Hi Yago, When you create a new core (via API or Web UI), you specify the collection name and shard id, in my example "cloud" and "shard1_0". When the core initializes in SolrCloud mode, it recognizes that the collection exists and adds itself as a replica to the shard. Then the main replica recovery process kicks in; try PeerSync, realize too far out of date, try snapshot replication from leader. The following core API command led to the same result as using the UI:
curl -v "http://localhost:8986/solr/admin/cores?action=CREATE&collection=cloud&shard=shard1_0&name=cloud_shard1_0_replica3"; The only trick here is you need to set the name of the core, which from what I can tell can be arbitrary but I chose to use the same naming standard as the other cores Cheers, Timothy Potter Sr. Software Engineer, LucidWorks www.lucidworks.com ________________________________________ From: Yago Riveiro <yago.rive...@gmail.com> Sent: Monday, December 16, 2013 9:32 AM To: solr-user@lucene.apache.org Subject: Re: How can you move a shard from one SolrCloud node to another? Tim, Can you explain how the replication snapshot is done using the coreAdminAPI? -- Yago Riveiro Sent with Sparrow (http://www.sparrowmailapp.com/?sig) On Monday, December 16, 2013 at 4:23 PM, Tim Potter wrote: > Hi Chris, > > The easiest approach is to just create a new core on the new machine that > references the collection and shard you want to migrate. For example, say you > split shard1 of a collection named "cloud", which results in having: shard1_0 > and shard1_1. Now let's say you want to migrate shard 1_0 over to the new > machine. > > First, fire off a q=*:*&distrib=false query to the shard you're migrating so > that you know how many docs it has (which will be used to verify the > migration was clean below). > > Next, bring up the new machine in cloud mode (-zkHost=?) and then go to the > admin console on that server. Nav to the core admin page and create a new > core, specifying the collection and shard1_0 in the form; note: the form > leads you to believe you need to create the directory on the local system but > you actually don't need to worry about doing that as the config will get > pulled from ZK and the directory will get created on the fly (at least that's > what happened in my env using branch_4x). > > When the new core initializes, it will use good ol' snapshot replication to > pull the index from the leader. Verify the new core is happy by executing the > q=*:*&distrib=false query again. Once you're satisfied, you can unload the > core you migrated. > > Btw ... you can do all this with the core admin API instead of the Web UI if > you want to script it. > > Cheers, > > Timothy Potter > Sr. Software Engineer, LucidWorks > www.lucidworks.com (http://www.lucidworks.com) > > ________________________________________ > From: cwhi <chris.whi...@gmail.com (mailto:chris.whi...@gmail.com)> > Sent: Sunday, December 15, 2013 3:43 PM > To: solr-user@lucene.apache.org (mailto:solr-user@lucene.apache.org) > Subject: How can you move a shard from one SolrCloud node to another? > > Let's say I want to rebalance a SolrCloud collection. I call SPLITSHARD to > split an existing shard, and then I'd like to move one of the subshards to a > new machine so the index is more balanced. Can this be done? If not, how > do you rebalance an existing SolrCloud collection? > > > > -- > View this message in context: > http://lucene.472066.n3.nabble.com/How-can-you-move-a-shard-from-one-SolrCloud-node-to-another-tp4106815.html > Sent from the Solr - User mailing list archive at Nabble.com > (http://Nabble.com). > >
