[ClusterLabs] Antw: Corosync ring marked as FAULTY

[Vladislav Bogdanov]

22.02.2017 11:40, Denis Gribkov wrote:
> Hi,
>
>
> On 22/02/17 10:35, bliu wrote:
>> Did you specify interface with "-i <your private network card>" when
>> you are using tcpdump. If you did, corosync is not talking with the
>> multicast address, you need to check if your private network support
>> multicast.
>
> Yes, I have used command:
>
> # tcpdump -i em2 udp port 5505 -vv -X
>
> Thanks for your advice, I'll ask network engineers about the issue.

That could be igmp querier issue. Corosync does not follow "common" 
model of mcast usage - one sender/router in a segment and many 
receivers. Instead, all corosync nodes are mcast senders and receivers. 
For that to work reliably, both IGMP snooping should be enabled and 
*work* in a segment and IGMP querier exists there (in absence of a mcast 
router). Also, if switches differ in that two segments, then issue could 
be with IGMPv2 vs IGMPv3 snooping support in that in ring0 segment. Not 
all switches support IGMPv3 (linux default) snooping, and to be on a 
safest side it could be needed to downgrade used linux IGMP version to 
v2 (/proc/sys/net/ipv4/conf/<intf>/force_igmp_version).


>
> --
> Regards Denis Gribkov
>
>
>
> _______________________________________________
> Users mailing list: Users at clusterlabs.org
> http://lists.clusterlabs.org/mailman/listinfo/users
>
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://bugs.clusterlabs.org
>