Improve check of memory usage in logParser
The current check for the memory usage is using with last -evt vmem
, which should give an estimate for VmPeak
.
Warnings are thrown when the memory usage is > 4 GB and < 8 GB and an error for > 8GB.
A job launched here: https://bigpanda.cern.ch/job?pandaid=6173479916 succeeded, while logParser would complain because the VM usage is > 8 GB.
From https://iopscience.iop.org/article/10.1088/1742-6596/898/5/052004/pdf and https://indico.cern.ch/event/440821/contributions/1931073/attachments/1218986/1781069/jambo.pdf it is not completely clear to me whether with Prodsys2 we need this test at all.
The questions are:
- do we need to monitor the memory usage of test jobs before submiting production jobs to the grid? (or does the pilot do that and then select sites appropriately?)
- if the answer to the above is yes, which metric should we use?
VmPeak
to my understanding is RAM+swap, but perhaps we should be monitoring only the RAM usage. In that case maxPSS
would seem the best metric but this is not reported in the log. In the log we have VmRSS
which for the job above is not far from maxPSS
but for other cases it might be?