[
https://issues.apache.org/jira/browse/LUCENE-6185?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14279068#comment-14279068
]
Michael McCandless commented on LUCENE-6185:
--------------------------------------------
Hmm one problem with TopDocs.merge is that it doesn't re-base the docIDs.
Instead, it sets shardIndex for each hit. I think this patch should sometimes
fail tests, when newSearcher swaps in an ExecutorService?
TopDocs.merge does this so that you can merge across indices that sum to > 2.1B
docs. But in this usage, the number of docs will be < 2.1B ... so maybe we
need an option to TopDocs.merge to rebase? Or we rebase afterwards in
IndexSearcher?
> Fix IndexSearcher with threads to not collect documents out of order
> --------------------------------------------------------------------
>
> Key: LUCENE-6185
> URL: https://issues.apache.org/jira/browse/LUCENE-6185
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Adrien Grand
> Assignee: Adrien Grand
> Priority: Blocker
> Fix For: 5.0, Trunk
>
> Attachments: LUCENE-6185.patch
>
>
> When created with an executor, IndexSearcher searches all leaves in a
> different task and eventually merges the results when all tasks are
> completed. However, this merging logic involves a TopFieldCollector which is
> collected out-of-order. I think it should just use TopDocs.merge?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
