On 2010-05-28 14:20, Vadym Chepkov wrote: > > On May 28, 2010, at 8:12 AM, Florian Haas wrote: > >> On 2010-05-28 14:01, Vadym Chepkov wrote: >>> Hi, >>> >>> I want to convert our home-made application to be managed by pacemaker >>> cluster. >>> The way it works now: application starts, discovers all IPs configured on >>> the system and if it sees preconfigured IP it becomes "master" and will >>> serve configuration requests, >>> if not - "node" and will try to connect to "master" node to get >>> configuration data. Other then that, application instances are absolutely >>> the same on all "cluster" servers and >>> they start collecting their own unique data. >>> >>> I started to write a resource agent in "stateful" manner, but the problem >>> is - I can't promote or demote and the documentation says : application has >>> to start in "slave" mode, >>> which I can't supply. So I thought of unique clones instead, but I need to >>> create collocation/ordering constraint with IP for the clone:0 and again >>> documentation says I shouldn't do it. >>> >>> As a workaround I could probably create a separate "master" instance and >>> clones with clone-max=node-1 and create -INFINITY collocation constraint >>> between them, but it's more of a hack it seems. >> >> Show us your RA and I'm sure quite a few people will come up with >> helpful suggestions. >> > > Imperative word was "started". You think I still should go multi-state RA for > this application?
If the application which that RA applies to distinguishes between roles equivalent to a Master and a Slave, and you want the RA to manage those, then probably yes. What makes you think you can't start the RA in Slave mode? Florian
signature.asc
Description: OpenPGP digital signature
_______________________________________________ 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
