[
https://issues.apache.org/jira/browse/KYLIN-5308?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jiale He updated KYLIN-5308:
----------------------------
Description:
kylin fails to write data to the new transaction because the old transaction is
not interrupted during the Renew Epoch Retry. As a result, Kylin loses the
epoch.
When Epoch Renew occurs, the write to the database times out and tries again.
As the old transaction is not rolled back during the retry, the old transaction
is successfully written to the database and MVCC+1 occurs. When a new
transaction writes to the database, it will check the MVCC. If the detection
fails, kylin will think that the Epoch is lost.
!image-2022-11-28-16-32-00-356.png!
See attachment for RC and Fix Design
was:
kylin fails to write data to the new transaction because the old transaction is
not interrupted during the Renew Epoch Retry. As a result, Kylin loses the
epoch.
When Epoch Renew occurs, the write to the database times out and tries again.
As the old transaction is not rolled back during the retry, the old transaction
is successfully written to the database and MVCC+1 occurs. When a new
transaction writes to the database, it will check the MVCC. If the detection
fails, kylin will think that the Epoch is lost.
!image-2022-11-28-16-32-00-356.png!
RC and Fix Design
> The old transaction is not interrupted when Epoch renew retry
> -------------------------------------------------------------
>
> Key: KYLIN-5308
> URL: https://issues.apache.org/jira/browse/KYLIN-5308
> Project: Kylin
> Issue Type: Bug
> Reporter: Jiale He
> Assignee: Jiale He
> Priority: Critical
> Attachments: KYLIN-5308 Epoch Renew 事务超时 Design.pdf,
> image-2022-11-28-16-32-00-356.png, image-2022-11-29-14-29-31-681.png
>
>
> kylin fails to write data to the new transaction because the old transaction
> is not interrupted during the Renew Epoch Retry. As a result, Kylin loses the
> epoch.
>
> When Epoch Renew occurs, the write to the database times out and tries again.
> As the old transaction is not rolled back during the retry, the old
> transaction is successfully written to the database and MVCC+1 occurs. When a
> new transaction writes to the database, it will check the MVCC. If the
> detection fails, kylin will think that the Epoch is lost.
> !image-2022-11-28-16-32-00-356.png!
>
> See attachment for RC and Fix Design
>
>
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
