Replicating Large Indexes
Wondering if anyone has experience with replicating large indexes. We have a Solr deployment with 1 master, 1 master/slave and 5 slaves. Our index contains 15+ million articles and is ~55GB in size. Performance is great on all systems. Debian Linux Apache-Tomcat 100GB disk 6GB RAM 2 proc on VMWare ESXi 4.0 We notice however that whenever the master is optimized, the complete index is replicated to the slaves. This causes a 100%+ bloat in disk requirements. Is this normal? Is there a way around this? Currently our optimize is configured as such: curl 'http://localhost:8080/solr/update?optimize=true&maxSegments=1&waitFlush=true&expungeDeletes=true' Willing to share our experiences with Solr. Thanks, Jason
Re: Replicating Large Indexes
Thanks Robert. We optimize less frequently than we used to. Down to twice a month from once a day. Without optimizing the search speed stays the same, however the index size increases to 70+ GB. Perhaps there is a different way to restrict disk usage. Thanks, Jason Robert Stewart wrote: Optimization merges index to a single segment (one huge file), so entire index will be copied on replication. So you really do need 2x disk in some cases then. Do you really need to optimize? We have a pretty big total index (about 200 million docs) and we never optimize. But we do have a sharded index so our largest indexes are only around 10 million docs. We have merge factor of 2. We run replication every minute. In our tests search performance was not very much better with optimization, but that may be specific to our types of searches, etc. You may have different results. Bob On Nov 1, 2011, at 12:46 AM, Jason Biggin wrote: > Wondering if anyone has experience with replicating large indexes. We have a > Solr deployment with 1 master, 1 master/slave and 5 slaves. Our index > contains 15+ million articles and is ~55GB in size. > > Performance is great on all systems. > > Debian Linux > Apache-Tomcat > 100GB disk > 6GB RAM > 2 proc > > on VMWare ESXi 4.0 > > > We notice however that whenever the master is optimized, the complete index > is replicated to the slaves. This causes a 100%+ bloat in disk requirements. > > Is this normal? Is there a way around this? > > Currently our optimize is configured as such: > > curl > 'http://localhost:8080/solr/update?optimize=true&maxSegments=1&waitFlush=true&expungeDeletes=true' > > Willing to share our experiences with Solr. > > Thanks, > Jason
RE: Replicating Large Indexes
Thanks Erick, Will take a look at this article. Cheers, Jason -Original Message- From: Erick Erickson [mailto:erickerick...@gmail.com] Sent: Tuesday, November 01, 2011 8:05 AM To: solr-user@lucene.apache.org Subject: Re: Replicating Large Indexes Yes, that's expected behavior. When you optimize, all segments are copied over to new segments(s). Since all changed/new segments are replicated to the slave, you'll (temporarily) have twice the data on your disk. You can stop optimizing, it's often not really very useful despite its name. That said, due to how segments are merged you will always have the potential for replicating your entire index to the slave if you happen to hit the magic segment merge event. And *that* said, there's quite a bit of control you can exercise over how segments are merged, here's a place to start: http://juanggrande.wordpress.com/2011/02/07/merge-policy-internals/ Merge Policy lets you control some of this behavior, but I'd still be nervous if I had less space on my disk than would allow a full copy of the index to be there for a while. Best Erick On Tue, Nov 1, 2011 at 12:46 AM, Jason Biggin wrote: > Wondering if anyone has experience with replicating large indexes. We have a > Solr deployment with 1 master, 1 master/slave and 5 slaves. Our index > contains 15+ million articles and is ~55GB in size. > > Performance is great on all systems. > > Debian Linux > Apache-Tomcat > 100GB disk > 6GB RAM > 2 proc > > on VMWare ESXi 4.0 > > > We notice however that whenever the master is optimized, the complete index > is replicated to the slaves. This causes a 100%+ bloat in disk requirements. > > Is this normal? Is there a way around this? > > Currently our optimize is configured as such: > > curl > 'http://localhost:8080/solr/update?optimize=true&maxSegments=1&waitFlush=true&expungeDeletes=true' > > Willing to share our experiences with Solr. > > Thanks, > Jason >
