Andrew Beekhof @ 15/04/2011 03:31 -0300 dixit:
On Thu, Apr 14, 2011 at 6:39 PM, Carlos G Mendioroz <t...@huapi.ba.ar> wrote:
Hi,
I've a doubt on the best way (tm) to deal with HA in the case of services
that are somehow proxies to some other resource (i.e.
stateless in some sense).
The problem seems to be that the clients need to know its identity
and location, which usually boils down to IP address.
Even though some services might run well with anycast models,
many are TCP based so anycast does not fit that good.
"moving" the IP address seems the usual mechanism that
pacemaker based resources use. This implies that the service itself
has to be restarted, because many depend on the service IP address
being attached to the host for the configuration to be accepted.
Has anybody thought about having a local (i.e. loppback) address
and then switching ARP responsibility over the LAN as an alternative ?
Has this been implemented already and I'm rediscovering a wheel ? :)
I think there's something like that in the IPaddr agent (for
ldirectord support IIRC)
Yup, sure thing. lvs_support option.
Wheels already rolling :) Thanks.
There's an issue though, that I don't really know how much of an impact
it may have. The code does remove and add the ip address to different
interfaces. This cuould be felt by the service, as the address will
disappear for a while. Moving ARP responsibility (proxy ARP for
the loopback, that is) might be easier on the upper layers.
Have to test it to be sure.
-Carlos
--
Carlos G Mendioroz <t...@huapi.ba.ar> LW7 EQI Argentina
_______________________________________________
Pacemaker mailing list: Pacemaker@oss.clusterlabs.org
http://oss.clusterlabs.org/mailman/listinfo/pacemaker
Project Home: http://www.clusterlabs.org
Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
Bugs: http://developerbugs.linux-foundation.org/enter_bug.cgi?product=Pacemaker