[
https://issues.apache.org/jira/browse/LUCENE-10500?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17517693#comment-17517693
]
ASF subversion and git services commented on LUCENE-10500:
----------------------------------------------------------
Commit 796a19b457691249b0f3c2fc837f82a378d1d2c6 in lucene's branch
refs/heads/main from Luca Cavanna
[ https://gitbox.apache.org/repos/asf?p=lucene.git;h=796a19b4576 ]
LUCENE-10500: StringValueFacetCounts to not rely on sequential collection (#788)
StringValueFacetCounts should use the segment ordinal instead of the current
index when looping through the matching hits, as when search is multi-threaded
the order of the matching hits (one per segment) is not deterministic.
> StringValueFacetCounts relies on sequential collection
> ------------------------------------------------------
>
> Key: LUCENE-10500
> URL: https://issues.apache.org/jira/browse/LUCENE-10500
> Project: Lucene - Core
> Issue Type: Bug
> Reporter: Luca Cavanna
> Priority: Major
> Time Spent: 20m
> Remaining Estimate: 0h
>
> We recently moved some of the facets tests to use IndexSearcher#search(Query,
> CollectorManager) providing a FacetsCollectorManager instead of a
> FacetsCollector. Whenever newIndexSearcher(IndexReader) is used in tests,
> concurrent search may now be exercised while it was not before.
> This caused some build failures on TestStringValueFacetCounts:
> {code:java}
> java.lang.ArrayIndexOutOfBoundsException: Index 1 out of bounds for length 1
> at
> __randomizedtesting.SeedInfo.seed([ED8BF8281FCE5C02:9FC7DD27AEAEEA71]:0)
> at
> org.apache.lucene.core@10.0.0-SNAPSHOT/org.apache.lucene.util.packed.Packed64.get(Packed64.java:81)
> at
> org.apache.lucene.core@10.0.0-SNAPSHOT/org.apache.lucene.index.OrdinalMap$2.get(OrdinalMap.java:346)
> at
> org.apache.lucene.facet.StringValueFacetCounts.countOneSegment(StringValueFacetCounts.java:440)
> at
> org.apache.lucene.facet.StringValueFacetCounts.count(StringValueFacetCounts.java:295)
> at
> org.apache.lucene.facet.StringValueFacetCounts.<init>(StringValueFacetCounts.java:123)
> at
> org.apache.lucene.facet.TestStringValueFacetCounts.checkFacetResult(TestStringValueFacetCounts.java:349)
> at
> org.apache.lucene.facet.TestStringValueFacetCounts.testRandom(TestStringValueFacetCounts.java:325)
> {code}
> This looks like a real bug, as StringValueFacetCounts#countOneSegment is
> called once providing the index of the current loop instead of the ordinal
> taken from the matching hits that we are analyzing. That works fine with
> single threaded collection as we will go sequentially and the two indices
> will always be the same. With multi-threaded search, the order of the
> returned matching hits (one per segment) is not deterministic.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
