You are viewing a plain text version of this content. The canonical link for it is here.
Posted to jira@kafka.apache.org by "Sophie Blee-Goldman (Jira)" <ji...@apache.org> on 2020/02/04 19:53:00 UTC
[jira] [Updated] (KAFKA-9501) Promote Standby tasks to active tasks
without closing them
[ https://issues.apache.org/jira/browse/KAFKA-9501?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Sophie Blee-Goldman updated KAFKA-9501:
---------------------------------------
Description:
The purpose of StandbyTasks in Kafka Streams is fast failover. However, atm we close StandbyTasks (and create a new active task) if they are promoted to active.
While this works ok for persistent state stores, it renders hot standbys for in-memory state stores useless, because we drop the in-memory state when we close the StandbyTask and thus the new active tasks needs to reread the changelog topic to recreate the in-memory state.
Hence, we should promote StandbyTasks to active tasks without closing them. This will not only fix the issue for in-memory stores, but will make rebalancing faster for persistent state stores, too, because closing and reopening RocksDB has significant overhead.
was:
The purpose of StandbyTasks in Kafka Streams is fast failover. However, atm we close StandbyTasks (and create a new active task) if they are promoted to active.
While this works ok for persistent state stores, it renders hot standbys for in-memory state stores useless, because we drop the in-memory state when we close the StandbyTask and thus the new active tasks needs to reread the changelog topic to recreate the in-memory state.
Hence, we should promote StandbyTasks to active tasks without closing them. This will not fix the issue for in-memory stores, but will make rebalancing faster for persistent state stores, too, because closing and reopening RocksDB has significant overhead.
> Promote Standby tasks to active tasks without closing them
> ----------------------------------------------------------
>
> Key: KAFKA-9501
> URL: https://issues.apache.org/jira/browse/KAFKA-9501
> Project: Kafka
> Issue Type: Bug
> Components: streams
> Reporter: Matthias J. Sax
> Priority: Critical
>
> The purpose of StandbyTasks in Kafka Streams is fast failover. However, atm we close StandbyTasks (and create a new active task) if they are promoted to active.
> While this works ok for persistent state stores, it renders hot standbys for in-memory state stores useless, because we drop the in-memory state when we close the StandbyTask and thus the new active tasks needs to reread the changelog topic to recreate the in-memory state.
> Hence, we should promote StandbyTasks to active tasks without closing them. This will not only fix the issue for in-memory stores, but will make rebalancing faster for persistent state stores, too, because closing and reopening RocksDB has significant overhead.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)