[
https://issues.apache.org/jira/browse/SOLR-14904?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17205047#comment-17205047
]
David Smiley commented on SOLR-14904:
-------------------------------------
I'm thinking for a threshold use the smaller of documentCache.getMaxSize and
queryResultMaxDocsCached, considering the possibility of a -1 for either
(unsupported or unspecified) and use Integer.MAX_VALUE. I'd rather not
introduce yet another threshold configuration for minutia like this, and I
think this threshold algorithm seems safe/conservative. What we have today (no
threshold) has been there for all of Solr's existence.
Also, if you're specifying cursorMark beyond the first page, then we shouldn't
consult the documentCache either.
> Don't use documentCache for large result sets
> ---------------------------------------------
>
> Key: SOLR-14904
> URL: https://issues.apache.org/jira/browse/SOLR-14904
> Project: Solr
> Issue Type: Improvement
> Security Level: Public(Default Security Level. Issues are Public)
> Reporter: David Smiley
> Assignee: David Smiley
> Priority: Major
>
> Some users ask Solr to return many documents (high rows param), even though
> this is an anti-pattern. Sometimes there is some sense to it, and even Solr
> itself will do it in some cases like "bin/solr export" and perhaps some
> streaming-expressions cases. If there is a documentCache, these queries have
> a tendency to completely thrash it -- dump it and fill it with poor cache
> candidates. I've even seen the cache's existence for such queries become a
> bottleneck of the query -- granted for the now old LRUCache and in a
> particularly high abuse-case.
> I propose that if the number of documents to be returned is above some
> fraction of the documentCache's size limit, then don't use the documentCache
> at all. Maybe half size is sufficient? Or quarter-size? Maybe at least
> queryWindowSize big (thus at least 20 typically)? I see in solrconfig a
> queryResultMaxDocsCached option used for the queryResultCache but it could be
> made to apply to populating the documentCache as well. Code default is
> infinite but the default and most configs set to 200.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
