We are seeing some unexpected behavior with git that we'd like to better
understand. We are running git 18.104.22.168 on the remote repo server
(CentOS 6.3-x86) and the clients (mostly Windows 7), and we are running
gitolite 3.04 on the server.
At times, our repos can get many unreachable loose objects, and if we
don't clean them up, we will see instances where a push to the remote
repo does not work correctly. This almost always occurs when we try to
delete a remote branch with a "git push origin :branch" call, but this
week we saw two occurrences where a push failed to create a branch (as
well as about a dozen cases where a push failed to delete the branch).
This behavior does not occur for every push, however: I'd guess 10% of
Here is output from a user's machine:
d:\git\jedi\jedifw>git push origin
Counting objects: 59, done.
Delta compression using up to 16 threads.
Compressing objects: 100% (29/29), done.
Writing objects: 100% (30/30), 129.69 KiB, done.
Total 30 (delta 24), reused 1 (delta 0)
Auto packing the repository for optimum performance.
warning: There are too many unreachable loose objects; run 'git prune'
to remove them.
After a couple of hours, that user was concerned that their branch
hadn't started processing, so they sent an email to support and the
re-pushed. Here are the lines from the gitolite log showing both of
gitolite-2013-01-07.log:2013-01-07.15:57:58 23285 update
jedi/jedifw user.n...@hp.com W
gitolite-2013-01-07.log:2013-01-07.19:57:16 30374 update
jedi/jedifw user.n...@hp.com W
There are no log entries for that branch or that SHA in between those
two lines. My understanding is that both of those lines show a new
branch being created on the remote repo. I'm not sure what happened to
the first push, though: I expected that the branch would have been
created at that time, and subsequent fetches from the remote repo would
show the branch. However, fetches from the remote repo did not show the
branch until after the second push.
We've been running git for about 3 years, and this happened occasionally
when we first started with git, but we always found it was related to a
huge number of unreachable loose objects, which triggered an auto-gc on
the remote repo. When we performed manual gc --aggressive and prune
operations on the remote repo, the problem stopped. It happened this
week because we'd deleted over 4,000 refs/topics branches on
Friday/Saturday, which left a huge number of unreachable loose objects.
Our cleanup script was only pruning objects older than 2.weeks.ago,
and when I pruned objects older than 2.days.ago the problem again stopped.
Is this expected behavior for the interaction between pushes and auto-gc
on the remote repo?
Also, I went through the release notes for the latest versions of git,
and I found this for 22.214.171.124:
* When "git push" triggered the automatic gc on the receiving end, a
message from "git prune" that said it was removing cruft leaked to
the standard output, breaking the communication protocol.
Could that be related to what we're seeing?
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html