You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@lucene.apache.org by "Alan Woodward (JIRA)" <ji...@apache.org> on 2016/10/17 16:17:58 UTC

[jira] [Commented] (SOLR-9566) Can we avoid doing recovery when collections are first created?

    [ https://issues.apache.org/jira/browse/SOLR-9566?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15582651#comment-15582651 ] 

Alan Woodward commented on SOLR-9566:
-------------------------------------

I got a verbal +1 from [~markrmiller@gmail.com] at Revolution, so I'll commit this shortly.  Anybody who disagrees should speak up now...

> Can we avoid doing recovery when collections are first created?
> ---------------------------------------------------------------
>
>                 Key: SOLR-9566
>                 URL: https://issues.apache.org/jira/browse/SOLR-9566
>             Project: Solr
>          Issue Type: Improvement
>      Security Level: Public(Default Security Level. Issues are Public) 
>            Reporter: Alan Woodward
>         Attachments: SOLR-9566.patch
>
>
> When a core starts up as part of a collection, and it's not a shard leader, it goes into recovery, part of which involves a 7 second wait (set in SOLR-7141) to ensure that updates being sent from the leader don't get missed in between buffering and replication.
> This has the unfortunate side-effect of adding a 7-second pause to collection creation whenever the replication factor is 2 or more, which slows down tests massively - for example, DeleteReplicaTest takes about 54 seconds to execute on my machine, 28 seconds of which is just pauses - over 50% of execution time.  It's not actually possible to add documents to a collection before the creation request has returned, so the recovery stage here isn't strictly speaking necessary.  
> I think we could try adding a parameter to a CoreAdmin create request that says the core is being created as part of a new collection, so it doesn't need to try and recover from it's leader when it starts up.  Does this sound sensible, or am I missing something here?



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org