[
https://issues.apache.org/jira/browse/FLINK-19439?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17210645#comment-17210645
]
Dawid Wysakowicz commented on FLINK-19439:
------------------------------------------
I am confident the regression happens only for the first record, because of
slightly heavier initialization of Avro readers. I will close this ticket as
not a problem.
I also executed an experiment of increasing the number of incoming records in
benchmarks, to possibly minimize the impact of the initialization on the
throughput. Here are the results ([^jmh-result.csv]) . I am not entirely sure
though if we want to commit that to flink-benchmarks. Take notice that pretty
much all serializers report higher throughput than the baseline with more
incoming records. My concern is it will make our performance history useless.
WDYT [~AHeise] [~pnowojski]
> Performance regression on 24.09.2020
> ------------------------------------
>
> Key: FLINK-19439
> URL: https://issues.apache.org/jira/browse/FLINK-19439
> Project: Flink
> Issue Type: Bug
> Components: API / Type Serialization System, Benchmarks
> Affects Versions: 1.12.0
> Reporter: Piotr Nowojski
> Assignee: Dawid Wysakowicz
> Priority: Blocker
> Attachments: jmh-result.csv
>
>
> http://codespeed.dak8s.net:8000/timeline/?ben=serializerAvro&env=2
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
