Just been reading about a 5 possible set up: all indexes in the cluster are slaves to a master index outside of the cluster. building and maintaining the index happens outside of the cluster which would only be used for queries. are their any issues with this set up ?
thanks lee On 24 November 2010 11:53, lee carroll <lee.a.carr...@googlemail.com> wrote: > Hi We are investigating / looking to deploy solr on to a weblogic cluster > of 4 servers. > The cluster has no concept of a master / slave configuration so we are > thinking of the following solutions (new to solr so some or all may be bad > ideas:-) > > 1) all 4 servers run their own index. we re-index each server individually > this works best in terms of fail over but we worry about machine indexs > being slightly different due to re-index timings. can we index one server > and then copy index? > what is the effect on a solr server when an index is copied but it is still > reading the index? > > 2) all 4 servers use the same index > not very good for fail over, performance and we worry about multiple > servers updating. although if we controlled re-indexing to always use one > server and then commit on all 4 would this > solve this issue? > > 3) set up master - slave as per docs for 1.4.X http replication. > the down side here is we introduce a master server in our cluster which > should be HA - what happens if we loose the master will the slaves still be > operable ? > is it justa case of having stale indexes until the master is brought back > into the cluster > is limiting updates to the master server good enough > > 4) set up repeaters (master/slave servers) as per docs > this fits in better with a weblogic HA cluster but seems to be not the > exact match to the solution repeaters are aimed at (reducing data transfers > across a wan). ie will we bring down the cluster with endless updates with > the follwoing config or > will the set up stop propigating the changes ? > > server1 (master, slave of server 2,3,4) > server2 (master, slave of server 1,3,4) > server3 (master, slave of server 1,2,4) > server4 (master, slave of server 1,2,3) > > mmph this one sounds very whacky > > Anyways if we can index a single instance and copy index seemlessly to > others this seems maybe the best approach if not maybe option3 and deal with > stale indexes if the master drops out (will have a tough time convincing the > police on this one maybe) how have others solved solrs master slave model > within a HA cluster ? (particularly weblogic clusters.. > > Thanks for all help so far (this is 3rd question posted this week :-) > > lee > > > > > > >
