[Gluster-devel] Persistent AFR changelog attributes

[Pranith Kumar Karampuri]

----- Original Message -----
> From: "Pranith Kumar Karampuri" <pkarampu at redhat.com>
> To: "Anand Avati" <avati at gluster.org>
> Cc: "Ravishankar N" <ravishankar at redhat.com>, "Gluster Devel" <gluster-devel at nongnu.org>
> Sent: Wednesday, February 12, 2014 11:16:57 AM
> Subject: Re: [Gluster-devel] Persistent AFR changelog attributes
> 
> 
> 
> ----- Original Message -----
> > From: "Anand Avati" <avati at gluster.org>
> > To: "Ravishankar N" <ravishankar at redhat.com>
> > Cc: "Gluster Devel" <gluster-devel at nongnu.org>
> > Sent: Wednesday, February 12, 2014 10:30:29 AM
> > Subject: Re: [Gluster-devel] Persistent AFR changelog attributes
> > 
> > Ravi,
> > We had earlier discussed a solution for this (sometime last year) by making
> > volgen remember xlator names and not reassign them. Copying KP with who I
> > had discussed this to quite a level of detail. Have you guys spoken about
> > this? IIRC the solution KP and I discussed was more generic and could also
> > support retaining user made changes/customizations to the volfiles.
> 
> How will new machines that are joining the cluster will know about this
> modified graph? One way to achieve it is to send the skeleton structure of
> the graph to other machines. I wonder how this will address snapshot
> volumes' graph. May be even in the case of snapshot volumes, the skeleton
> has to be copied over to snapshot volume. So instead of storing just the
> client xlator names, the generic solution will have to store the entire
> skeleton and should keep it in sync at the time of probe, snapshot. Is that
> the rough algo you guys discussed? Did I miss anything?

If this indeed is the approach, I don't see an easy way out for generating brick volfiles according to versions. Brick processes don't have graph switching capability yet. Because of this, new versions may need to generate new xlators (Ex: barrier xlator that is going to come in for 3.6) in new versions but not for old versions. In essence the skeleton that is stored for one version could be incompatible for other versions. Then we will have to start having some sort of conversion mechanisms of the skeletons from one version to other. It seems a bit complicated :-(. Any suggestions?

> 
> Pranith
> 
> > 
> > Thanks,
> > Avati
> > 
> > 
> > On Tue, Feb 11, 2014 at 8:33 PM, Ravishankar N < ravishankar at redhat.com >
> > wrote:
> > 
> > 
> > Hello,
> > 
> > As you might perhaps be aware, AFR translator currently uses the client
> > translator names as the name of it's changelog extended attributes.
> > 
> > i)This can be a problem when a remove-brick operation is performed when
> > there
> > are pending heals happening because remove-brick causes a graph change
> > wherein the client translator names become different.
> > 
> > ii) Also, for gluster snapshot volumes to work correctly, there needs to be
> > a
> > persistent mapping of AFR changelog attributes to the bricks.
> > 
> > After some internal discussions, we have come up with a new naming
> > mechanism
> > that ensures backward compatibility. Details on the aforementioned problems
> > and the proposed solution are detailed in a feature page [1] for GlusterFS
> > 3.6.
> > Please feel free to go through it and give your comments/ critiques.
> > 
> > Thanks and regards,
> > Ravi
> > 
> > [1] https://www.gluster.org/ community/documentation/index.
> > php/Features/persistent-AFR- changelog-xattributes
> > 
> > ______________________________ _________________
> > Gluster-devel mailing list
> > Gluster-devel at nongnu.org
> > https://lists.nongnu.org/ mailman/listinfo/gluster-devel
> > 
> > 
> > _______________________________________________
> > Gluster-devel mailing list
> > Gluster-devel at nongnu.org
> > https://lists.nongnu.org/mailman/listinfo/gluster-devel
> > 
>