Details
-
Bug
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
-
None
-
None
Description
PROBLEM: When restarting the NodeManager via Ambari, the Ambari agent conducts
a recursive chown on the YARN local cache configuration directory.
Something like this: chown -R yarn:hadoop /hadoop/yarn/local
I have seen this in Ambari 2.0 and Ambari 2.0.1 - where 2.0.1 was upgraded
from 2.0
BUSINESS IMPACT: The problem is that user directories exist under
/hadoop/yarn/local/usercache/<username> and subsequent containers will not
have rights to the user's directory after the NM restart, in a secure cluster.
STEPS TO REPRODUCE: N/A
SUPPORT ANALYSIS: As seen in the Ambari logs for starting the NM:
u"Directory['/hadoop/yarn/local']"
{'group': 'hadoop', 'recursive': True, 'cd_access': 'a', 'ignore_failures': True, 'mode': 0775, 'owner': 'yarn¹}
Attachments
Issue Links
- links to