EESSI test suite sync meetings

Planning

• every 2 weeks on Thursday at 14:00 CE(S)T
• next meetings:
  • 9 January 15:30-16:30 => should work for Sam, Satish, Caspar (not for Lara)

---

Meeting (2024-12-09)

Attending: Sam, Satish, Lara, Caspar

• Merged PRs

  • Collect additional info in mixin #206
  • Enable Vega GPU #215
  • Fix (known) MPI issue (on Azure) with workaround #216

• Open PRs

  • General config and run_reframe.sh for local and EESSI stack #200
    • Lara just processed comments today
  • use mixin class for QuantumESPRESSO #212
    • Needs review
  • use mixin class for PyTorch #213
    • Needs review
  • Add support for setting the exact required memory #214
    • Needs review

• Discussion

  • Should we move configs to a separate repository? I regularly need to update them to implement fixes, add partitions, etc (e.g. #215 and #216)

    • Current issue: either I let daily runs use master for configs anyway, or I'm stuck with broken runs until next release
    • Downside: a test suite will log its version, but not the version of the config file used
    • Conclusion: adapt the run_reframe.sh to also do a sparse checkout of the config files. Then, use those when running reframe.
      • Also: we could add the commit hash as a variable, and set that variable on the command line. (question: can I add a variable on command line if none of the tests 'know' that variable? And is it logged then? => otherwise I'll add it to eessi_mixin)
      • In the config, we could use the python-git to figure out the commit of the current git-tree we are in.

  • API docs: Caspar will probably implement the changes himself

  • Satish will explore Stream

  • To get more adoption we should probably present the test suite on the EB user meeting, and promote it also as something that can test local module stacks

  • Regarding data staging: approach is described here. We'll change eessi_mixin to enforce the users to think about making things `read_only``=> Sam will have a look at this