[
https://issues.apache.org/jira/browse/FLINK-23399?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Yanfei Lei updated FLINK-23399:
-------------------------------
Description:
We notice that rescaling is not covered in the current state benchmark, so we'd
like to introduce a benchmark for rescaling in flink-benchmark.
The benchmark process is:
(1) generate some states,
(2) change the parallelism of the operator, and restore from these states
generate before.
The implementation of this benchmark is based on
{{RocksIncrementalCheckpointRescalingTest}}, and *_AverageTime_* is used to
measure the rescaling performance on each subtask.
And this benchmark does not conflict with
`RocksIncrementalCheckpointRescalingBenchmarkTest` in
[PR|[https://github.com/apache/flink/pull/14893]]. Compare with
`RocksIncrementalCheckpointRescalingBenchmarkTest`, this benchmark supports
testing rescaling on different state backends, and has a finer granularity.
was:
We notice that rescaling is not covered in the current state benchmark, so we'd
like to introduce a benchmark for rescaling in flink-benchmark.
The benchmark process is:
(1) generate some states,
(2) change the parallelism of the operator, and restore from these states
generate before.
The implementation of this benchmark is based on
{{RocksIncrementalCheckpointRescalingTest}}, and *_AverageTime_* is used to
measure the rescaling performance on each subtask.
And this benchmark does not conflict with
RocksIncrementalCheckpointRescalingBenchmarkTest in
[PR|[https://github.com/apache/flink/pull/14893]]. Compare with
RocksIncrementalCheckpointRescalingBenchmarkTest, this benchmark supports
testing rescaling on different state backends, and has a finer granularity.
> Add a benchmark for rescaling
> -----------------------------
>
> Key: FLINK-23399
> URL: https://issues.apache.org/jira/browse/FLINK-23399
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / State Backends
> Affects Versions: 1.14.0
> Reporter: Yanfei Lei
> Priority: Minor
>
> We notice that rescaling is not covered in the current state benchmark, so
> we'd like to introduce a benchmark for rescaling in flink-benchmark.
> The benchmark process is:
> (1) generate some states,
> (2) change the parallelism of the operator, and restore from these states
> generate before.
> The implementation of this benchmark is based on
> {{RocksIncrementalCheckpointRescalingTest}}, and *_AverageTime_* is used to
> measure the rescaling performance on each subtask.
>
> And this benchmark does not conflict with
> `RocksIncrementalCheckpointRescalingBenchmarkTest` in
> [PR|[https://github.com/apache/flink/pull/14893]]. Compare with
> `RocksIncrementalCheckpointRescalingBenchmarkTest`, this benchmark supports
> testing rescaling on different state backends, and has a finer granularity.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
