hi,
I use 5 second to enable gluster-nfs ip shift when nodes poweroff,
but in heavy IO load, client often timeout. maybe need longer timeout.
> The reason for the long (42 second) ping-timeout is because
> re-establishing fd's and locks can be a very expensive operation.
> Allowing a longer time to reestablish connections is logical, unless you
> have servers that frequently die.
>
> If you shut down a server through the normal kill process, the TCP
> connections will be closed properly. The client will be aware that the
> server is going away and there will be no timeout. This allows server
> maintenance without encountering that issue.
>
> One issue with a 42 second timeout is that ext4 may detect an error and
> remount itself read only should that happen while the VM is running. You
> can override this behavior by specifying the mount option,
> "errors=continue" in fstab ("errors=remount-ro" is the default). The
> default can be changed, as well, by changing the superblock option with
> tune2fs.
>
> On 10/22/2013 03:12 AM, John Mark Walker wrote:
> >
> > Hi JC,
> >
> > Yes, the default is a 42-second timeout for failover. You can
> > configure that to be a smaller window.
> >
> > -JM
> >
> > On Oct 22, 2013 10:57 AM, "JC Putter" <jcputter at gmail.com
> > <mailto:jcputter at gmail.com>> wrote:
> >
> > Hi,
> >
> > I am new to GlusterFS, i am trying to accomplish something which i am
> > not 100% sure is the correct use case but hear me out.
> >
> > I want to use GlusterFS to host KVM VM's, from what I've read this was
> > not recommended due to poor write performance however since
> > libgfapi/qemu 1.3 this is now viable ?
> >
> >
> > Currently i'am testing out GlusterFS with two nodes, both running as
> > server and client
> >
> > i have the following Volume:
> >
> > Volume Name: DATA
> > Type: Replicate
> > Volume ID: eaa7746b-a1c1-4959-ad7d-743ac519f86a
> > Status: Started
> > Number of Bricks: 1 x 2 = 2
> > Transport-type: tcp
> > Bricks:
> > Brick1: glusterfs1.example.com:/data
> > Brick2: glusterfs2.example.com:/data
> >
> >
> > and mounting the brick locally on each server as /mnt/gluster,
> > replication works and everything but as soon as i kill one node, the
> > directory /mnt/gluster/ becomes unavailable for 30/40 seconds
> >
> > log shows
> >
> > [2013-10-22 11:55:48.055571] W [socket.c:514:__socket_rwv]
> > 0-DATA-client-0: readv failed (No data available)
> >
> >
> > Thanks in advance!
> > _______________________________________________
> > Gluster-users mailing list
> > Gluster-users at gluster.org <mailto:Gluster-users at gluster.org>
> > http://supercolony.gluster.org/mailman/listinfo/gluster-users
> >
> >
> >
> > _______________________________________________
> > Gluster-users mailing list
> > Gluster-users at gluster.org
> > http://supercolony.gluster.org/mailman/listinfo/gluster-users
>
