Here are more details about our setup: Zookeeper: * 3 separate hosts in same rack as Solr cluster * Zookeeper hosts do not run any other processes
Solr: * total servers: 24 (plus 2 cold standbys in case of host failure) * physical memory: 65931872 kB (62 GB) * max JVM heap size: -Xmx10880m ( 10 GB) * only one Solr per host On the Œindex¹ directory size front, I am seeing some differences in the disk usage between leaders and replicas. In 1 / 12 shards, there is no difference in size between the leader and replica. In 6 / 12 shards, there 1 1 G difference in size between the leader and replica. Both have one index directory. In 5 / 12 shards, the replica is a multiple of the leader size, due to multiple index directories on disk. For example, shard 1 leader has a directory named 'index.20140624071707699¹ 30 G in size. The replica has two directories: 'index.20150108052156468¹ at 31G and Œindex.20140624071556270¹ at 32G. Thanks, Lindsay On 2015-01-09, 5:01 PM, "Shawn Heisey" <apa...@elyograg.org> wrote: >On 1/9/2015 4:54 PM, Lindsay Martin wrote: >> I am experiencing a problem where Solr nodes go into recovery following >>an update cycle. > ><snip> > >> For background, here are some details about our configuration: >> * Solr 4.10.2 (problem also observed with Solr 4.6.1) >> * 12 shards with 2 nodes per shard >> * a single updater running in a separate subnet is posting updates >>using the SolrJ CloudSolrServer client. Updates are triggered hourly. >> * system is under continuous query load >> * autoCommit is set to 821 seconds >> * autoSoftCommit is set to 303 seconds > >I would suspect some kind of performance problem that likely results in >the zkClientTimeout expiring. I have a standard set of questions for >performance problems. > >Questions about zookeeper: > >How many ZK nodes? Is zookeeper on separate hardware? If it's on the >same hardware as Solr, is its database on the same disk spindles as the >Solr index, or separate spindles? Is zookeeper standalone or embedded >in Solr? If it's standalone, do you happen to know the java max heap >for the zookeeper processes? > >Questions about Solr and the hardware: > >How many total Solr servers? How much RAM is installed on each one? >What is the max size of the Java heap? Are you running more than one >Solr (JVM/container) instance per machine? > >If you add up all the "index" directories on a server, how much disk >space does it take? Is the amount of disk space used similar on all of >the servers? > >Thanks, >Shawn >