Sync meeting on EESSI test suite (2024 10 10)

Meeting (2024-10-10)

Attending:

• Merged PRs

  • Added Mixin class #177
  • Lower configured memory for Azure zen4 nodes #186
  • Enforce use of zen4 stack on Snellius zen4 nodes #187 and #188
  • Release v0.4.0 #189
  • Update lammps hook name #191

• Open PRs

  • MetalWalls:
    • Tested on Snellius, all 1_node runs pass now
    • Now skips correctly, message is a bit misleading though ("Number of tasks per node 384 exceeds maximum core count 256 for hackathonGPU/benchmark6"). This should probably be something like "Number of tasks [XYZ] exceeds maximum task count (256) for [benchmark_name]"
    • It seems to report a correct total_elapsed_time performance metric, but extract_time is always 0s. Bug? Or is that timing something that is just really fast?
  • Automatic versioning:
    • Comment from Sam: it's a lot of code for automatic versioning. Might be a maintenance burden
    • Caspar: most of that is CI checks, that I think won't need much maintenance. The code itself is relatively minimal. Complexity comes from the fact that we a) still support python 3.6 and b) want to have correct versions in all scenarios: installed package from pipy, installed directly from release tarball, or running directly from git clone. The CI does prove that all of these work :)
    • If it becomes to troublesome, it's easy to take it out again.

• Need docs on new mixin class (Caspar)

  • For now, the lammps test is the example, it has been adjusted to use the eessi_mixin class.

• Apply memory limits using memory hook for all tests

  • Much easier with the mixin class - it can print memory simply by setting a class variable

• Port all tests to use the mixin class