[
https://issues.apache.org/jira/browse/SOLR-7759?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17049447#comment-17049447
]
David Smiley commented on SOLR-7759:
------------------------------------
I'm royally confused. No matter what type of replica you have (NRT/TLOG/PULL),
a shard is a subset of the index as a whole and therefore might have skewed
stats, particularly for rare terms or if the sharding isn't purely uniform
(such as if you need to group types of documents together). Distributed-stats
is to ensure you get a global view. The replica type might influence how up to
date a specific replica is (NRT being fully up to date, TLOG & PULL not), and
maybe that could lead to out of date stats if distributed-stats talks to TLOG
or PULL replicas that are not leaders. But we're not discussing that here.
> DebugComponent's explain should be implemented as a distributed query
> ---------------------------------------------------------------------
>
> Key: SOLR-7759
> URL: https://issues.apache.org/jira/browse/SOLR-7759
> Project: Solr
> Issue Type: Bug
> Reporter: Varun Thacker
> Priority: Major
> Attachments: SOLR_7759.patch
>
>
> Currently when we use debugQuery to see the explanation of the matched
> documents, the query fired to get the statistics for the matched documents is
> not a distributed query.
> This is a problem when using distributed idf. The actual documents are being
> scored using the global stats and not per shard stats , but the explain will
> show us the score by taking into account the stats from the shard where the
> document belongs to.
> We should try to implement the explain query as a distributed request so that
> the scores match the actual document scores.
--
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
