[
https://issues.apache.org/jira/browse/HBASE-1724?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12737081#action_12737081
]
Jean-Daniel Cryans commented on HBASE-1724:
-------------------------------------------
FYI, there will always be some level of dataloss in HBase as long as the
appends in HDFS are not fully functional (suppose to be ready for 0.21). Until
that time there are some knobs you can play with but the situation you
described will at least not happen with the latest fixes. For example, we were
not rolling the logs so you could easily lost data (as it did to you) after
even days. Now it happens every hour and it is configurable. Also there is a
max size a log can have until it is rolled.
> Data loss after `kill -9` region server
> ---------------------------------------
>
> Key: HBASE-1724
> URL: https://issues.apache.org/jira/browse/HBASE-1724
> Project: Hadoop HBase
> Issue Type: Bug
> Components: regionserver
> Affects Versions: 0.19.3
> Environment: OpenSUSE 11.1, HBase 0.19.3, Hadoop HDFS 0.19.1
> Reporter: Kirill Balyasnikov
> Attachments: log_fragment.txt
>
>
> I have a 3 node cluster setup each running hadoop and hbase. I have created
> 'accounts' table and loaded some data into it (about 3000 rows).
> Some days later one of the region servers died and after i restarted it there
> were no records in the table at all. I saw HLOG file with my records
> in HDFS but as I understand the file was not used by HBase to recover the
> table.
> I tried to emulate the situation and uploaded another 2000 records into my
> table and killed the region server holding 'accounts' table region.
> In the HDFS I found file with my records but some time later it was replaced
> by another empty directory. As i suspected after killed region
> server startup the data was not recovered.
> Everything is lost again and there is no any exceptions in the logs...
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
