Michael McCandless created LUCENE-9405:
------------------------------------------
Summary: IndexWriter incorrectly calls closeMergeReaders twice
when the merged segment is 100% deleted
Key: LUCENE-9405
URL: https://issues.apache.org/jira/browse/LUCENE-9405
Project: Lucene - Core
Issue Type: Bug
Components: core/index
Reporter: Michael McCandless
This is the first spinoff from a [controversial PR to add a new index-time
feature to Lucene to merge small segments during
commit|https://github.com/apache/lucene-solr/pull/1552]. This can
substantially reduce the number of small index segments to search.
See specifically [this discussion
there|https://github.com/apache/lucene-solr/pull/1552#discussion_r440298695].
{{IndexWriter}} seems to be missing a {{success = true}} inside {{mergeMiddle}}
in the case where all segments being merged have 100% deletions and the
segments will simply be dropped.
In this case, in master today, I think we are incorrectly calling
{{closeMergedReaders}} twice, first with {{suppressExceptions = false}} and
second time with {{true}}.
There is a [dedicated test case here showing the
issue|https://github.com/apache/lucene-solr/commit/cab5ef5e6f2bdcda59fd669a298ec1377777af9d],
but that test case relies on changes in the controversial feature (added
{{MergePolicy.findFullFlushMerges}}). I think it should be possible to make
another test case show the bug without that controversial feature, and I am
unsure why our existing randomized tests have not uncovered this yet ...
--
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
