[Gluster-users] Gluster v 3.3 with KVM and High Availability

[Mark Nipper]

On 11 Jul 2012, Brian Candler wrote:
> > 	But changing this to writeback or writethrough is still
> > considered acceptable and safe, correct?
> 
> cache=writethrough is the default (slow) mode, and is safe.
> 
> cache=writeback is is positively unsafe, and only acceptable if you don't
> care about possible data loss.
> 
> http://www.linux-kvm.org/wiki/images/5/59/Kvm-forum-2011-performance-improvements-optimizations-D.pdf
> "Cache=writeback - Not supported"

	Okay, good to know.  Writethrough is what we ended up
using (it seemed like the more appropriate choice based on the
concept of a write through cache in general, i.e. making sure
stuff is in fact written to disk before moving on).  But I wanted
to ask because one of our VM's did experience some disk
corruption (out of the 17 we have running on Gluster at the
moment) when we first moved it over.  But I've since rebuilt that
machine's QCOW2 disk image from scratch and it hasn't happened
there or on any other VM since, so the assumption at this point
is that there was a problem with the original iSCSI image (which
had been running on a DRBD/HA Linux cluster setup) we converted
from initially using qemu-img.

> > But my understanding from reading previous posts on this
> > list is that using something other than a cache mode of none is
> > acceptable and safe with Gluster at least.
> 
> cache=none is definitely what we want, but doesn't currently work with
> glusterfs.

	Would that be using something like O_DIRECT which FUSE
doesn't support at the moment?

-- 
Mark Nipper
nipsy at bitgnome.net (XMPP)
+1 979 575 3193
-
"We enter into life naked and howling, covered with blood.  The
fun doesn't have to end there".
 -- _The Book of Madness_