[Openais] Corosync UDP ports

[Christine Caulfield]

On 15/03/10 09:53, Colin wrote:
> Hi All,
>
> in a test that we started last week we have two Pacemaker+Corosync
> clusters, each with three hosts, where all six hosts are on the same
> network(s). The two clusters are identically configured, with one
> execption: the mcastport is 688 for one, and 689 for the other.
>
> This morning I found the clusters in a strange state, none of the
> hosts could see any of the others, i.e. Pacemaker output was "as if"
> Corosync wasn't running on the other  nodes, although the network was
> fine, as I could easily verify with a ping etc.
>
> I then noticed in the lsof output that Corosync seems to also use the
> port below the configured mcastport, which leads me to my questions:
>
> Is this normal? It doesn't seem to be documented in
> http://corosync.org/doku.php?id=faq:configure_openais and
> corosync.conf(5).
> Is this overlap created by the additional port a likely cause for the
> cluster conking out?
>

Yes, corosync uses both port <n> and <n-1>, so if you have two clusters 
in the same multicast address you will need to take this into account.

I haven't tried setting up cluster like you describe but I can easily 
imagine that using ports like that would cause extreme confusion!

Chrissie