cyrodiilsavior / hpc Goto Github PK

When checking if output files exist please bump the polling interval to at least one second or make it configurable (at

). Reason is that those output files might end up on a parallel filesystem and they usually don't like metadata operations.

And a question regarding the GetOutputSlurm() function: the way this looks to me we wait until there is an output file created by Slurm and if there is, we read it, relay to the gitlab instance and remove it.
What happens in the case where the file is created, but the job is still running and Slurm is still writing output to it? If I understand correctly it will read the output up until this point in time, remove the file and miss the rest, as Slurm does not copy the file over once it is finished, but does incremental writes.
I see two options: either checking for job completion and then reading the output, or relaying the output as it comes and and on job complete remove the file. Thoughts?