Arun, Yes, changing the solrconfig.xml to point to the new master could require a restart. However, if you use logical addresses (VIPs in the Load Balancer or even local hostname aliases if you don't have a LB) then you just need to point those VIPs/aliases to new IPs and the Solr slave won't have to be restarted.
Otis ----Sematext :: http://sematext.com/ :: Solr - Lucene - Nutch Lucene ecosystem search :: http://search-lucene.com/ ----- Original Message ---- > From: Arunkumar Ayyavu <arunkumar.ayy...@gmail.com> > To: solr-user@lucene.apache.org > Sent: Sun, October 10, 2010 1:57:34 PM > Subject: Re: Multiple masters and replication between masters? > > On Mon, Oct 4, 2010 at 4:58 PM, Upayavira <u...@odoko.co.uk> wrote: > > On Mon, 2010-10-04 at 00:25 +0530, Arunkumar Ayyavu wrote: > >> I'm looking at setting up multiple masters for redundancy (for index > >> updates). I found the thread in this link > >> >(http://www.lucidimagination.com/search/document/68ac303ce8425506/multiple_masters_solr_replication_1_4) > > >> discussed this subject more than a year back. Does Solr support such > >> configuration today? > > > > Solr does not support master/master replication. When you commit > > documents to SOLR, it adds a segment to the underlying Lucene index. > > Replication then syncs that segment to your slaves. To do master/master > > replication, you would have to pull changes from each master, then merge > > those changed segments into a single updated index. This is more complex > > than what is happening in the current Solr replication (which is not > > much more than an rsync of the index files). > > > > Note, if you commit your changes to two masters, you cannot switch a > > slave between them, as it is unlikely that the two masters will have > > matching index files. If you did so, you would probably trigger a pull > > of the entire index across the network, which (while it would likely > > work) would not be the most efficient action. > > > > What you can do is think cleverly about how you organise your > > master/slave setup. E.g. have a slave that doesn't get queried, but > > exists to take over the role of the master in case it fails. The index > > on a slave is the same as that in a master, and can immediately take on > > the role of the master (receiving commits), and upon failure of your > > master, you could point your other slaves at this new master, and things > > should just carry on as before. > Wouldn't this require restart of Solr instances? > > Sorry, I couldn't respond to you earlier as I wasn't checking my mails > for sometime. > > > > > Also, if you have a lot of slaves (such that they are placing too big a > > load on your master), insert intermediate hosts that are both slaves off > > the master, and masters to your query slaves. That way, you could have, > > say, two boxes slaving off the master, then 20 or 30 slaving off them. > > > >> And does Solr support replication between masters? Otherwise, I'll > >> have to post the updates to all masters to keep the indexes of masters > >> in sync. Does SolrCloud address this case? (Please note it is too > >> early for me to read about SolrCloud as I'm still learning Solr) > > > > I don't believe SolrCloud is aiming to support master/master > > replication. > > > > HTH > > > > Upayavira > > > > > > > > > > -- > Arun >
