Re: [Evms-cluster] EVMS and Linux-HA

[Steve D. <st...@us...>]

Chris Rogers wrote:
> Does the lastest version of EVMS  (2.3.2) still have to run
> in admin mode when using Cluster Segment Manager in
> conjunction with the latest version of Linux-HA (v 1.2.1 or
> 1.3.0)? I know that in the past if one node was down, then
> the surviving node could not takeover the failed node's
> volumes unless EVMS was running in admin mode. Is this
> by design? Doesn't this kind of defeat the purpose of using
> EVMS and Linux-HA together? It seems to me that when
> one node has a critical failure is when you would most want
> the surviving tnode to take over the volumes but it seems
> that EVMS does not act this way unless it is running in
> admin mode.
>
> Thanks in advance for any wisdom you can impart.
>
>
> -- Chris Rogers

Hello, Chris.

IIRC, the problem was that when one node died the remaining node did not
have quorum.  That is, there was no cluster membership, even though the one
node was up and running.  That is a problem with the CCM (the component
that manages the cluster membership) used by Linux-HA.  EVMS is behaving
properly.  The EVMS Cluster Segment Manager will not allow changes to be
made if the node it is on does not have quorum, i.e., is not a member of
the cluster.  If the node is not a member of the cluster the CSM can't
safely change things since it cannot coordinate with the other nodes.
Running with admin_mode=yes ignores that safety restriction.

On the occasions when the remaining node has quorum (which happens on my
test machines) the failover works without having to run in admin mode.

I don't know if the latest HA releases have fixed the problem in the CCM.
I'm running with HA version 1.2.0 on one of my test clusters.  I haven't
noticed problems, but then I haven't tested failover very much on it
either.

Steve D.