You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hive.apache.org by "zhangbutao (Jira)" <ji...@apache.org> on 2022/01/22 07:40:00 UTC
[jira] [Commented] (HIVE-25401) Insert overwrite a table which location is on other cluster fail in kerberos cluster
[ https://issues.apache.org/jira/browse/HIVE-25401?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17480361#comment-17480361 ]
zhangbutao commented on HIVE-25401:
-----------------------------------
I think you can try this parameter and its value is mutiple cluster deFaultFS :
<property>
<name>mapreduce.job.hdfs-servers</name>
<value>hdfs://cluster1,hdfs://cluster2</value>
</property>
> Insert overwrite a table which location is on other cluster fail in kerberos cluster
> --------------------------------------------------------------------------------------
>
> Key: HIVE-25401
> URL: https://issues.apache.org/jira/browse/HIVE-25401
> Project: Hive
> Issue Type: Bug
> Components: Hive
> Affects Versions: 2.3.0, 3.1.2
> Environment: hive 2.3
> hadoop3 cluster with kerberos
> Reporter: Max Xie
> Assignee: Max Xie
> Priority: Minor
> Labels: pull-request-available
> Attachments: HIVE-25401.patch, image-2021-07-29-14-25-23-418.png
>
> Time Spent: 1.5h
> Remaining Estimate: 0h
>
> we have tow hdfs clusters with kerberos security, it means that mapreduce task need delegation tokens to authenticate namenode when hive on mapreduce run.
> Insert overwrite a table which location is on other cluster fail in kerberos cluster. For example,
> # yarn cluster's default fs is hdfs://cluster1
> # tb1's location is hdfs://cluster1/tb1
> # tb2's location is hdfs://cluster2/tb2
> # sql `INSERT OVERWRITE TABLE tb2 SELECT * from tb1` run on yarn cluster will fail
>
> reduce task error log:
> !image-2021-07-29-14-25-23-418.png!
> How to fix:
> After dig it, web found mapreduce job just obtain delegation tokens for input files in FileInputFormat. But Hive context get extendal scratchDir base on table's location, If the table 's location is on other cluster, the delegation token will not be obtained.
> So we need to obtaine delegation tokens for hive scratchDirs before hive submit mapreduce job.
>
> How to test:
> no test
>
>
--
This message was sent by Atlassian Jira
(v8.20.1#820001)