Maybe it would be better if Solr checked the live nodes and not all the existing nodes in zk. If a server dies and you need to start a new one, it would go straight to the correct shard without one needing to specify it manually. Of course, the problem could be if a server goes down for a minute and then comes back up, maybe a new node was added to the shard in the interim, but I still think it would be better this way.
Tomás On Mon, Nov 19, 2012 at 1:51 PM, Mark Miller <markrmil...@gmail.com> wrote: > > On Nov 19, 2012, at 11:24 AM, joe.cohe...@gmail.com wrote: > > > Hi > > I have the following scenario: > > I have 1 collection across 10 servers. Num of shards: 10. > > Each server has 2 solr instances running. replication is 2. > > > > I want to move one of the instances to another server. meaning, kill the > > solr process in server X and start a new solr process in server Y > instead. > > When I kill the solr process in server X, I can still see that instance > in > > the solr-cloud-graph (marked differently). > > When I run the instance on server Y, it get attahced to another shard, > > instead of getting into the shard that is now actually missing an > instance. > > > > 1. Any way to tell solr/zookeeper - "Forget about that instance"? > > Unload the SolrCores involved. > > > 2. when running a new solr instance - any way to tell solr/zookeper - > "add > > this instance to shard X"? > > Specify a shardId when creating the core or configuring it in solr.xml and > make it match the shard you want to add to. > > - Mark > >
