[Pacemaker] [DRBD-user] DRBD is not syncing over my routed network.

[Florian Haas]

[CCing the Pacemaker list here, in case we have people interested in
following this discussion there]

On 09/08/2009 05:52 PM, Gary.Webb at opengi.co.uk wrote:
> Hi.      I am having a problem whilst carrying out some preliminary
> testing of DRBD with OpenAIS.  My test setup was all working fine. I
> have now introduced it to a simulated wide area network by changing the
> subnet on each system, modifying the config files accordingly, adding
> the relevant routes and gateway, and using an established Linux box that
> we use for throttling bandwidth for such testing. The network config is
> sound as I can ping to and from each node through the router system. I
> also have full key authentication setup between them. Nothing else has
> changed.

Gary,

If you actually get this to work with OpenAIS, be sure to let us know
and share your findings. AFAICS this is bound to fail as OpenAIS
currently isn't capable of unicast messaging, so unless you actually
manage to get multicast routing set up correctly between your sites,
you're out of luck. The Heartbeat communication stack does support
unicast, but it too requires that you have super-reliable links between
sites. And, regardless of the communication layer beneath it, Pacemaker
currently does not support split-site clustering.

If you want to do a split-site setup with what is available today, then
that effectively amounts to having two separate clusters, which happen
to share a DRBD replication link between them. A work-in-progress
writeup can be found at
http://www.drbd.org/users-guide/s-pacemaker-floating-peers.html --
please be patient; the information is still incomplete. It is set to be
completed with the next maintenance release of the User's Guide, which
is scheduled for shortly after the DRBD 8.3.3 release.

> Now, the DRBD nodes are failing to sync. I am getting no obvious errors.
> If I watch the /proc/drbd file, it goes through the motion of syncing,
> and the progress bar briefly is seen. It then flips back.

Probably just a result of the cluster manager detecting one cluster
partition after another, as the underlying communication stack can't see
its peer.

Cheers,
Florian

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 260 bytes
Desc: OpenPGP digital signature
URL: <https://lists.clusterlabs.org/pipermail/pacemaker/attachments/20090909/7f3d0e82/attachment-0003.sig>