[DRBD-user] Three-way replication setup
lars.ellenberg at linbit.com
Wed Jun 6 11:04:50 CEST 2012
On Tue, Jun 05, 2012 at 10:14:00PM +0200, Arnold Krille wrote:
> On 05.06.2012 21:06, Florian Haas wrote:
> > On Tue, Jun 5, 2012 at 8:34 PM, Arnold Krille <arnold at arnoldarts.de> wrote:
> >> Luca Fornasari <luca.fornasari at upprovider.it> schrieb:
> >>> I have already setup a two node HA Proxmox cluster configuring as per
> >>> <http://pve.proxmox.com/wiki/Two-Node_High_Availability_Cluster>
> >>> http://pve.proxmox.com/wiki/DRBD
> >>> So I have a primary/primary DRBD resource (yes I configured fencing
> >>> correctly).
> >> I believe that three nodes including dual-primary is probably a nightmare to setup and an even worse nightmare configuring a cluster to automate it...
> > It's not a nightmare. It's just not possible. You can sleep soundly. :)
> >> As far as I was told, linbit is working on three-way drbd-replication on a single-layer. But there are other systems both on the block- and on the fs-layer that do n-way replication...
> > Yup. Well actually I'm not aware of anything that does n-way
> > multi-master at the block level, although you could say that RBD does
> > (not exactly correct, but close enough). Multi-master n-way filesystem
> > replication is available in GlusterFS.
> I would count rbd of ceph and also sheepdog as n-way replication at the
> block-level. Although its the block-devices of virtual machines. And who
> needs block-level replication for something else then VMs anyway? :-)
> I tried HA-nfs with drbd below but went back to just providing drbd to a
> vm for the nfs and (re)start that vm from corosync/pacemaker. Its a
> bigger problem with samba as windows doesn't seem to like it when the
> server crashes and restarts, probably because some packets in between
> are answered with 'rejected'. nfs and smtp/imap are far more tolerant.
> But somehow I fear my eviction from this list when I always answer posts
> with a list of alternatives... Hint to the list-maintainers: I do have
> drbd running in production, I do not (yet) have one of the other in
DRBD is a tool.
It is the right tool for some jobs.
It is the wrong tool for others.
Rather use some "alternative" that better suits your use case,
than trying to force-fit DRBD, just because it works well in some
"similar" use case.
That would only cause frustration for those who try to force-fit DRBD,
they then tell all who would listen that DRBD was a bad idea, period,
just because even after applying force it did not quite fit...
If you follow me on that, then pointing out other tools
that may better fit the specific use case even helps DRBD.
So please keep it coming ;-)
: Lars Ellenberg
: LINBIT | Your Way to High Availability
: DRBD/HA support and consulting http://www.linbit.com
More information about the drbd-user