[
https://issues.apache.org/jira/browse/HBASE-4065?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13088320#comment-13088320
]
Brock Noland commented on HBASE-4065:
-------------------------------------
In the mapred package, HTable is created inside getRecordWriter where it can
throw an IOE. In mapreduce HTable is created in setConf where an IOE cannot be
thrown. I am fine with moving the HTable creation to getRecordWriter like
mapred, but that still leaves us with the fact that
ZKUtil.applyClusterKeyToConf throws IOE.
My assumption was that an IOE from ZKUtil.applyClusterKeyToConf should be
considered fatal as well? That may certainly not be true and we could simply
move HTable creation to getRecordWriter.
> TableOutputFormat ignores failure to create table instance
> ----------------------------------------------------------
>
> Key: HBASE-4065
> URL: https://issues.apache.org/jira/browse/HBASE-4065
> Project: HBase
> Issue Type: Bug
> Affects Versions: 0.90.3
> Reporter: Todd Lipcon
> Assignee: Brock Noland
> Fix For: 0.94.0
>
> Attachments: HBASE-4065.1.patch
>
>
> If TableOutputFormat in the new API fails to create a table, it simply logs
> this at ERROR level and then continues on its way. Then, the first write() to
> the table will throw a NPE since table hasn't been set.
> Instead, it should probably rethrow the exception as a RuntimeException in
> setConf, or do what the old-API TOF does and not create the HTable instance
> until getRecordWriter, where it can throw an IOE.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
