[
https://issues.apache.org/jira/browse/PHOENIX-3149?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15408672#comment-15408672
]
Sergey Soldatov commented on PHOENIX-3149:
------------------------------------------
It's interesting question. To get it 100% reproducible may be tricky. In my
setup I had to use very small memstore size to increase the number of HFiles,
load data using csv bulkload and run create index with relatively large number
of columns (to get size of index big) + flush + split immediately. I will
check whether it's possible to fit it into IT test.
> Local index got corrupted if intermediate compaction happen during the split.
> -----------------------------------------------------------------------------
>
> Key: PHOENIX-3149
> URL: https://issues.apache.org/jira/browse/PHOENIX-3149
> Project: Phoenix
> Issue Type: Bug
> Affects Versions: 4.8.0
> Reporter: Sergey Soldatov
> Assignee: Sergey Soldatov
> Attachments: PHOENIX-3149.patch
>
>
> The scenario is quite simple:
> If we trigger split after index creation when we have a lot of HFiles not yet
> compacted, the compaction may happen for part of those files and the scan
> type will be COMPACT_RETAIN_DELETES, but not COMPACT_DROP_DELETES as we
> expect in preCompactScannerOpen. In this case the regular scanner will be
> used during compaction and we get incorrectly data in the local index.
> Discussed it with [~enis] offline and it seems that the check for
> COMPACT_DROP_DELETES is not necessary and can/should be deleted.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
