[SOLVED] Mixed kernels 2.6 & 3.10 : no quorum

[Florent]

Hi everyone,

I have a cluster with 5 nodes, all running 3.10 Proxmox kernel.

I want to add another node, but on 2.6.32 Proxmox kernel (due to some problems, I want to come back to 2.6.32).

If I start this node on 2.6.32 kernel, cman can't get quorum.
If I start this node on 3.10, cman get quorum !

Of course network is working in both cases. Multicast too.

What could be the problem ?

Thank you a lot.

 

[RandomK]

Are date and time synced on all nodes? I've had that bite be in the past.

 

[noway]

The same for me.
Using 3.10 kernel in 6 node cluster with 2.6 kernels makes cluster no quorum.
With following error flood in logs:

corosync[2939]: [TOTEM ] Retransmit List: 2ec 2ed

 

[Florent]

Are date and time synced on all nodes? I've had that bite be in the past.

Yes, times are synced on all nodes. Handled by ntpd.

The same for me.
Using 3.10 kernel in 6 node cluster with 2.6 kernels makes cluster no quorum.
With following error flood in logs:

I don't have this error in logs... I don't have any error Just messages like this :
corosync[3326]: [TOTEM ] A processor joined or left the membership and a new membership was formed.
corosync[3326]: [CPG ] chosen downlist: sender r(0) ip(10.111.0.2) ; members(old:1 left:0)

 

[Florent]

If I run Debian 3.2 kernel, quorum is OK.

I cannot see any difference in kernel config about Bridge or Multicast.

The only differences I see between 2.6.32 and 3.10 is multicast_querier and multicast_snooping. But If I set the same values, it does not get quorum neither..

 

[noway]

I gave up on trying to find out what was the reason...
Moved back on rebuilt 2.6 kernel with deleted newest and bugged nic drivers (had an issue with bridge+4nics bond kernel panics).

 

[Florent]

I think I have found the solution for this problem.

Do not touch to 3.10 nodes, but on 2.6.32 nodes, disable multicast_snooping & multicast_querier on all bridges :

find /sys/devices/virtual/net -name 'multicast_snooping' -exec sh -c 'echo 0 > {}' \;
find /sys/devices/virtual/net -name 'multicast_querier' -exec sh -c 'echo 0 > {}' \;
service cman stop; service cman start; service pve-cluster restart

 

[ScOut3R]

Have you tried KVM live migration between a 2.6.32 and a 3.10 node?