[
https://issues.apache.org/jira/browse/HIVE-27797?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Taraka Rama Rao Lethavadla updated HIVE-27797:
----------------------------------------------
Description:
+Scenario:+
Let's there are 100 transactions opened. These 100 will be logged in
notification_log and when replicated, they will get created in target cluster.
Now 50 out of these 100 transactions got aborted due to timeout and got removed
from HMS. In this step, we are not logging those transactions in to
notification_log.
So next time when we do replication, these 50 aborted transactions will not be
replicated.
As a result in the target cluster the transactions that got created earlier
will only get removed after number of days configured in config
{code:java}
hive.repl.txn.timeout (11 days default){code}
Actually, we have the logic to log aborted transactions if they got aborted for
some other reason but not for those that are getting timed out.
was:
+Scenario:+
Let's there are 100 transactions opened. These 100 will be logged in
notification_log and when replicated, they will get created in target cluster.
Now 50 out of these 100 transactions got aborted due to timeout and got removed
from HMS. In this step, we are not logging those transactions in to
notification_log.
So next time when we do replication, these 50 aborted transactions will not be
replicated.
As a result in the target cluster the transactions that got created earlier
will stay forever without getting cleaned
Actually, we have the logic to log aborted transactions if they got aborted for
some other reason but not for those getting timed out.
> Transactions that got timed out are not getting logged as ABORTED
> transactions in NOTIFICATION_LOG
> --------------------------------------------------------------------------------------------------
>
> Key: HIVE-27797
> URL: https://issues.apache.org/jira/browse/HIVE-27797
> Project: Hive
> Issue Type: Bug
> Components: repl, Transactions
> Reporter: Taraka Rama Rao Lethavadla
> Assignee: Taraka Rama Rao Lethavadla
> Priority: Major
>
> +Scenario:+
> Let's there are 100 transactions opened. These 100 will be logged in
> notification_log and when replicated, they will get created in target
> cluster.
> Now 50 out of these 100 transactions got aborted due to timeout and got
> removed from HMS. In this step, we are not logging those transactions in to
> notification_log.
> So next time when we do replication, these 50 aborted transactions will not
> be replicated.
> As a result in the target cluster the transactions that got created earlier
> will only get removed after number of days configured in config
> {code:java}
> hive.repl.txn.timeout (11 days default){code}
> Actually, we have the logic to log aborted transactions if they got aborted
> for some other reason but not for those that are getting timed out.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
