[
https://issues.apache.org/jira/browse/HBASE-10546?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13902900#comment-13902900
]
Lars Hofhansl commented on HBASE-10546:
---------------------------------------
Also see HBASE-6069, which *added* the initialize we're removing now (to avoid
an NPE with Hive).
With that in mind, we should not remove the initialize() calls.
Let's just fix restart(...)
> Two scanner objects are open for each hbase map task but only one scanner
> object is closed
> ------------------------------------------------------------------------------------------
>
> Key: HBASE-10546
> URL: https://issues.apache.org/jira/browse/HBASE-10546
> Project: HBase
> Issue Type: Bug
> Reporter: Vasu Mariyala
> Fix For: 0.96.2, 0.98.1, 0.99.0, 0.94.17
>
> Attachments: 0.94-HBASE-10546.patch, trunk-HBASE-10546.patch
>
>
> Map reduce framework calls createRecordReader of the
> TableInputFormat/MultiTableInputFormat to get the record reader instance. In
> this method, we are initializing the TableRecordReaderImpl (restart method).
> This initializes the scanner object. After this, map reduce framework calls
> initialize on the RecordReader. In our case, this calls restart of the
> TableRecordReaderImpl again. Here, it doesn't close the first scanner. At the
> end of the task, only the second scanner object is closed. Because of this,
> the smallest read point of HRegion is affected.
> We don't need to initialize the RecordReader in the createRecordReader method
> and we need to close the scanner object in the restart method. (incase if the
> restart method is called because of exceptions in the nextKeyValue method)
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
