Re: Repository corruption

From: Yann Dirson <ydirson@altern.org>
Date: Mar 15, 2008, 13:04 UTC
Message-ID: <20080315130441.GF4079@nan92-1-81-57-214-146.fbx.proxad.net>
In-Reply-To: <20080315121903.GY10335@machine.or.cz>

On Sat, Mar 15, 2008 at 01:19:03PM +0100, Petr Baudis wrote:

Show 9 quoted lines

>   Hi,
> 
> On Mon, Sep 10, 2007 at 11:25:50PM +0200, Yann Dirson wrote:
> > Not sure if you saw my note in IRC: it looks like tagua/ydirson.git is
> > definitely corrupted - not sure it would be useful to keep a copy
> > around for investigation, but I'd appreciate if you could nuke the
> > contents, so I can push again to it.
> 
>   I have nuked it now. :-)

Thanks :)

Any idea what caused the corruption ?

BTW - although I have not reported it on the list, I recently had a repo corrupted by a "git gc" with 1.5.4.3. Not sure there would be a link between those 2 events, but who knows...

It may not be easy to investigate, since I do not have an exact copy of it before the corruption (fortunately I was able to salvage most of the work that had no backup copy), since that repo is a private one holding source code I am not allowed to publish, and since I will be leaving in 2 weeks the position that gives me access to that code.

The corruptions were of several kinds:
- commits for unapplied stgit patches were all lost, although they
were reachable from refs/patches/master/*
- seemingly-random objects in the history (commits, trees, probably
blobs as well) were missing

Does anyone have ideas about post-mortem analysis on that repo ?

That tree was managed with stgit, using a rebase/commit workflow, and accumulated objects reported by git-count-objects over the months kept growing, even after "git repack -a" and "git prune", which always looked suspect to me.

Although it is possible that I had already run git-gc on that tree, that would not be recently - it was basically a "let's try it and see if we perform better afterwards".

Best regards,

-- 
Yann