[
https://issues.apache.org/jira/browse/SOLR-16242?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17551284#comment-17551284
]
Mikhail Khludnev commented on SOLR-16242:
-----------------------------------------
Linked PR has implementation and basic test coverage. I need go/no-go feedback
for it. So far I want to check how it works with cloud collections. And tackle
unifying classic and score join query under single abstract class via separate
issue. Right now they are both supported but it requires ugly instanceofs.
> Regenerate heavy joins in background
> ------------------------------------
>
> Key: SOLR-16242
> URL: https://issues.apache.org/jira/browse/SOLR-16242
> Project: Solr
> Issue Type: Improvement
> Security Level: Public(Default Security Level. Issues are Public)
> Components: query
> Reporter: Mikhail Khludnev
> Assignee: Mikhail Khludnev
> Priority: Major
>
> * When join queries specifies fromIndex result docset might be cached let's
> say in toIndex filter cache. Usually joins are heavy, take a while to
> execute. But after result docset is cached it respond fast.
> * if docs updated in fromIndex and commit occurs (in fromIndex) the next
> request (for the same join query) misses the toIndex filter cache and takes a
> while to regenerate this join query entry.
> * However, we might be ok to use old docset in toIndex if it doesn't harm
> query latency, and fine if changes in fromIndex will be joined little bit
> later.
> This issue proposes to introduce
> # \{!join cacheEventually=true ..}..
> # separate user cache for these join query docsets and
> # update processor to regenerate them explicitly
> /solr/fromIndex/update?stream.body=%3Ccommit/%3E&post-processor=refresh-join-caches
>
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@solr.apache.org
For additional commands, e-mail: issues-h...@solr.apache.org
