On 2010-09-26, at 10:28 AM, Nico Kadel-Garcia wrote:
> On Fri, Sep 24, 2010 at 5:23 PM, Daniel Shahaf <d.s_at_daniel.shahaf.name> wrote:
>> Luis Rojas wrote on Fri, Sep 24, 2010 at 15:16:17 -0400:
>
>>>
>>> We actually suspect the disk, but we don't have enough evidence to
>>> suggest the disk is faulty, we are a small development team, and is
>>> very rarely we submit at the same time, i am not saying it cant happen
>>> but is very unlikely. also, as far as i know, no other process where
>>> accessing the repository at the time.
>>>
>>
>> "No other process" --- did you mean: no other process except for the
>> post-commit hook script?
>
> Could be a broken pre-commit or post-commit script: I've actually seen
> such scripts miswritten and atempt to "rsync -av --delete $srcdir/
> $targetdir/', and forget to set $targetdir. Hilarity ensued.
Could be, but our post-commit is very straight forward, just a simple mantis integration and mail notification. less than 10 lines long and has been working ok for years.
>
> The safe way to rebuild and update a full repository is:
>
> mv $repo $repo.save
> svnadmin hotcopy $repo.save $repo.new
> rsync -av --delete $repo.save/ $repo.new/ ---dry-run # confirm changes, and objects like symlinks, which older Subversion does not handle
> mv $repo.new $repo
>
> That gets your files on an entirely new chunk of disk, which can help
> protect you from disk failure issues.
Thanks for this, i will give this a try, although i thought dump worked better than hot-copy... any reason you prefer doing it with hot-copy ?
Luis R. Rojas
SysAdmin
Rogue Research Inc
T: (514) 284-3888
F: (514) 284-6750
luis_at_rogue-research.com
Received on 2010-09-27 16:10:57 CEST