[jira] [Commented] (SOLR-13668) Solr cores should release cache memory when inactive

["Erick Erickson (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SOLR-13668?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16897171#comment-16897171 ] 

Erick Erickson commented on SOLR-13668:
---------------------------------------

I'm a little iffy on this. One advantage with filling up the caches to the limit then keeping it at the limit is that unreasonable limits are more like to be found during testing. Here's the scenario:

I set up filterCache with a size of 1,000,000

I test for a while with something like fq=timestamp:[NOW-1DAY TO NOW]

I'm indexing a lot and bouncing the server so the cache never grows too large

I put it in production on an unchanging index and generate OOMs.

I've seen this case "in the wild" so it's not entirely theoretical.

Similar situations could occur with aging out entries, it works fine until the system comes under heavy load where all the cache entries are relatively young.

That said, reducing the memory pressure is always a good thing. Perhaps where we eventually end up is with some heuristics around dynamically resizing caches based on memory pressure or aging out cache entries more quickly based on memory pressure etc., in which case my worries evaporate.

> Solr cores should release cache memory when inactive
> ----------------------------------------------------
>
>                 Key: SOLR-13668
>                 URL: https://issues.apache.org/jira/browse/SOLR-13668
>             Project: Solr
>          Issue Type: Improvement
>      Security Level: Public(Default Security Level. Issues are Public) 
>            Reporter: Andrzej Bialecki 
>            Assignee: Andrzej Bialecki 
>            Priority: Major
>             Fix For: 8.3
>
>
> (This optimization has been identified also in SOLR-13579).
> Solr caches expand to fill the max limit of the cache without ever releasing the memory (unless a commit happens, in which case autowarming fills them again to some extent).
> It would be nice to be able to release some of that heap if the core has not had any query requests for a configured amount of time.



--
This message was sent by Atlassian JIRA
(v7.6.14#76016)

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