Fedora 33: reduce concurrency

... to avoid OOM. E.g.:

as: out of memory allocating 4064 bytes after a total of 396447744 bytes

marked this merge request as ready

assigned to @jbrost and unassigned @aklimov

Why only for Fedora 33? How do our runners work? Do jobs have a guaranteed amount of RAM (i.e. with only on job per VM)? Or are there multiple concurrent jobs per VM that have to share the memory and Fedora 33 just had bad luck in one run?

What is the default of RPM_BUILD_NCPUS? Number of cores? For the two Fedora jobs I compared, the build time only increased slightly from 25:22 minutes to 26:55, so are we flooding the runners with too many concurrent jobs at the moment which doesn't even give us some real speedup?

assigned to @aklimov and unassigned @jbrost

Because only it has the problem. I think we have one VM per job. No, this problem is systematic, just have a look at the latest master pipeline.

Yes, number of cores. Yes, you seem to be right.

Okay, pushed retry and it also got killed when only this one job was running. Do we have any memory statistics for the runners? Like are all jobs using most of the memory already and some change (new compiler version requiring a bit more memory or something like that) pushed it over the limit?

approved this merge request

I guess no.

merged

mentioned in commit eec09614