HPC: Broken/Insufficient Testing of Load Balancer Functionality

[Schlevidon]

Mistakes related to the examples in hpc/test:

• Helix/Makefile is outdated and unusable in its current state.
• MultiplyBy2/client.py sends a request with too many input dimensions and crashes.
• minimal and MultiplyBy2 do almost the same thing on the server side and should be merged into a single example or renamed appropriately.
• Currently only Evaluate requests are tested.

Tasks:

• Clean up and fix the current examples.
• Add tests to ensure coverage of the entire UM-Bridge model interface, i.e
  • GetInputSizes and GetOutputSizes
  • Evaluate, Gradient, ApplyJacobian and ApplyHessian
  • SupportsEvaluate, SupportsGradient, SupportsApplyJacobian and SupportsApplyHessian
• If possible: Add CI to automatically run tests.
  • It should be possible to use HQ without SLURM by using hq worker start instead of automatic allocation.
  • Alternatively, it might be possible to run a virtual SLURM cluster?

[linusseelinger]

We might actually consider removing the examples in hpc entirely. There was a point in having them initially when the hpc code was in a separate repo, but now they are mostly duplicates of the existing (and CI tested) example models.

Running hq without SLURM sounds good for CI!