[Gluster-users] Shard in production

Krutika Dhananjay kdhananj at redhat.com
Thu Jan 21 04:42:17 UTC 2016


Right. I'd also like to just add that in its current form, sharding does not support zerofill and discard operations and their development is underway. 
As far as fallocate is concerned, it should be possible to use the operation on a sharded volume if it is invoked through posix_fallocate() API. 

Also, in our testing we have found 512MB shard size to provide good performance as far as both IO and self-heal are concerned. I hope that works for you too! 

-Krutika 

----- Original Message -----

> From: "Lindsay Mathieson" <lindsay.mathieson at gmail.com>
> To: "Krutika Dhananjay" <kdhananj at redhat.com>, "Bello Florent"
> <florent.bello at ville-kourou.fr>
> Cc: gluster-users at gluster.org
> Sent: Thursday, January 21, 2016 6:22:09 AM
> Subject: Re: [Gluster-users] Shard in production

> Heh, thanks Krutika.

> I believe its pretty much there, have been stress testing it on our
> production VM cluster for weeks, though not with critical VM's. It is a huge
> improvement when for the heal process when dealing with large VM's. Failure
> modes have been very reliable to.

> Nb. All my testing has been VM's - e.g a few large files with only one
> process accessing them. I can't speak to the many thousands of dirs/files
> use case.

> If you are using it for VM hosting (as you are) then it is essential that you
> set the "virt" group properties, other wise you will end up with VM
> corruption. These are the settings I've evolved over my testing:

> > Options Reconfigured:
> 
> > performance.readdir-ahead: off
> 
> > features.shard: on
> 
> > features.shard-block-size: 512MB
> 
> > cluster.self-heal-window-size: 256
> 
> > server.event-threads: 4
> 
> > client.event-threads: 4
> 
> > cluster.quorum-type: auto
> 
> > cluster.server-quorum-type: server
> 
> > performance.low-prio-threads: 32
> 
> > performance.io-thread-count: 32
> 
> > performance.cache-refresh-timeout: 4
> 
> > nfs.disable: on
> 
> > nfs.addr-namelookup: off
> 
> > nfs.enable-ino32: off
> 
> > performance.write-behind: off
> 
> > performance.strict-write-ordering: on
> 
> > performance.stat-prefetch: off
> 
> > performance.quick-read: off
> 
> > performance.read-ahead: off
> 
> > performance.io-cache: off
> 
> > cluster.eager-lock: enable
> 
> > network.remote-dio: enable
> 
> > cluster.data-self-heal-algorithm: full
> 

> On 21/01/16 02:06, Krutika Dhananjay wrote:

> > Adding Lindsay Mathieson, who has tested sharding with VM store
> > extensively.
> > :)
> 

> > -Krutika
> 
> > ----- Original Message -----
> 

> > > From: "Bello Florent" <florent.bello at ville-kourou.fr>
> > 
> 
> > > To: gluster-users at gluster.org
> > 
> 
> > > Sent: Wednesday, January 20, 2016 8:12:27 PM
> > 
> 
> > > Subject: [Gluster-users] Shard in production
> > 
> 

> > > Hi,
> > 
> 

> > > i see recently the Shard feature is reached to stable. Soon, i want to
> > > install oVirt with Gluster Replica Storage and i guess shard feature is
> > > good
> > > for me. Can i use this feature for my VM Storage production or you guess
> > > it's not mature ?
> > 
> 
> > > _______________________________________________
> > 
> 
> > > Gluster-users mailing list
> > 
> 
> > > Gluster-users at gluster.org
> > 
> 
> > > http://www.gluster.org/mailman/listinfo/gluster-users
> > 
> 

> --
> Lindsay Mathieson
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.gluster.org/pipermail/gluster-users/attachments/20160120/5ac226ec/attachment.html>


More information about the Gluster-users mailing list