[ 
https://issues.apache.org/jira/browse/SOLR-12233?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442577#comment-16442577
 ] 

Shawn Heisey commented on SOLR-12233:
-------------------------------------

bq. In our use case we're loading some 60,000 solrcore objects

That many cores on one machine is going to present a LOT of scalability issues, 
some of which cannot be improved, even when using transient cores.  Anything 
you can do to decrease this number is going to help.  One idea is to combine 
data from multiple cores into one core, and use filtering on fields in the 
document.  Another is to add more servers, so there are fewer cores on each one.

Sharing internal objects like this issue aims to do will help, but that can't 
be done with everything.  A SolrCore instance is not a small object, and takes 
a little bit of time to initialize.

bq. the other being creating the handlers which is a whole other problem 

There are a lot of implicit request handlers created even without 
configuration.  Quite a lot of them were added after the 4.x versions.  It 
might be that we need a way to disable implicit handlers when the user is 
facing scalability issues and is absolutely certain that they won't ever need 
some of them.

https://lucene.apache.org/solr/guide/7_3/implicit-requesthandlers.html


> QParserPlugin maintains a list of classes recreated every time a Solrcore 
> object is created
> -------------------------------------------------------------------------------------------
>
>                 Key: SOLR-12233
>                 URL: https://issues.apache.org/jira/browse/SOLR-12233
>             Project: Solr
>          Issue Type: Bug
>      Security Level: Public(Default Security Level. Issues are Public) 
>    Affects Versions: 7.1.1
>            Reporter: Jeff Miller
>            Assignee: Erick Erickson
>            Priority: Minor
>              Labels: Performance, qparserplugin
>          Time Spent: 10m
>  Remaining Estimate: 0h
>
> QParserPlugin maintains a static map of Class Names to Class objects and 
> everytime we create a SolrCore object this causes a lot of overhead doing 
> classloader lookups.  Our system runs a lot of cores and the classloader gets 
> bogged down when a lot of threads are creating solrcore objects.  
> There's no need to create these objects every time, similar classes such as 
> TransformerFactory store the object one time and reference it over and over 
> again



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

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

Reply via email to