Open boegel opened 1 year ago
A problem arose today when trying to ingest a tarball that included updates hooks (2023.06/init/easybuild/eb_hooks.py
), since that messed up the way in which the CPU architecture is being checked, which is based on this:
tar_first_file=$(tar tf "${tar_file}" | head -n 1)
@bedroge Do you think it's OK if we add grep -v '/init/'
in there, before the head
?
I guess we should also make sure that $tar_first_file
is not empty then...
It would be interesting if the
ingest-tarball.sh
script would support ingesting a combined tarball of software installations for the software layer, in which a small number of installations across all (or at least multiple) CPU targets are included.For
OSU-Micro-Benchmarks
(see PR #201), we had to create and ingest 10 separate small tarballs, which only adds overhead (since for each tarball, a PR to thestaging
repo is created, and needs to be reviewed/merged). The bot could combine all tarballs into a single one when it's instructed to do the deploy.The ingest script currently makes a couple of assumptions that are incorrect for such a combined tarball, including that the tarball only includes software installations for a single CPU target, and that
tar tf ${tarball} | head -1
allows to determine whether it's an ingesting forsoftware
/compat
/init
.