[ 
https://issues.apache.org/jira/browse/SOLR-5025?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13704500#comment-13704500
 ] 

Steve Rowe commented on SOLR-5025:
----------------------------------

+1

Another implementation possibility: Re-partition the hash ring with the desired 
number of shards, then split each existing shard along the new partition 
boundaries.  This would require:

* the ability to split shards at arbitrary points in the hash ring, creating 
unequally sized sub-shards
* the ability to split shards into more than two (SOLR-5004)
* merging new shards whose target section of the new hash ring partition falls 
across old shard boundary/ies

In addition to growing the number of shards in a collection, this strategy 
could also be used to *reduce* the number of shards.
                
> Implement true re-sharding for SolrCloud
> ----------------------------------------
>
>                 Key: SOLR-5025
>                 URL: https://issues.apache.org/jira/browse/SOLR-5025
>             Project: Solr
>          Issue Type: Wish
>          Components: SolrCloud
>            Reporter: Shawn Heisey
>
> Shard splitting is an incredibly nice thing to have, but it doesn't 
> completely address the idea of re-sharding.
> Let's say that you currently have three shards, only your index is three or 
> four times as big as you ever expected it to get when you first built it.  
> You've added nodes, which helps, but doesn't address the fundamental fact 
> that each of your shards is too big for an individual server.  If you had 
> created eight shards up front, everything would be smooth.  It's not possible 
> with shard splitting to go from three equal size shards to eight equal size 
> shards.
> A new feature to accomplish true re-sharding would solve this.  One 
> implementation possibility:  Create a new collection with the new numShards, 
> split all the documents accordingly to the new replicas, then rename/swap the 
> collection and core names.
> There are a number of sticky points to iron out regardless of the 
> implementation method chosen, some of which could be really hairy.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to