[
https://issues.apache.org/jira/browse/FLINK-21731?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Zhilong Hong updated FLINK-21731:
---------------------------------
Description:
We notice that {{DefaultScheduler#allocateSlotsAndDeploy}} is not covered in
the current scheduler benchmark. When we are trying to implement a benchmark
related to this procedure, we think that it's better to implement a benchmark
that covers the entire procedures of DefaultScheduler's creation, scheduling
and deploying instead.
This can avoid missing any parts that may affect the performance of scheduling.
Also in this way we don't need to add benchmarks for every sub-procedure
related to scheduling, which makes the benchmark heavy and hard to maintain.
Then we can just focus on the performance-sensitive procedures, as the existing
benchmarks do.
The new benchmark item is implemented based on
{{DefaultSchedulerBatchSchedulingTest}}.
was:
We notice that {{DefaultScheduler#allocateSlotsAndDeploy}} is not covered in
the current scheduler benchmark. When we are trying to implement a benchmark
related to this procedure, we think that it's better to implement a benchmark
that covers the entire {{DefaultScheduler#startScheduling}} procedure instead.
This can avoid missing any parts that may affect the performance of scheduling.
Also in this way we don't need to add benchmarks for every sub-procedure
related to scheduling, which makes the benchmark heavy and hard to maintain.
Then we can just focus on the performance-sensitive procedures, as the existing
benchmarks do.
The new benchmark item is implemented based on
{{DefaultSchedulerBatchSchedulingTest}}.
> Add benchmarks for DefaultScheduler's creation, scheduling and deploying
> ------------------------------------------------------------------------
>
> Key: FLINK-21731
> URL: https://issues.apache.org/jira/browse/FLINK-21731
> Project: Flink
> Issue Type: Improvement
> Components: Benchmarks, Runtime / Coordination
> Affects Versions: 1.13.0
> Reporter: Zhilong Hong
> Priority: Major
> Labels: pull-request-available
> Fix For: 1.13.0
>
>
> We notice that {{DefaultScheduler#allocateSlotsAndDeploy}} is not covered in
> the current scheduler benchmark. When we are trying to implement a benchmark
> related to this procedure, we think that it's better to implement a benchmark
> that covers the entire procedures of DefaultScheduler's creation, scheduling
> and deploying instead.
> This can avoid missing any parts that may affect the performance of
> scheduling. Also in this way we don't need to add benchmarks for every
> sub-procedure related to scheduling, which makes the benchmark heavy and hard
> to maintain. Then we can just focus on the performance-sensitive procedures,
> as the existing benchmarks do.
> The new benchmark item is implemented based on
> {{DefaultSchedulerBatchSchedulingTest}}.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
