Re: [Evms-devel] Please name new maintainers

[Robert W. <raw...@no...>]

John,

Thanks for offering more details. You asked "what did it accomplish?"
When working as designed, EVMS is supposed to ensure all nodes have
a consistent view of shared storage, and tools are designed so that
you can't get into accidental trouble if carving disks on one node in your
cluster, and someone else tries the same on a different node, and=20
somehow they collide. The cluster container is a separate part, and is
not required, unless one also objects to having devnodes visible across
all servers in the cluster. Private containers are intended to reduce the
risk of doing something bad to a devnode from one server in the
cluster, it it's actually in use on some other. This is an additional =
layer
of protection which can be ignored assuming admins are careful to e.g.
not mkfs on a devnode if used by some other server.

I'm sorry you've had trouble with your cluster, as far as I know we've
fixed and pushed upstream all the bugs we've encountered during our
QA and by working with other customers.=20

But the question remains - who is maintaining EVMS - and if it slowly
goes away, what would people like to use instead? (assuming the same
features mentioned above are desirable)...

Thanks,
Robert



>>> John Lange <joh...@op...> 12/05/07 9:57 AM >>>
On Wed, 2007-12-05 at 07:58 -0700, Robert Wipfel wrote:

> Only one node can start an EVMS engine at a time - i.e. the GUI - which
> is IBM's design for cluster awareness. That engine coordinates with =
daemons
> on other nodes to synchronize their view of shared storage. All devnodes
> show up on all nodes, unless you are using private cluster containers =
which
> are designed to hide the devnodes on all cluster nodes except the one =
that
> has it active. Shared cluster containers are intended for e.g. cluster =
file
> systems, that by design expect to be able to access shared devnodes over
> all nodes for parallel access to storage. If you don't use the cluster =
container
> then the devnodes are visible on all nodes...

That's all fine but it heartbeat is supposed to handle starting EVMS and
it doesn't work. It used to work when it was originally setup and the 4
nodes ran find for quite some time but something happened and now after
a recent reboot it will only start on one node at a time. Plus the EVMS
GUI won't start either.

As mentioned; EVMS has been a nightmare from the start. Just getting it
working with heartbeat in the first place was a huge undertaking and
what did it accomplish? We could have done the same thing just using
LVM.

Live and learn I guess.

> We have lots of customers using it successfully - do you have a =
support=20
> agreement?

This client has 4 licensed versions of SLES10 but not a support
agreement.

By the way, when I said "I blame Novell", it was tongue-and-cheek. I
just meant the only reason we tried it was because it was in Novell
"Cool Solutions". I don't blame Novell for EVMS & Heartbeat problems.

John