Ævar Arnfjörð Bjarmason [off-list ref] writes:
[I didn't find an ideal message to reply to in this thread, but this
seemed to probably be the best]
Just an update on this since I went back and looked at this thread,
GitLab about ~1yr ago turned on core.fsyncObjectFiles=true by
default.
The reason is detailed in [1], tl;dr: empty loose object file issue on
ext4 allegedly caused by a lack of core.fsyncObjectFiles=true, but I
didn't do any root cause analysis. Just noting it here for for future
reference.
1. https://gitlab.com/gitlab-org/gitlab-foss/-/issues/51680#note_180508774
Thanks for bringing the original discussion back. I do recall the
discussion and the list of issues to think about raised by Peff back
then in [2] is still relevant.
2. https://public-inbox.org/git/20180117205509.GA14828@sigill.intra.peff.net/