[
https://issues.apache.org/jira/browse/HDFS-17476?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17846316#comment-17846316
]
ASF GitHub Bot commented on HDFS-17476:
---------------------------------------
KeeProMise commented on PR #6747:
URL: https://github.com/apache/hadoop/pull/6747#issuecomment-2110243175
> Thanks involving me.
>
> @KeeProMise I'm just doubt what scenarios could cause a negative
`clientStateId`?
@ZanderXu thanks for your review, under normal circumstances, this problem
does not occur; because when I wrote a single test, I forgot to pass the
transaction id, which caused this problem;
I think the code logic here itself lacks consideration of long overflow (in
our environment, if the transaction id is not passed, the default is
long.minvalue), which may cause problems in future iterations.
> fix: False positive "Observer Node is too far behind" due to long overflow.
> ---------------------------------------------------------------------------
>
> Key: HDFS-17476
> URL: https://issues.apache.org/jira/browse/HDFS-17476
> Project: Hadoop HDFS
> Issue Type: Bug
> Reporter: Jian Zhang
> Assignee: Jian Zhang
> Priority: Critical
> Labels: pull-request-available
> Attachments: HDFS-17476.patch, image-2024-04-18-10-57-10-481.png
>
>
> In the code GlobalStateIdContext#receiveRequestState(), if clientStateId is a
> small negative number, clientStateId-serverStateId may be greater than
> （ESTIMATED_TRANSACTIONS_PER_SECOND due to overflow
> * TimeUnit.MILLISECONDS.toSeconds(clientWaitTime)
> * ESTIMATED_SERVER_TIME_MULTIPLIER）,
> resulting in false positives that Observer Node is too far behind.
> !image-2024-04-18-10-57-10-481.png|width=742,height=110!
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
