2ee5a1fae71e8fd06ffc8279ca34f931c0). The tests failing now don't seem to have been caused issues in the EasyBuild test clusters and we've only been able to reproduce the issue so far in build jobs with the bot (https://github.com/EESSI/software-layer/pull/404#issuecomment-1836162681) and not interactively (https://github.com/EESSI/software-layer/pull/404#issuecomment-1836439946).

Since, as far as we can see, the problem only happens when building for ARM on EESSI bot jobs, the way to go is skipping the offending tests through a pre test hook on eb_hooks.py.

Info also in the EESSI support portal.

Neves-P commented 11 months ago

I split #404 into smaller PRs for some dependencies, and weirdly enough, in the one for netCDF (#431) the tests passed without issue for neoverse_v1 without the hook (which I'd forgotten to add there https://github.com/EESSI/software-layer/pull/431#issuecomment-1862647169). Hard to say what happened, but I agree with @ocaisa's suggestion on the same thread to merge that PR as skipping the tests shouldn't affect functionality anyway.