This doesn't seem right. A leader will ask a replica to recover only when an update request could not be forwarded to it. Can you check your leader logs to see why updates are not being sent through to replicas?
On Fri, Jun 21, 2013 at 7:03 AM, Sven Stark <sven.st...@m-square.com.au> wrote: > Hello, > > first: I am pretty much a Solr newcomer, so don't necessarily assume basic > solr knowledge. > > My problem is that in my setup SolrCloud seems to create way too much > network traffic for replication. I hope I'm just missing some proper config > options. Here's the setup first: > > * I am running a five node SolrCloud cluster on top of an external 5 node > zookeeper cluster, according to logs and clusterstate.json all nodes find > each other and are happy > * Solr version is now 4.3.1, but the problem also existed on 4.1.0 ( I > thought upgrade might solve the issue because of > https://issues.apache.org/jira/browse/SOLR-4471) > * there is only one shard > * solr.xml and solrconfig.xml are out of the box, except for the enabled > soft commit > > <autoSoftCommit> > <maxTime>1000</maxTime> > </autoSoftCommit> > > * our index is minimal at the moment (dev and testing stage) 20-30Mb, about > 30k small docs > > The issue is when I run smallish load tests against our app which posts ca > 1-2 docs/sec to solr, the SolrCloud leader creates outgoing network traffic > of 20-30Mbyte/sec and the non-leader receive 4-8MByte/sec each. > > The non-leaders logs are full of entries like > > INFO - 2013-06-21 01:08:58.624; > org.apache.solr.handler.admin.CoreAdminHandler; It has been requested that > we recover > INFO - 2013-06-21 01:08:58.640; > org.apache.solr.handler.admin.CoreAdminHandler; It has been requested that > we recover > INFO - 2013-06-21 01:08:58.643; > org.apache.solr.handler.admin.CoreAdminHandler; It has been requested that > we recover > INFO - 2013-06-21 01:08:58.651; > org.apache.solr.handler.admin.CoreAdminHandler; It has been requested that > we recover > INFO - 2013-06-21 01:08:58.892; > org.apache.solr.handler.admin.CoreAdminHandler; It has been requested that > we recover > INFO - 2013-06-21 01:08:58.893; > org.apache.solr.handler.admin.CoreAdminHandler; It has been requested that > we recover > > So my assumption is I am making config errors and the cloud leader tries to > push the index to all non-leaders over and over again. But I couldn't > really find much doco on how to properly configure SolrCloud replication > online. > > Any hints and help much appreciated. I can provide more info or data, just > let me know what you need. > > Thanks in advance, > Sven -- Regards, Shalin Shekhar Mangar.
