[
https://issues.apache.org/jira/browse/FLINK-30670?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Lijie Wang updated FLINK-30670:
-------------------------------
Description: Currently, we include the broadcast bytes in the
"jobmanager.adaptive-batch-scheduler.avg-data-volume-per-task" when calculating
the parallelism (see [PR17952|https://github.com/apache/flink/pull/17952] for
details), and set a cap ratio(0.5) for the broadcast ratio. Considering that
the broadcast bytes are generally relatively small, we can ignore the broadcast
bytes to simplify the logic. (was: Currently, we include the broadcast bytes
in the "jobmanager.adaptive-batch-scheduler.avg-data-volume-per-task" when
calculating the parallelism, and set a cap ratio(0.5) for the broadcast ratio.
Considering that the broadcast bytes are generally relatively small, we can
ignore the broadcast bytes to simplify the logic.)
> Ignore broadcast bytes when computing parallelism and input infos
> -----------------------------------------------------------------
>
> Key: FLINK-30670
> URL: https://issues.apache.org/jira/browse/FLINK-30670
> Project: Flink
> Issue Type: Sub-task
> Components: Runtime / Coordination
> Reporter: Lijie Wang
> Assignee: Lijie Wang
> Priority: Major
> Fix For: 1.17.0
>
>
> Currently, we include the broadcast bytes in the
> "jobmanager.adaptive-batch-scheduler.avg-data-volume-per-task" when
> calculating the parallelism (see
> [PR17952|https://github.com/apache/flink/pull/17952] for details), and set a
> cap ratio(0.5) for the broadcast ratio. Considering that the broadcast bytes
> are generally relatively small, we can ignore the broadcast bytes to simplify
> the logic.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
