[
https://issues.apache.org/jira/browse/HIVE-12058?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14953534#comment-14953534
]
Yongzhi Chen commented on HIVE-12058:
-------------------------------------
[~jxiang], when the hbase mapredcp fails, without 2>/dev/null, the error will
go to stderr.log or console in our system, it will not become input for the for
loop, so remving "2>/dev/null" does no harm. Hive still can startup. I think
the only thing 2>/dev/null does is hiding the failure.
With/Without the 2>/dev/null, when hbase mapredcp fails, there is no hbase
related library added to the HADOOP_CLASSPATH. So there is
no functional change in Hive server by removing the redirection, and we get the
error log to analyze the hbase failure for extra.
> Change hive script to record errors when calling hbase fails
> ------------------------------------------------------------
>
> Key: HIVE-12058
> URL: https://issues.apache.org/jira/browse/HIVE-12058
> Project: Hive
> Issue Type: Bug
> Components: Hive, HiveServer2
> Affects Versions: 0.14.0, 1.1.0, 2.0.0
> Reporter: Yongzhi Chen
> Assignee: Yongzhi Chen
> Attachments: HIVE-12058.1.patch
>
>
> By default hive will try to find out which jars need to be added to the
> classpath in order to run MR jobs against an HBase cluster, however if hbase
> can't be found or if hbase mapredcp fails, the hive script will fail
> silently and ignore some of the jars to be included into the. That makes very
> difficult to analyze the real problem.
> Hive script should record the error not just simply redirect two hbase
> failures:
> HBASE_BIN=$
> {HBASE_BIN:-"$(which hbase 2>/dev/null)"}
> $HBASE_BIN mapredcp 2>/dev/null
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
