hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "jeongmin kim (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-13473) deleted cells come back alive after the stripe compaction
Date Thu, 16 Apr 2015 16:31:59 GMT

    [ https://issues.apache.org/jira/browse/HBASE-13473?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14498245#comment-14498245
] 

jeongmin kim commented on HBASE-13473:
--------------------------------------

public static final String FLUSH_TO_L0_KEY = "hbase.store.stripe.compaction.flushToL0";

I'm not 100% sure yet. but is this means that -
when FLUSH_TO_L0_KEY config is set to false (which is the default), 
 there's going to be no HFile on L0? 
so canDropDeletesWithoutL0 is always true?

I think it depends on this if we should assert hasAllFiles or do a minor compaction.

> deleted cells come back alive after the stripe compaction
> ---------------------------------------------------------
>
>                 Key: HBASE-13473
>                 URL: https://issues.apache.org/jira/browse/HBASE-13473
>             Project: HBase
>          Issue Type: Bug
>          Components: Compaction
>    Affects Versions: 2.0.0, 1.0.1, 1.1.0, 0.98.12
>            Reporter: jeongmin kim
>            Assignee: jeongmin kim
>            Priority: Blocker
>              Labels: patch
>         Attachments: HBASE-13473.patch, HBASE_13473-v2.patch
>
>
> during the STRIPE compaction,
> deletes(tombstones) are being dropped in 2 cases.
> 1. Compaction including L0  (includeL0 == true)
> 2. L0 has no files (canDropDeletesWithoutL0 == true)
> To drop delete marker and keep the consistency during compaction, 
> All of HFiles in the stripe has to be selected, just like major compaction.
> otherwise, after the compaction only delete markers would be gone, 
> and deleted cells (which is in the not-selected HFiles) are going to be alive again.
> In my cluster, there was no file on L0(canDropDeletesWithoutL0==true) and not all files
are selected for compaction, so some of deleted rows have come back alive and appears when
i get or scan after compactions.
> I made a patch about it.
> it checks if all files are selected before we set the majorRange of compaction request
.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message