Hi,
I've got a very small and very lightly utilized corosync
1.4.1[-7.el6_3.1.x86_64] configuration on 4 EL6 KVM virtual nodes (which
are running on FC17 as the KVM host). The libvirt networking I am using
for the 4 nodes is a completely virtual (i.e. not bridged with any real
interface) network.
I have the 4 nodes configured into 2 corosync clusters on multicast
ports 5405 and 5410 respectively.
On top of corosync I am running pacemaker 1.1.7[-6.el6.x86_64].
What I am finding is that while the clusters will come up even just the
slightest amount of pacemaker activity will start to cause corosync to
report the following sorts of messages:
Nov 26 15:43:52 mds2 corosync[1293]: [TOTEM ] Retransmit List: 1c8 1c9
Nov 26 15:43:53 mds2 corosync[1293]: [TOTEM ] Retransmit List: 1c8 1c9
Nov 26 15:43:53 mds2 corosync[1293]: [TOTEM ] Retransmit List: 1c8 1c9
Nov 26 15:43:53 mds2 corosync[1293]: [TOTEM ] Retransmit List: 1c8 1c9
Nov 26 15:43:53 mds2 corosync[1293]: [TOTEM ] Retransmit List: 1c8 1c9
Nov 26 15:43:53 mds2 corosync[1293]: [TOTEM ] Retransmit List: 1c8 1c9
Nov 26 15:43:54 mds2 corosync[1293]: [TOTEM ] Retransmit List: 1c8 1c9
Nov 26 15:43:54 mds2 corosync[1293]: [TOTEM ] Retransmit List: 1c8 1c9
Some research seems to yield that typical causes of this problem are:
* machines that are quite different in performance
* slow networks
I don't seem to have either of those. Given that my machines are VMs on
KVM all built from the same template and running on a KVM host with 16GB
or RAM and 8 CPU cores there should be plenty of resources for them and
they should get fairly equal access to those resources, ruling out the
first item.
The virtual network the nodes are using is entirely unused but for the
corosync traffic and bit of SSH, etc. so I can't imagine there being any
serious shortage there. I'm happy to be proven wrong there though.
Any ideas on how to resolve this situation?
FWIW, this configuration of VMs on the particular KVM host has worked
just fine it the past. The only thing that could be different from when
it has worked just fine would be upgraded packages on the EL6 VMs. Has
anything resembling this possibly been introduced into the most recent
of packages that would come from an up-to-date EL6?
Cheers,
b.
_______________________________________________
discuss mailing list
discuss@xxxxxxxxxxxx
http://lists.corosync.org/mailman/listinfo/discuss
