Warning problem with Python

[leacmichel]

Hi,
We just installed the GA4HPC on the HPC of our institute. The code worked perfectly when I tried it on my cluster (I use mostly R), but it created a warning for one of my colleagues that uses Python.
We're unsure how to solve this.

[camilo1729]

Hello,

I dont know if you figure it out, but we have the same problem in our cluster. This happens with jobs that took less than 1 sec. We solved like this:

        if x.TotalCPUtime_.total_seconds() > 1:
            # The total CPU time of the job may exceed the job's elapsed time for jobs that include multiple job steps.
            # https://slurm.schedmd.com/sacct.html
            assert x.TotalCPUtime_ <= x.CPUwallclocktime_

You need to put that on file backend/slurm_extract.py

[avchavanne]

It seems the issue in our cluster (with @leacmichel) was accounting for the runtime of multi-core jobs. In the end we solved it this way in the slurm_extract.py :

def calc_CPUusage2use(self, x): 
    max_cpu_time = x.CPUwallclocktime_ * x.NCPUS_ 
    if max_cpu_time.total_seconds() == 0: 
        return max_cpu_time
    cpu_time = min(x.TotalCPUtime_, max_cpu_time) 
    return pd.to_timedelta(cpu_time, unit="s")