[Pacemaker] TOTEM Retransmit list in logs when a node gets up

Christine Caulfield ccaulfie at redhat.com
Fri Nov 14 08:58:52 UTC 2014


On 13/11/14 17:54, Daniel Dehennin wrote:
> Digimer <lists at alteeve.ca> writes:
>
>> This generally happens if the network is slow or congested. It is
>> corosync saying it needs to resend some messages. It is not uncommon
>> for it to happen now and then, but that is a fairly large amount of
>> retransmits.
>
> Thanks for the explanation.
>
>> Is your network slow or saturated often? It might be that the traffic
>> from the join is enough to push a congested network to the edge.
>
> Not really:
>
> - two physical hosts with:
>    + one 1Gb/s network card for OS (corosync network)
>    + three 1Gb/s network cards in LACP bonding included in an Open
>      vSwitch
>
> - one physical host with:
>    + one 10Gb/s network card for the OS (corosync network)
>    + three 10Gb/s network cards in LACP bonding included in an Open
>      vSwitch
>
> - one KVM guest (quorum node) with:
>    + one virtio card (corosync network)
>
> - one KVM guest with:
>    + one virtio card for service (OpenNebula web frontend)
>    + one virtio card for corosync communications
>
> With tcpdump I can see packets flying around on all nodes, but it looks
> like there is something with my two cards KVM guest, when I start
> pacemaker on it I begin to see Retransmit messages in other nodes logs.
>
> Is there a may to know which nodes is responsible of the resend of
> theses messages?
>

If its only happening at startup it could be the switch/router learning 
the ports for the nodes and building its routing tables. Switching to 
udpu will then get rid of the message if it's annoying

Chrissie







More information about the Pacemaker mailing list