[LUCENE-8458] Carry-over hard-deletes after merge may not adjust soft-delete count

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 7.5, 8.0
• Fix Version/s: 7.5, 8.0
• Component/s: core/index
• Lucene Fields: New, Patch Available
• Assignee: Unassigned
• Reporter: Nhat Nguyen
• Created: 19/Aug/18 02:28
• Updated: 28/Aug/22 15:34
• Resolved: 24/Aug/18 13:15

Description

Attached is a test that can trip PendingDeletetes assertion around 5%.

The assertion is violated because we do not reduce soft-deletes count accordingly when carrying over hard-deletes after a merge in IndexWriter#carryOverHardDeletes. If the newly merged segment has soft-deleted documents, its PendingDeletes requires a segment reader to "transfer" soft-deletes count to hard-deletes accordingly.

testSoftDeleteWhileMergeSurvives (introduced in LUCENE-8293) always passes as a segment warmer used in that test forces ReadersAndUpdates to open a new segment reader.

NOTE: reproduce with: ant test  -Dtestcase=TestSoftDeletesRetentionMergePolicy -Dtests.method=testMixedSoftDeletesAndHardDeletes -Dtests.seed=FFED48B49B9F6AA5 -Dtests.slow=true -Dtests.badapples=true -Dtests.locale=zh-Hans -Dtests.timezone=Atlantic/South_Georgia -Dtests.asserts=true -Dtests.file.encoding=UTF-8
8月 19, 2018 12:11:10 上午 com.carrotsearch.randomizedtesting.RandomizedRunner$QueueUncaughtExceptionsHandler uncaughtException
警告: Uncaught exception in thread: Thread[Lucene Merge Thread #0,5,TGRP-TestSoftDeletesRetentionMergePolicy]
org.apache.lucene.index.MergePolicy$MergeException: java.lang.AssertionError: softDeleteCount doesn't match 21 != 19
	at __randomizedtesting.SeedInfo.seed([FFED48B49B9F6AA5]:0)
	at org.apache.lucene.index.ConcurrentMergeScheduler.handleMergeException(ConcurrentMergeScheduler.java:704)
	at org.apache.lucene.index.ConcurrentMergeScheduler$MergeThread.run(ConcurrentMergeScheduler.java:684)
Caused by: java.lang.AssertionError: softDeleteCount doesn't match 21 != 19
	at org.apache.lucene.index.PendingSoftDeletes.onNewReader(PendingSoftDeletes.java:87)
	at org.apache.lucene.index.ReadersAndUpdates.getReader(ReadersAndUpdates.java:173)
	at org.apache.lucene.index.ReadersAndUpdates.getLatestReader(ReadersAndUpdates.java:237)
	at org.apache.lucene.index.PendingSoftDeletes.ensureInitialized(PendingSoftDeletes.java:189)
	at org.apache.lucene.index.PendingSoftDeletes.isFullyDeleted(PendingSoftDeletes.java:200)
	at org.apache.lucene.index.ReadersAndUpdates.isFullyDeleted(ReadersAndUpdates.java:744)
	at org.apache.lucene.index.IndexWriter.isFullyDeleted(IndexWriter.java:5161)
	at org.apache.lucene.index.IndexWriter.commitMerge(IndexWriter.java:3926)
	at org.apache.lucene.index.IndexWriter.mergeMiddle(IndexWriter.java:4592)
	at org.apache.lucene.index.IndexWriter.merge(IndexWriter.java:4058)
	at org.apache.lucene.index.ConcurrentMergeScheduler.doMerge(ConcurrentMergeScheduler.java:625)
	at org.apache.lucene.index.ConcurrentMergeScheduler$MergeThread.run(ConcurrentMergeScheduler.java:662)

NOTE: reproduce with: ant test  -Dtestcase=TestSoftDeletesRetentionMergePolicy -Dtests.method=testMixedSoftDeletesAndHardDeletes -Dtests.seed=FFED48B49B9F6AA5 -Dtests.slow=true -Dtests.badapples=true -Dtests.locale=zh-Hans -Dtests.timezone=Atlantic/South_Georgia -Dtests.asserts=true -Dtests.file.encoding=UTF-8

com.carrotsearch.randomizedtesting.UncaughtExceptionError: Captured an uncaught exception in thread: Thread[id=16, name=Lucene Merge Thread #0, state=RUNNABLE, group=TGRP-TestSoftDeletesRetentionMergePolicy]

	at __randomizedtesting.SeedInfo.seed([FFED48B49B9F6AA5:B2667DCCD81812E2]:0)
Caused by: org.apache.lucene.index.MergePolicy$MergeException: java.lang.AssertionError: softDeleteCount doesn't match 21 != 19
	at __randomizedtesting.SeedInfo.seed([FFED48B49B9F6AA5]:0)
	at org.apache.lucene.index.ConcurrentMergeScheduler.handleMergeException(ConcurrentMergeScheduler.java:704)
	at org.apache.lucene.index.ConcurrentMergeScheduler$MergeThread.run(ConcurrentMergeScheduler.java:684)
Caused by: java.lang.AssertionError: softDeleteCount doesn't match 21 != 19
	at org.apache.lucene.index.PendingSoftDeletes.onNewReader(PendingSoftDeletes.java:87)
	at org.apache.lucene.index.ReadersAndUpdates.getReader(ReadersAndUpdates.java:173)
	at org.apache.lucene.index.ReadersAndUpdates.getLatestReader(ReadersAndUpdates.java:237)
	at org.apache.lucene.index.PendingSoftDeletes.ensureInitialized(PendingSoftDeletes.java:189)
	at org.apache.lucene.index.PendingSoftDeletes.isFullyDeleted(PendingSoftDeletes.java:200)
	at org.apache.lucene.index.ReadersAndUpdates.isFullyDeleted(ReadersAndUpdates.java:744)
	at org.apache.lucene.index.IndexWriter.isFullyDeleted(IndexWriter.java:5161)
	at org.apache.lucene.index.IndexWriter.commitMerge(IndexWriter.java:3926)
	at org.apache.lucene.index.IndexWriter.mergeMiddle(IndexWriter.java:4592)
	at org.apache.lucene.index.IndexWriter.merge(IndexWriter.java:4058)
	at org.apache.lucene.index.ConcurrentMergeScheduler.doMerge(ConcurrentMergeScheduler.java:625)
	at org.apache.lucene.index.ConcurrentMergeScheduler$MergeThread.run(ConcurrentMergeScheduler.java:662)

 

Issue Links

• links to: GitHub Pull Request #453