From: Lars Marowsky-Bree <lmb@suse.de>
To: Daniel Phillips <phillips@istop.com>
Cc: David Teigland <teigland@redhat.com>,
Steven Dake <sdake@mvista.com>,
linux-kernel@vger.kernel.org, akpm@osdl.org
Subject: Re: [PATCH 1b/7] dlm: core locking
Date: Thu, 28 Apr 2005 14:55:12 +0200 [thread overview]
Message-ID: <20050428125512.GR21645@marowsky-bree.de> (raw)
In-Reply-To: <200504280249.04735.phillips@istop.com>
On 2005-04-28T02:49:04, Daniel Phillips <phillips@istop.com> wrote:
> > Just some food for thought how this all fits together rather
> > neatly.
>
> It's actually the membership system that glues it all together. The
> dlm is just another service.
Membership is one of the lowest level and high privileged inputs to the
whole picture, of course.
However, "membership" is already a pretty broad term, and one must
clearly state what one is talking about. So we're clearly focused on
node membership here, which is a special case of group membership; the
top-level, sort of.
Then every node has it's local view of node membership, constructed
typically from observing node heartbeats.
Then the nodes communicate to reach concensus on the coordinated
membership, which will usually be a set of nodes with full N:N
connectivity (via the cluster messaging mechanism); and they'll also
usually aim to identify the largest possible set.
Eventually, there'll be a membership view which also implies certain
shared data integrity guarantees if appropriate (ie, fencing in case a
node didn't go down cleanly, and granting access on a clean join).
These steps but the last one usually happen completely internal to the
membership layer; the last one requires coordination already, because
the fencing layer itself might need recovery before it can fence
something after a node failure.
And then there's quorum computation.
Certainly you could also try looking at it from a membership-centric
angle, but the piece which coordinates the recovery of the various
components which makes sure the right kind of membership events are
delivered in the proper order, and errors during component recovery are
appropriately handled, is, I think, pretty much distinct from the
"membership" and a higher level component.
So I'm not sure I'd buy "the membership is what glues it all together"
on eBay even for a low starting bid.
Sincerely,
Lars Marowsky-Brée <lmb@suse.de>
--
High Availability & Clustering
SUSE Labs, Research and Development
SUSE LINUX Products GmbH - A Novell Business
next prev parent reply other threads:[~2005-04-28 12:57 UTC|newest]
Thread overview: 48+ messages / expand[flat|nested] mbox.gz Atom feed top
2005-04-25 16:58 [PATCH 1b/7] dlm: core locking David Teigland
2005-04-25 18:34 ` Nikita Danilov
2005-04-25 20:44 ` Daniel Phillips
2005-04-25 22:27 ` Nikita Danilov
2005-04-26 1:34 ` Daniel Phillips
2005-04-25 20:41 ` Jesper Juhl
2005-04-26 5:00 ` Daniel Phillips
2005-04-25 21:54 ` Steven Dake
2005-04-26 5:49 ` David Teigland
2005-04-26 17:40 ` Steven Dake
2005-04-26 22:24 ` Daniel Phillips
2005-04-26 23:04 ` Steven Dake
2005-04-27 0:53 ` Daniel Phillips
2005-04-27 1:50 ` Steven Dake
2005-04-27 4:21 ` Daniel Phillips
2005-04-27 3:02 ` David Teigland
2005-04-27 13:41 ` Lars Marowsky-Bree
2005-04-27 14:26 ` David Teigland
2005-04-28 12:33 ` Lars Marowsky-Bree
2005-04-28 16:39 ` Daniel McNeil
2005-04-28 16:45 ` Lars Marowsky-Bree
2005-04-29 8:25 ` Daniel Phillips
2005-05-02 20:45 ` Lars Marowsky-Bree
2005-05-02 23:23 ` Daniel Phillips
2005-04-29 4:01 ` David Teigland
2005-04-29 22:58 ` Daniel McNeil
2005-04-30 4:29 ` David Teigland
2005-04-30 9:09 ` Daniel Phillips
2005-04-30 10:32 ` Lars Marowsky-Bree
2005-04-30 11:12 ` Daniel Phillips
2005-05-02 20:51 ` Lars Marowsky-Bree
2005-05-02 22:21 ` Daniel Phillips
2005-05-05 12:25 ` Stephen C. Tweedie
2005-05-05 12:40 ` copy_to_user question linux
2005-05-05 13:13 ` Richard B. Johnson
2005-05-05 19:29 ` [PATCH 1b/7] dlm: core locking Daniel Phillips
2005-04-28 2:52 ` Daniel Phillips
2005-04-28 12:37 ` Lars Marowsky-Bree
2005-04-28 23:43 ` Daniel Phillips
2005-04-28 6:49 ` Daniel Phillips
2005-04-28 12:55 ` Lars Marowsky-Bree [this message]
2005-04-29 0:26 ` Daniel Phillips
2005-04-29 2:52 ` David Teigland
2005-04-29 3:49 ` Daniel Phillips
2005-05-02 21:00 ` Lars Marowsky-Bree
2005-05-03 2:54 ` David Teigland
2005-04-27 12:33 ` Domen Puncer
2005-04-27 13:30 ` David Teigland
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20050428125512.GR21645@marowsky-bree.de \
--to=lmb@suse.de \
--cc=akpm@osdl.org \
--cc=linux-kernel@vger.kernel.org \
--cc=phillips@istop.com \
--cc=sdake@mvista.com \
--cc=teigland@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox