← Back to team overview

holland-coredev team mailing list archive

Re: Holland iterative purge

 

Sounds clever to me! Also, I haven't forgotten about getting WordPress up and running - just haven't had time. Should have more time either this week or next.

On May 27, 2011, at 5:57 PM, Andrew Garner wrote:

> I pushed a tentative change to holland-backup/master earlier this week
> to improve purge backups on ext3.   This does a gradual ftruncate() on
> files before doing an unlink().   This was done to address some
> performance issues when holland has large files (particularly large
> tarballs) in $backup_directory and simply unlink()ing those on ext3
> causes a huge io spike and often io stalls, particularly if something
> like mysql, happens to be doing synchronous io on the same filesystem.
> 
> This is essentially the same issue discussed here:
> 
> http://www.mysqlperformanceblog.com/2009/06/16/slow-drop-table/
> 
> Although not directly related to MySQL in this case. :)   XFS
> significantly improves this but you can still see stalls of a few
> seconds (but much less dramatic than the 2+ minutes of ext3), so I
> think the approach is generally useful in unix environments.
> 
> This change came in here:
> 
> https://github.com/holland-backup/holland/commit/8010ab72525e1be02f14c4f39a1f06083fb3d425
> 
> I am throwing it out here in case anyone has any comments or concerns
> with this approach.
> 
> ~Andy
> 
> _______________________________________________
> Mailing list: https://launchpad.net/~holland-coredev
> Post to     : holland-coredev@xxxxxxxxxxxxxxxxxxx
> Unsubscribe : https://launchpad.net/~holland-coredev
> More help   : https://help.launchpad.net/ListHelp
> 



References