Mailinglist Archive: opensuse-ha (30 mails)

< Previous Next >
Re: [opensuse-ha] ext4 failure on cluster
On Mon, 2014-08-18 at 21:04 +0400, Andrei Borzenkov wrote:
В Mon, 18 Aug 2014 18:56:44 +0200
steve <steve@xxxxxxxxxxxx> пишет:

On Mon, 2014-08-18 at 13:16 +0200, Kai Dupke wrote:
On 08/18/2014 10:51 AM, steve wrote:
Thanks for the input. It's not performance we want. We're never going to
get that. It's reliability. What do you mean by fail over? I think we
already have that. If one node fails, there is always the other one
already up. If you mean have only one node available at a time, what's
the advantage of that? Also, how are we going to serve a windows domain
without ctdb?

I assume you can handle a windows domain with Samba and don't need CTDB?

fail-over means to have samba either running on A xor B.

The advantage is noOCFS2, which means less communication overhead and
less complex setup.

'Already the other is up' does not help really as CTDB does not provide
a transparent fail-over for a client connection AFIK. The time the
system needs to identify the issue and switch from node A to B isn't
that long. Usually it isn't an issue if not happens multiple times a week.

I'm not sure what you mean by transparent fail-over.

After failover TCP connection is broken and client has to reconnect.
Depending on client application it may have more or less impact
(Windows Explorer will probably reconnect; database will likely crash).

Ah, I see. So it depends upon the client reconnecting. I thought it was
the cluster's responsibility to make the takeover invisible. We can
still work on LibreOffice documents for example and disable and enable
nodes at will. I Imagine that to be the exception rather than the norm
though. Or, more likely, we've just been lucky.

It has nothing to do with server, but simply with the way SMB works and
was the same also using Windows Server cluster. There is limited
support for durable/persistent shares in SMB 3.0, do not know whether
SAMBA implements it.
OK. So there's really no such thing as foolproof HA.
S pp li

Do you mean that
the IP is not taken over without user intervention? Or that smbd on the
other node does not take over the share? Both, or something else?

What do your SLES tests show?
L x

Kai Dupke
Senior Product Manager
Server Product Line
Phone: +49-(0)5102-9310828 Mail: kdupke@xxxxxxxx
Mobile: +49-(0)173-5876766 WWW:

SUSE Linux Products GmbH - Maxfeldstr. 5 - 90409 Nuernberg (Germany)
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer, HRB 16746 (AG Nurnberg)

To unsubscribe, e-mail: opensuse-ha+unsubscribe@xxxxxxxxxxxx
To contact the owner, e-mail: opensuse-ha+owner@xxxxxxxxxxxx

< Previous Next >
List Navigation
Follow Ups