[
https://issues.apache.org/jira/browse/LUCENE-10060?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17405269#comment-17405269
]
Adrien Grand commented on LUCENE-10060:
---------------------------------------
bq. I assume what's going on here is that the caching ignores deleted docs and
expects they'll be handled later
This is correct. The query cache caches based on the so-called "segment core"
(see SegmentReader#core), ie. everything but live docs and doc-value updates.
This allows cache entries to be reused across reopens as long as a segment
hasn't been merged away. But this requires cache entries to be computed on the
entire segment and follow-up searches to apply deletes on top of cache entries.
DrillSidewaysQuery is a bit abusing the API, it shouldn't be a Query. It would
be nice to refactor it to something else than a Query at some point.
> Ensure DrillSidewaysQuery instances don't get cached
> ----------------------------------------------------
>
> Key: LUCENE-10060
> URL: https://issues.apache.org/jira/browse/LUCENE-10060
> Project: Lucene - Core
> Issue Type: Bug
> Components: modules/facet
> Reporter: Greg Miller
> Priority: Minor
> Fix For: main (9.0), 8.10
>
> Time Spent: 1h
> Remaining Estimate: 0h
>
> We need to make sure DSQ instances don't end up in the query cache. -It's
> important that the {{DrillSidewaysScorer}} (bulk scorer implementation)
> actually runs during query evaluation in order to populate the "sideways"
> {{FacetsCollector}} instances with "near miss" docs. If it gets cached, this
> won't happen.-
> There may also be an implication around {{acceptDocs}} getting honored as
> well. [~zacharymorn] may be able to provide more details.
> UPDATE: The original issue I detailed above isn't actually an issue since
> {{DrillDownQuery}} doesn't implement {{equals}}, so the cache always misses
> and it always executes the {{BulkScorer}} ( {{DrillSidewaysScorer}} ).
> Tricky! There is a separate issue found by Zach (as mentioned above) related
> to "acceptDocs" though. See below conversation and link off to the separate
> [PR
> conversation|https://github.com/apache/lucene/pull/240#discussion_r692154001]
> for more details.
--
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
