[
https://issues.apache.org/jira/browse/FLINK-15637?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Yu Li updated FLINK-15637:
--------------------------
Release Note: After FLINK-15637, we change the default timer store from
heap to RocksDB for RocksDBStateBackend to support asynchronous snapshots for
timer state and better scalability, with less than 5% performance cost. If you
find the performance decline critical, please set
"state.backend.rocksdb.timer-service.factory" to "HEAP" in flink-conf.yaml to
restore the old behavior.
Update release note accordingly.
> For RocksDBStateBackend, make RocksDB the default store for timers
> ------------------------------------------------------------------
>
> Key: FLINK-15637
> URL: https://issues.apache.org/jira/browse/FLINK-15637
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / State Backends
> Reporter: Stephan Ewen
> Assignee: Congxian Qiu(klion26)
> Priority: Blocker
> Labels: pull-request-available
> Fix For: 1.10.0
>
> Time Spent: 20m
> Remaining Estimate: 0h
>
> Set the {{state.backend.rocksdb.timer-service.factory}} to {{ROCKSDB}} by
> default. Also ensure that the programmatic default value becomes the same.
> We need to update the performance tuning guide to mention this.
>
> We need to update the release notes to mention this.
> This is a decision concluded from [ML
> discussion|http://apache-flink-mailing-list-archive.1008284.n3.nabble.com/DISCUSS-Change-default-for-RocksDB-timers-Java-Heap-gt-in-RocksDB-td36720.html]
> (as well as a [twitter
> vote|https://twitter.com/StephanEwen/status/1214594273565388801]).
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
