hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ashu Pachauri (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (HBASE-17117) Reversed scan returns deleted versions and breaks RegionLocator
Date Sat, 19 Nov 2016 00:43:58 GMT

     [ https://issues.apache.org/jira/browse/HBASE-17117?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel

Ashu Pachauri updated HBASE-17117:
    Attachment: reverseScanMetaDeleteFamily

reverseScanMetaDeleteFamily : Output of raw, forward and reversed scan on meta during the
I have replaced the start and end keys with dummy values (but in correct order).

We did not back up the store files for meta, and I am unable to reproduce the issue just by
creating a delete family marker in meta and doing a reverse scan. This is not happening consistently
during splits either.  So, I am dropping the priority down to major from blocker. I'll try
to research edge cases in the scanner code and come up with a unit test for reproducing this.
Any ideas are welcome.

> Reversed scan returns deleted versions and breaks RegionLocator
> ---------------------------------------------------------------
>                 Key: HBASE-17117
>                 URL: https://issues.apache.org/jira/browse/HBASE-17117
>             Project: HBase
>          Issue Type: Bug
>          Components: Scanners
>    Affects Versions: 1.3.0
>            Reporter: Ashu Pachauri
>            Priority: Blocker
>             Fix For: 1.3.0
>         Attachments: reverseScanMetaDeleteFamily
> We started seeing clients persistently throwing errors as they were trying to talk to
a region that was non existent (split a few days ago). We verified that the region was deleted
from meta when the split happened.
> On performing a raw scan on meta, the deleted version for the split region appears, which
also does on performing a normal reversed scan. Since MetaScanner uses a reversed scan, this
explains why clients see non existent regions.
> We also verified that there was no in-memory corrupt state by failing over the master.
When we trigger major compaction on meta, the problem goes away further confirming the fact
that we were seeing deleted versions.

This message was sent by Atlassian JIRA

View raw message