[Gluster-devel] AFR bricks - how does it resync question
Anand Avati
avati at zresearch.com
Tue Jul 31 20:18:36 UTC 2007
Hi,
the 'createtime' is only used to detect 'recrection' of files. if an
existing file is edited, the syncing logic maintains its own cvs-like file
versioning in the extended attributes and does not rely upon mtime. we felt
that dependency on mtime is risky, a pain to setup (synchronizing system
clock etc), and also the granularity of mtime is limited.
thanks,
avati
2007/8/1, DeeDee Park <deedee6905 at hotmail.com>:
>
> Thanks that clears up a lot about what is going on. I am not familiar with
> xattr. I tried getfattr -d <somefile> on both the brick, and glusterfs
> client, and nothing showed up.
>
> Does this information also exist in for directories?
>
> Also, I see the algorithm checks for latest createtime. I think something
> is
> missing. I would think that lastmodified time needs to also be kept track
> of to know which brick has the latest version. For Example, if I have
> Brick1
> and Brick2, and they both exist when a file is created, then B2 goes down.
> Let's say that <somefile> got modified in the interim while B2 was down.
> When B2 comes back up, both files will have the same file create time, but
> B1 will have a newer file modified time. I assume under the current
> algorithm, both files on B1 and B2 would be considered as the same file by
> glusterfs. Oh. my bad, I see now that you are using the
> trusted.afr.version
> to keep track of which is modified later. I see some of the tradeoffs
> between lastmodified time vs version. Are these times relative to the
> server
> clock or client clock?
>
> I recall with DFS, that the clock or time management was a huge part of
> the
> design. I'm using a bunch of test boxes right now (in a production
> environment, this might be less of a problem with well managed boxes), but
> sometimes the boxes are off a little on the clocks as until I remember to
> run NTP. Is there a big impact to glusterfs re-syncing if the clocks are
> off?
>
>
> >From: "Krishna Srinivas" <krishna at zresearch.com>
> >To: "DeeDee Park" <deedee6905 at hotmail.com>
> >CC: gluster-devel at nongnu.org
> >Subject: Re: [Gluster-devel] AFR bricks - how does it resync question
> >Date: Tue, 31 Jul 2007 21:02:05 +0530
> >
> >DeeDee,
> >
> >http://lists.gnu.org/archive/html/gluster-devel/2007-07/msg00008.html
> >
> >See if that helps.
> >
> >Regards
> >Krishna
> >
> >On 7/31/07, DeeDee Park <deedee6905 at hotmail.com> wrote:
> > > If I have two bricks B1, and B2, and they are replicas using AFR.
> > > If one of the bricks goes down and has an older copy of a file, and
> then
> > > comes backup,
> > > if a file is being appended, how does the AFR know which copy to
> update?
> > >
> > > _________________________________________________________________
> > > http://newlivehotmail.com
> > >
> > >
> > >
> > > _______________________________________________
> > > Gluster-devel mailing list
> > > Gluster-devel at nongnu.org
> > > http://lists.nongnu.org/mailman/listinfo/gluster-devel
> > >
>
> _________________________________________________________________
> Don't get caught with egg on your face. Play Chicktionary!
> http://club.live.com/chicktionary.aspx?icid=chick_hotmailtextlink2
>
>
>
> _______________________________________________
> Gluster-devel mailing list
> Gluster-devel at nongnu.org
> http://lists.nongnu.org/mailman/listinfo/gluster-devel
>
--
Anand V. Avati
More information about the Gluster-devel
mailing list