[
https://issues.apache.org/jira/browse/PHOENIX-7591?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17950348#comment-17950348
]
Tanuj Khurana commented on PHOENIX-7591:
----------------------------------------
One scenario where data inconsistency can happen is when there are multiple
batches trying to concurrently update the same row and the first batch makes
progress but the second batch update fails to obtain the row locks in a timely
manner and gets a timeout. Then the third concurrent batch will generate the
index mutations based on the data row state read from the disk and not from the
data row state in the memory generated by the first batch.
> Concurrent updates to tables with indexes can cause data inconsistency
> ----------------------------------------------------------------------
>
> Key: PHOENIX-7591
> URL: https://issues.apache.org/jira/browse/PHOENIX-7591
> Project: Phoenix
> Issue Type: Improvement
> Affects Versions: 5.3.0
> Reporter: Tanuj Khurana
> Assignee: Tanuj Khurana
> Priority: Major
>
> testConcurrentUpserts#ConcurrentMutationsExtendedIT andÂ
> testConcurrentUpsertsWithoutIndexedColumns#ConcurrentUpsertsWithoutIndexedColsIT
> have become flakey. Most likely, this has happened after
> [PHOENIX-7543|https://issues.apache.org/jira/browse/PHOENIX-7543]
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
