You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Seth Wiesman (Jira)" <ji...@apache.org> on 2021/10/06 23:25:00 UTC
[jira] [Updated] (FLINK-24464) Metrics sometimes report negative
backlog
[ https://issues.apache.org/jira/browse/FLINK-24464?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Seth Wiesman updated FLINK-24464:
---------------------------------
Description:
As report [here|[https://stackoverflow.com/questions/69441309/negative-backlog-in-apache-flink-stateful-functions]] the backlog metric sometimes reports a negative value. This is because the current count is stored in an in-memory variable, while the value used to decrement the count as the backlog clears is stored in Flink state. In the case of a job restart, the in-memory variable is reset to zero (it is a new instance) while the numbers in state are retained.
A negative backlog is non-sensical, confusing, and should be prevented.
was:
As report [here| [https://stackoverflow.com/questions/69441309/negative-backlog-in-apache-flink-stateful-functions]] the backlog metric sometimes reports a negative value. This is because the current count is stored in an in-memory variable, while the value used to decrement the count as the backlog clears is stored in Flink state. In the case of a job restart, the in-memory variable is reset to zero (it is a new instance) while the numbers in state are retained.
A negative backlog is non-sensical, confusing, and should be prevented.
> Metrics sometimes report negative backlog
> -----------------------------------------
>
> Key: FLINK-24464
> URL: https://issues.apache.org/jira/browse/FLINK-24464
> Project: Flink
> Issue Type: Improvement
> Components: Stateful Functions
> Reporter: Seth Wiesman
> Priority: Major
>
> As report [here|[https://stackoverflow.com/questions/69441309/negative-backlog-in-apache-flink-stateful-functions]] the backlog metric sometimes reports a negative value. This is because the current count is stored in an in-memory variable, while the value used to decrement the count as the backlog clears is stored in Flink state. In the case of a job restart, the in-memory variable is reset to zero (it is a new instance) while the numbers in state are retained.
>
> A negative backlog is non-sensical, confusing, and should be prevented.
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)