[DRBD-user] DRBD primary/primary for KVM - what is the best option?

Bart Coninckx bart.coninckx at telenet.be
Tue May 11 20:45:49 CEST 2010

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


On Tuesday 11 May 2010 19:12:23 Michael Iverson wrote:
> The interesting bit on the first link was the statement:
> 
> "If active-active isn't possible, maybe there is distance involved and it's
> doing asynchronous replication, then you will need to implement something
> like heartbeat to add the volume using ietadm to the running iSCSI target
> once drbd B becomes primary..."
> 
> The quote is a little short on implementation details, but based on
> it, and some snippets
> from the other links, it is the tool to dynamically add or remove
> volumes without messing
> with the remainder of the live volumes.
> 
> The only challenge I see is that any changes that are made are
> dynamic, and would not
> survive a reboot or a daemon restart. So, somehow, upon a restart, the
> ietd daemon needs
> a method to reliably determine which volumes it should or should not
> be serving, or be
> told what to server by heartbeat or the state of the drbd volume.
> 
> On Tue, May 11, 2010 at 12:52 PM, Bart Coninckx
> 
> <bart.coninckx at telenet.be> wrote:
> > On Tuesday 11 May 2010 16:09:08 Michael Iverson wrote:
> >> ietadm is the answer.
> >>
> >> These might help:
> >>
> >> http://old.nabble.com/IET-on-DRBD-howto--td20567810.html
> >> http://www.gossamer-threads.com/lists/linuxha/users/45280
> >> http://www.markround.com/archives/50-Building-a-redundant-iSCSI-and-NFS-
> >>clu ster-with-Debian-Part-4.html
> >>
> >> On Tue, May 11, 2010 at 9:51 AM, Bart Coninckx
> >> <bart.coninckx at telenet.be>
> >
> > wrote:
> >> > On Tuesday 11 May 2010 15:15:34 Michael Iverson wrote:
> >> >> I've done about zero research into this, but perhaps you could run
> >> >> two separate daemon instances, one listening on each IP.
> >> >>
> >> >> On Tue, May 11, 2010 at 8:55 AM, Bart Coninckx
> >> >
> >> > <bart.coninckx at telenet.be>wrote:
> >> >> > On Tuesday 11 May 2010 12:58:45 Michael Iverson wrote:
> >> >> > > I'd be quite interested as well, obviously. So this is what we
> >> >> > > would end up with:
> >> >> > >
> >> >> > > Host A is primary for drbd volume 1, and secondary for drbd
> >> >> > > volume 2. It acts as an iSCSI target for whatever's on volume 1.
> >> >> > >
> >> >> > > Host B is primary for volume 2, and secondary for volume 1. It
> >> >> > > acts as a target for whatever's on volume 2.
> >> >> > >
> >> >> > > If either node fails, the opposite node takes over the secondary
> >> >> > > volume, and exports its fallen comrade's iSCSI targets.
> >> >> > >
> >> >> > > This idea could possibly be extended with Ben's approach of one
> >> >> > > DRBD volume per iSCSI target. (Except that it would be in a
> >> >> > > primary/secondary role, instead of primary/primary.) This would
> >> >> > > make the process of rebalancing the load between the two nodes
> >> >> > > fairly trivial.
> >> >> > >
> >> >> > > Mike
> >> >> > >
> >> >> > > On Tue, May 11, 2010 at 5:09 AM, Bart Coninckx
> >> >> > > <bart.coninckx at telenet.be
> >> >> >
> >> >> > wrote:
> >> >> > > >> It is. I'm planning to showcase this in one of our upcoming
> >> >> > > >> webinars.
> >> >> > > >>
> >> >> > > >> Cheers,
> >> >> > > >> Florian
> >> >> > > >
> >> >> > > > Excellent, any timeframe on this? As it happens I'm dealing
> >> >> > > > with a
> >> >> >
> >> >> > setup
> >> >> >
> >> >> > > > now that could definitely benefit from this.
> >> >> > > >
> >> >> > > > B.
> >> >> > > > _______________________________________________
> >> >> > > > drbd-user mailing list
> >> >> > > > drbd-user at lists.linbit.com
> >> >> > > > http://lists.linbit.com/mailman/listinfo/drbd-user
> >> >> >
> >> >> > Agreed, but what might be less trivial is to convince a running
> >> >> > IETD target to
> >> >> > have the config for the "other" targets merged to the existing
> >> >> > targets and at
> >> >> > the same time bind to the new secondary IP address, preferably
> >> >> > while not breaking running operation. This all should be taken care
> >> >> > of by Heartbeat.
> >> >> >
> >> >> > I'm going to try to dive into the challenge and report back to the
> >> >> > list, unless the webinar would happen fairly soon.
> >> >> >
> >> >> >
> >> >> > B.
> >> >> > _______________________________________________
> >> >> > drbd-user mailing list
> >> >> > drbd-user at lists.linbit.com
> >> >> > http://lists.linbit.com/mailman/listinfo/drbd-user
> >> >
> >> > Not possible:
> >> >
> >> > http://sourceforge.net/mailarchive/message.php?msg_id=02dd01c8263f%244
> >> >496 ae60%245dd810d1%40e3demo
> >> >
> >> >
> >> > Rgds,
> >> >
> >> > B.
> >> > _______________________________________________
> >> > drbd-user mailing list
> >> > drbd-user at lists.linbit.com
> >> > http://lists.linbit.com/mailman/listinfo/drbd-user
> >
> > Building a HA lcuster with IETD and DRBD is not really challenging, has
> > been done numerous times. The challenge would be having a active/passive
> > one on which each node is both active for some LUNs and active for
> > others, especially at failover.
> >
> > I don't quite get the suggestion on the first link, having a
> > active-active one and both nodes serving stuff. But I guess it would not
> > distribute load in between two nodes, what my fist idea would do.
> >
> >
> > _______________________________________________
> > drbd-user mailing list
> > drbd-user at lists.linbit.com
> > http://lists.linbit.com/mailman/listinfo/drbd-user
> 

Heartbeat normally takes care of both things: failing over and starting all 
resources at boot. I guess we need a Heartbeat resource agent that can use 
ietdadm. I really wonder what the webinar is about.

B. 





More information about the drbd-user mailing list