RE: [Evms-cluster] evms_failover problems

[Chris R. <chr...@ne...>]

I also have the problem with the segfault in the evmsgui with a very =
similar setup. I ran evmsgui -d debug and have attached that log. I am =
running EVMS v2.3.2 and kernel 2.4.24=20

-----Original Message-----
From: evm...@li... =
[mailto:evm...@li...] On Behalf Of Steve =
Dobbelstein
Sent: Tuesday, May 11, 2004 9:21 AM
To: Alexander Kordecki
Cc: evm...@li...; evm...@li...
Subject: Re: [Evms-cluster] evms_failover problems





Alexander Kordecki <al...@ko...> wrote on 05/11/2004 07:23:43 AM:

> Hi Steve !
>
> Thanks for your reply.
>
> > You don't need admin_mode to do a failover.  Which version of=20
> > heartbeat
are
> > you using?  Earlier versions had a problem with producing a valid
cluster
> > membership when one node died.  The membership services said there=20
> > was
no
> > membership even when one node was still alive.  The EVMS Cluster
Segment
> > Manager (CSM) won't make changes to its containers unless it has a
valid
> > membership (or admin_mode is yes).  If there is no valid membership,
the
> > CSM can't safely make changes since it can't coordinate the changes
with
> > the other nodes in the cluster.
>
> I use heartbeat 1.2.0 ... Is this enough ?

That's what I'm using on one of my test clusters.  Once in a while I =
won't get a membership when I run evmsgui.  Sometimes if I close then =
restart evmsgui I get a membership.

> > If you do not run in admin_mode, any containers that are deported or
don't
> > belong to the node will not be produced by the CSM.
> >
> > On my test cluster, if I run without admin_mode, the directory
> > /dev/evms/.nodes/priv1 goes away on a "evms_failover priv1 stop". =20
> > If I
run
> > in admin_mode, the directory remains.
> >
>
> Thats the same on my cluster, but in non_admin mode I couldn't get=20
> back
the
> container if it is deported. Neither on the node where i run the=20
> "evms_failover container stop" nor on the other node.
>
> Every time I try it, I get the message: "WARNING: Container <name>=20
> failed
to
> start"
>
> After this evms seems the behave curious ... It hangs, couldn't=20
> connect
to
> the other cluster, etc.

I found a bug in the 2.3.2 version of evms_failover when I was trying to =
reproduce your problem.  The end symptom was that containers couldn't be =
started.  Sounds like your problem.  Try this patch.

(See attached file: evms_failover.patch)

> > Hmmm.  I can use evmsgui to change the container from deported to
private,
> > with or without admin_mode.  Can you run "evmsgui -d debug" and then
send
> > me log, /var/log/evms-engine.log, (gzipped, since it's big) after=20
> > the segfault.  I'll see if it has any clues as to what segfaulted. =20
> > You
could
> > also run evmsgui under gdb and then get a back trace after the
segfault.
> >
>
> I attached the log ... But sorry, I have no gdb installed.

I'm looking through the log.  I'll let you know if I find anything.

Steve D.