lars.ellenberg at linbit
Nov 16, 2009, 10:47 AM
Post #3 of 3
(91 views)
Permalink
|
On Mon, Nov 16, 2009 at 07:10:12PM +0100, Dejan Muhamedagic wrote:
> Hi,
>
> On Mon, Nov 02, 2009 at 12:07:59PM -0500, Sam Tran wrote:
> > Hi All,
> >
> > I followed the thread "Food for thought: add something like cutter to
> > IPaddr2 (or portblock?) RA"
> > (http://lists.linux-ha.org/pipermail/linux-ha-dev/2008-October/016196.html)
> > with great interest.
> >
> > I am working on a cluster of Master OpenLDAP servers using PaceMaker
> > and OpenAIS. The problem I have lies in the replication between the
> > master server that holds the IP address resource and a replica server.
> > In the "refreshAndPersist" replication mode that is being used, the
> > replica polls the master server for updates, then the connection
> > between the replica and the master server is maintained, and the
> > replica is waiting for subsequent updates from the master server. In
> > the event of a failure of the initial master the new master is taking
> > over the IP address resource, but doesn't know anything about the
> > previous persist stage, therefore is not able to send new updates to
> > the replica. An RST needs to be sent to the replica in order to
> > terminate the existing session and force a polling retry from the
> > replica, or the replica would wait for the session to time out.
> >
> > I was wondering whether some work has been done as far as the
> > implentation of the tickle ACK feature in IPaddr2 RA is concerned.
>
> Not to my knowledge. It would obviously be a good feature. The
> only thing which is not clear to me is who/how would
> keep/maintain/synchronize the connections database
As a "best effort" sort of thing, you could do a "depth=X" monitoring
action in the IPaddr2 RA, which would
grep "ESTABLISHED" /proc/net/nf_conntrack |
dd conv=fsync of=/somewhere/on/DRBD/or/NFS/or/iSCSI
On stop, it may (optionally?) truncate that state file.
On start, it would (optionally?) check that state file,
and send out "Tickle ACKs".
You will miss only those connections that have been
established since the last "grep", i.e. since the last
"monitor depth=X". If you want more, use conntrackd.
Volunteers?
--
: Lars Ellenberg
: LINBIT | Your Way to High Availability
: DRBD/HA support and consulting http://www.linbit.com
DRBD® and LINBIT® are registered trademarks of LINBIT, Austria.
_______________________________________________________
Linux-HA-Dev: Linux-HA-Dev[at]lists.linux-ha.org
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
|
