Ethan Li created STORM-3494:
-------------------------------
Summary: Use UserGroupInformation to login to HDFS only once per
process
Key: STORM-3494
URL: https://issues.apache.org/jira/browse/STORM-3494
Project: Apache Storm
Issue Type: Improvement
Reporter: Ethan Li
Assignee: Ethan Li
UserGroupInformation (UGI) loginUserFromKeytab should be used only once in a
process to login to hdfs because it overrides static fields. Also
loginUserFromKeytabAndReturnUGI function is also problematic according to
hadoop team. So the correct way to connect to hdfs is to use UGI
loginUserFromKeytab once and only in a process.
Currently we only use HDFS in hdfs-blobstore. It works correctly. But the code
is implemented in the hdfs-blobstore plugin. It will be problematic if we want
to add another plugin which also needs to connect to HDFS.
So the proposal here is to remove the login piece of code from hdfs-blobstore.
And explicitly login to hdfs once and only once when the server (nimbus,
supervisor, etc) or the client (storm cli command) launches. It can guarantee
one login per process.
The plugins like hdfs-blobstore then simply assume the process has already
logged in.
--
This message was sent by Atlassian Jira
(v8.3.2#803003)
