[
https://issues.apache.org/jira/browse/HDFS-6666?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14486180#comment-14486180
]
Chris Nauroth commented on HDFS-6666:
-------------------------------------
Thank you for the patch, [~vijaysbhat].
It's not necessary to call {{LOG.error}} before throwing the exceptions. The
exceptions will propagate out and get logged at a higher layer before the
process terminates. The log statements here would cause the error to get
logged twice.
On the NameNode side, I recommend changing the existing code in
{{BlockManager#createBlockTokenSecretManager}} instead of adding new code to
the {{NameNode}} class. The existing code currently logs an error. We'd just
need to change it to throw an exception.
In the test, I recommend using JUnit's {{ExpectedException}} with the expected
message set to something like "Security is enabled but block access tokens".
For an example of this, see
{{TestSaslDataTransfer#testDataNodeAbortsIfNoSasl}}. This will help us
guarantee that the exception is thrown from the right place and for the right
reason.
> Abort NameNode and DataNode startup if security is enabled but block access
> token is not enabled.
> -------------------------------------------------------------------------------------------------
>
> Key: HDFS-6666
> URL: https://issues.apache.org/jira/browse/HDFS-6666
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: datanode, namenode, security
> Affects Versions: 3.0.0, 2.5.0
> Reporter: Chris Nauroth
> Assignee: Vijay Bhat
> Priority: Minor
> Attachments: HDFS-6666.001.patch
>
>
> Currently, if security is enabled by setting hadoop.security.authentication
> to kerberos, but HDFS block access tokens are disabled by setting
> dfs.block.access.token.enable to false (which is the default), then the
> NameNode logs an error and proceeds, and the DataNode proceeds without even
> logging an error. This jira proposes that this it's invalid to turn on
> security but not turn on block access tokens, and that it would be better to
> fail fast and abort the daemons during startup if this happens.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
