Gluster users - Oct 2015 - New user, couple of questions (sharding+discard, arbiters, shutting down nodes)

I have a couple of 2-node clusters I'm hoping to move from drbd+ocfs2 to 
glusterfs. I've been testing with 3.7.4 and I have a few questions.

1) I gather that 2-node replicas have quorum issues, and if you disable 
quorum, then they have split-brain issues. Do split-brains happen even if the 
network link is very reliable - I have point-to-point 10Gbit, and the only 
clients will be the brick servers? If so would it make more sense to make this 
a 4-node cluster and use arbiter volumes?

2) Sharding looks pretty good, as the volumes will be used exclusively for 
large VM backing images. I've been testing with 1GB shard sizes and 
performance seems good. It seems that sharded volumes don't support discard,
though (ie. fstrim within VM guests). Is there a timeline on when that might 
be implemented?. Discard seems to work right on non-sharded volumes, but then 
heal times seem like they'll be an issue.

3) Is there operational documentation for maintenance procedures, like how to 
properly shut down nodes in a way that won't impact clients?  What I got
from
a recent mailing list post suggests something like:

killall glusterfs
killall glusterfsd
systemctl stop glusterd 

That seems to work in testing, the client VM's stay responsive. Is it safe?

Thanks in advance for any advice.

I've had zero issues using client quorum (cluster.quorum-type
=auto) and three nodes/bricks. Testing using node shutdown and node kills.

I'd really recommend going with three nodes - maybe one as a arbiter volume
if disk space is an issue.

On 29 October 2015 at 03:52, Alan Hodgson <ahodgson at simkin.ca> wrote:
> I have a couple of 2-node clusters I'm hoping to move from drbd+ocfs2
to
> glusterfs. I've been testing with 3.7.4 and I have a few questions.
>
> 1) I gather that 2-node replicas have quorum issues, and if you disable
> quorum, then they have split-brain issues. Do split-brains happen even if
> the
> network link is very reliable - I have point-to-point 10Gbit, and the only
> clients will be the brick servers? If so would it make more sense to make
> this
> a 4-node cluster and use arbiter volumes?
>
> 2) Sharding looks pretty good, as the volumes will be used exclusively for
> large VM backing images. I've been testing with 1GB shard sizes and
> performance seems good. It seems that sharded volumes don't support
> discard,
> though (ie. fstrim within VM guests). Is there a timeline on when that
> might
> be implemented?. Discard seems to work right on non-sharded volumes, but
> then
> heal times seem like they'll be an issue.
>
> 3) Is there operational documentation for maintenance procedures, like how
> to
> properly shut down nodes in a way that won't impact clients?  What I
got
> from
> a recent mailing list post suggests something like:
>
> killall glusterfs
> killall glusterfsd
> systemctl stop glusterd
>
> That seems to work in testing, the client VM's stay responsive. Is it
safe?
>
> Thanks in advance for any advice.
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-users
>


-- 
Lindsay
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://www.gluster.org/pipermail/gluster-users/attachments/20151029/63f3c4a7/attachment.html>

----- Original Message -----
> From: "Alan Hodgson" <ahodgson at simkin.ca>
> To: gluster-users at gluster.org
> Sent: Wednesday, October 28, 2015 11:22:24 PM
> Subject: [Gluster-users] New user, couple of questions (sharding+discard,
> arbiters, shutting down nodes)
> I have a couple of 2-node clusters I'm hoping to move from drbd+ocfs2
to
> glusterfs. I've been testing with 3.7.4 and I have a few questions.
> 1) I gather that 2-node replicas have quorum issues, and if you disable
> quorum, then they have split-brain issues. Do split-brains happen even if
the
> network link is very reliable - I have point-to-point 10Gbit, and the only
> clients will be the brick servers? If so would it make more sense to make
> this
> a 4-node cluster and use arbiter volumes?
> 2) Sharding looks pretty good, as the volumes will be used exclusively for
> large VM backing images. I've been testing with 1GB shard sizes and
> performance seems good. It seems that sharded volumes don't support
discard,
> though (ie. fstrim within VM guests). Is there a timeline on when that
might
> be implemented?. Discard seems to work right on non-sharded volumes, but
then
> heal times seem like they'll be an issue.
Thanks for the feedback, Alan. 

So 3.7.6 release is just two days away, and therefore it will not be possible to
get discard implementation
in by then. You can expect it to be available for 3.7.7 though (which would be
end of Nov).
Meanwhile, you can track its progress at
https://bugzilla.redhat.com/show_bug.cgi?id=1261841 .

-Krutika 
> 3) Is there operational documentation for maintenance procedures, like how
to
> properly shut down nodes in a way that won't impact clients? What I got
from
> a recent mailing list post suggests something like:
> killall glusterfs
> killall glusterfsd
> systemctl stop glusterd
> That seems to work in testing, the client VM's stay responsive. Is it
safe?
> Thanks in advance for any advice.
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://www.gluster.org/pipermail/gluster-users/attachments/20151028/b3cdbf75/attachment.html>