[HDDS-5019] Inconsistent tmp bucket name on different nodes when Kerberos is enabled

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 1.1.0
• Labels: pull-request-available
• Assignee: Chenyu Zheng
• Reporter: Chenyu Zheng
• Created: 23/Mar/21 10:08
• Updated: 31/Mar/21 20:45
• Resolved: 31/Mar/21 19:55

Description

I setup yarn cluster on ozone, and kerberos is enable. NodeManager can't download resource package in ContainerLocalizer stage, show error like this

2021-03-22 19:56:42,004 WARN org.apache.hadoop.yarn.server.nodemanager.containermanager.localizer.ResourceLocalizationService: \{ ofs://test1/tmp/hadoop-yarn/staging/$user/.staging/job_1616412423610_0008/job.splitmetainfo, 1616414199202, FILE, null } failed: hadoop-yarn/staging/$user/.staging/job_1616412423610_0008/job.splitmetainfo: No such file or directory!
java.io.FileNotFoundException: hadoop-yarn/staging/$user/.staging/job_1616412423610_0008/job.splitmetainfo: No such file or directory!

 

The reason is that this wrong way to get user. For kerberos principal in different machine are not same, so UserGroupInformation.getCurrentUser().getUserName() are not same. Because kerberos principal's formate is "user/[e-mail]"
 

Issue Links

• relates to: HDDS-5041 Use getShortUserName in getTrashRoot(s)
• links to: GitHub Pull Request #2075