[
https://issues.apache.org/jira/browse/FLINK-25908?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17491598#comment-17491598
]
jackwangcs commented on FLINK-25908:
------------------------------------
Hi [~gaborgsomogyi] , thanks for pointing out this issue. In fact, I also
noticed this before. In YarnLocalResourceDescriptor, it will pass the
`stagingDirFs`(constructed from yarnConfiguration) when it constructs the
YarnApplicationFileUploader see:
[here|https://github.com/apache/flink/blob/f1e17c9f39b876537ae2065c6331baa5302bbddf/flink-yarn/src/main/java/org/apache/flink/yarn/YarnClusterDescriptor.java#L804].
I think we need to ensure that we can obtain tokens for this Filesystem in
HadoopFSDelegationTokenProvider, it should work too in
[https://github.com/apache/flink/pull/14841.|https://github.com/apache/flink/pull/14841]
Utils.setTokenFor() will call TokenCache.obtainTokensForNamenodes(), and
TokenCache.obtainTokensForNamenodes() will also use the yarnConfiguration to
construct Fs for each provided paths under the
[hook|https://github.com/hanborq/hadoop/blob/99babec26a5f42021c0ec2a39ee7dbad4387b176/src/mapred/org/apache/hadoop/mapreduce/security/TokenCache.java#L91].
The behavior may change if it use ViewFS or DFS in hadoop?
I am not an expert in this part but I am glad to know more details.
> Move Hadoop FS token obtain functionality into HadoopFSDelegationTokenProvider
> ------------------------------------------------------------------------------
>
> Key: FLINK-25908
> URL: https://issues.apache.org/jira/browse/FLINK-25908
> Project: Flink
> Issue Type: Sub-task
> Affects Versions: 1.15.0
> Reporter: Gabor Somogyi
> Priority: Major
>
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
