ZFS HA cluster option
236 views
Skip to first unread message
adrian olson
Aug 19, 2016, 4:06:33 PM8/19/16
to esos-users
Marc,
I know you configured a cluster simular to this using the LSI syncro raid controllers. This may be interesting peice to look at as another HA option for this wonderful piece of software.
Adrian
Marc Smith
Aug 20, 2016, 12:30:10 PM8/20/16
to esos-...@googlegroups.com
Hi Adrian,
So, the problem is ZFS is not cluster-aware -- as far as I know, a ZFS
storage pool can not be activated/available on two nodes
simultaneously. In that article, it sounds as though they are doing
NFS and its an active/passive setup (one node is doing everything, the
other node is dormant waiting for a fail-over). The current designs I
use with ESOS and SCST are configured so storage from both nodes can
be "seen" by the initiators, and SCSI commands can be accepted at
either node (although we use ALUA to typically only allow I/O to one
of the nodes at a time).
I believe there are others that may have done SCST configurations
where no paths are visible on the second node until a fail-over
occurs. In my personal experience, this has not worked nearly as well
as the current model I use.
There are other options out there... if you just wanted to do RAID0,
LVM via clvmd can do stripped and linear sets. And now they have
md-cluster which supports RAID1... I suppose you could actually
combine md-cluster with clvmd to get RAID10 (stripped VG across RAID1
MD sets).
I believe there is a commercial vendor that provides a variant or
modification to ZFS so it is cluster-aware and can be used
concurrently from multiple hosts. Can't recall the name right now, but
it would be interesting to hear the pricing on that and combine it
with ESOS.
--Marc
> --
> You received this message because you are subscribed to the Google Groups
> "esos-users" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to esos-users+...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.
So, the problem is ZFS is not cluster-aware -- as far as I know, a ZFS
storage pool can not be activated/available on two nodes
simultaneously. In that article, it sounds as though they are doing
NFS and its an active/passive setup (one node is doing everything, the
other node is dormant waiting for a fail-over). The current designs I
use with ESOS and SCST are configured so storage from both nodes can
be "seen" by the initiators, and SCSI commands can be accepted at
either node (although we use ALUA to typically only allow I/O to one
of the nodes at a time).
I believe there are others that may have done SCST configurations
where no paths are visible on the second node until a fail-over
occurs. In my personal experience, this has not worked nearly as well
as the current model I use.
There are other options out there... if you just wanted to do RAID0,
LVM via clvmd can do stripped and linear sets. And now they have
md-cluster which supports RAID1... I suppose you could actually
combine md-cluster with clvmd to get RAID10 (stripped VG across RAID1
MD sets).
I believe there is a commercial vendor that provides a variant or
modification to ZFS so it is cluster-aware and can be used
concurrently from multiple hosts. Can't recall the name right now, but
it would be interesting to hear the pricing on that and combine it
with ESOS.
--Marc
> You received this message because you are subscribed to the Google Groups
> "esos-users" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to esos-users+...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.
Reply all
Reply to author
Forward
0 new messages