I've successfully made extensive use of load balancers in sharded, replicated slave setups - see [1].
My question is how that might work with a master. You can have a load balancer, but you'd need to configure it into a 'fail over but please don't fail back' configuration. I'm not sure if that is possible on the load balancers we have used. Otherwise, if your master had a five minute blip, you could have some content going to your backup, then traffic returning to your master, leading to master/backup out of sync and content missing from your master index. It seems to me, unless I am missing something, that while a load balancer can be useful, it is only as a part of a larger scheme when it comes to master replication. Or am I missing something? Upayavira [1] http://www.slideshare.net/sourcesense/sharded-solr-setup-with-master On Sun, 19 Dec 2010 22:41 -0800, "Lance Norskog" <goks...@gmail.com> wrote: > If you have a load balancer available, that is a much cleaner solution > than anything else. After the main indexer comes back, you have to get > the current index state to it to start again. But otherwise > > On Sun, Dec 19, 2010 at 10:39 AM, Upayavira <u...@odoko.co.uk> wrote: > > > > > > On Sun, 19 Dec 2010 10:20 -0800, "Tri Nguyen" <tringuye...@yahoo.com> > > wrote: > >> How do we tell the slaves to point to the new master without modifying > >> the config files? Can we do this while the slave is up, issuing a > >> command to it? > > > > I believe this can be done (details are in > > http://wiki.apache.org/solr/SolrReplication), but I've not actually done > > it. > > > > Upayavira > > > >> --- On Sun, 12/19/10, Upayavira <u...@odoko.co.uk> wrote: > >> > >> > >> From: Upayavira <u...@odoko.co.uk> > >> Subject: Re: master master, repeaters > >> To: solr-user@lucene.apache.org > >> Date: Sunday, December 19, 2010, 10:13 AM > >> > >> > >> We had a (short) thread on this late last week. > >> > >> Solr doesn't support automatic failover of the master, at least in > >> 1.4.1. I've been discussing with my colleague (Tommaso) about ways to > >> achieve this. > >> > >> There's ways we could 'fake it', scripting the following: > >> > >> * set up a 'backup' master, as a replica of the actual master > >> * monitor the master for 'up-ness' > >> * if it fails: > >> * tell the master to start indexing to the backup instead > >> * tell the slave(s) to connect to a different master (the backup) > >> * then, when the master is back: > >> * wipe its index (backing up dir first?) > >> * configure it to be a backup of the new master > >> * make it pull a fresh index over > >> > >> But, Jan Høydahl suggested using SolrCloud. I'm going to follow up on > >> how that might work in that thread. > >> > >> Upayavira > >> > >> > >> On Sun, 19 Dec 2010 00:20 -0800, "Tri Nguyen" <tringuye...@yahoo.com> > >> wrote: > >> > Hi, > >> > > >> > In the master-slave configuration, I'm trying to figure out how to > >> > configure the > >> > system setup for master failover. > >> > > >> > Does solr support master-master setup? From my readings, solr does not. > >> > > >> > I've read about repeaters as well where the slave can act as a master. > >> > When the > >> > main master goes down, do the other slaves switch to the repeater? > >> > > >> > Barring better solutions, I'm thinking about putting 2 masters behind a > >> > load > >> > balancer. > >> > > >> > If this is not implemented already, perhaps solr can be updated to > >> > support a > >> > list of masters for fault tolerance. > >> > > >> > Tri > >> > > > > > > -- > Lance Norskog > goks...@gmail.com >
