[test] Add MPI version of the eat memory check

[jgphpc]

thanks @ekouts

[codecov-io]

Codecov Report

Merging #1613 (5ec66bc) into master (26da6e4) will decrease coverage by 0.00%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #1613      +/-   ##
==========================================
- Coverage   87.54%   87.54%   -0.01%     
==========================================
  Files          44       44              
  Lines        7292     7288       -4     
==========================================
- Hits         6384     6380       -4     
  Misses        908      908              

Impacted Files	Coverage Δ
reframe/core/config.py	90.90% <0.00%> (-0.10%)	⬇️
reframe/frontend/cli.py	74.94% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 26da6e4...5ec66bc. Read the comment docs.

[ekouts]

Even though none of the tests fail there is an error printed in the end:

[  PASSED  ] Ran 15 test case(s) from 10 check(s) (0 failure(s))
[==========] Finished on Fri Nov 20 15:11:26 2020 
./bin/reframe: rfm_MemoryMpiCheck_job.out: No such file or directory

It is coming from these lines, but I am not sure how you should write the test to avoid it

        regex_mem = r'^Currently avail memory: (\d+)'
        self.reference_meminfo = \
            sn.extractsingle(regex_mem, self.stdout, 1,
                             conv=lambda x: int(int(x) / 1024**3))

Also this part doesn't need to be in a hook, since you use the deferable, right?

[jgphpc]

👍 I moved it out of the hook.
I still get: rfm_MemoryMpiCheck_job.out: No such file or directory

[vkarak]

What I cannot understand very well is the value of this test. Why running the single node eat memory test on multiple nodes wouldn't do the same?

[jgphpc]

value of this test

It's an extension of the MemoryOverconsumptionCheck:

it tests a slurm flag with a fixed value:
https://github.com/eth-cscs/reframe/blob/master/cscs-checks/system/slurm/slurm.py#L189
I am trying to test values closer to the node limit which I guess is more flexible.
Also, running on all cores is faster (not a huge difference) + some performance reporting.

[vkarak]

@jgphpc I did some minor enhancements to your test, but on Dom it fails to meet some references. Can you check? And then we can merge.

Changes done and we need to merge it today.