[
https://issues.apache.org/jira/browse/FLINK-36556?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Piotr Nowojski updated FLINK-36556:
-----------------------------------
Release Note: When using buffer debloating, during job's startup or
recovery, when we don't know what's the actual records throughput, initial
buffer size is now by default 1024B. Previously it was 32 KB, which could lead
to jobs choking with large buffers during the startup that took hours to fully
process. The starting buffer size can be changed using
`taskmanager.memory.starting-segment-size` (was: When using buffer debloating,
during job's startup intiall
)
> Allow to configure starting buffer size when using buffer debloating
> --------------------------------------------------------------------
>
> Key: FLINK-36556
> URL: https://issues.apache.org/jira/browse/FLINK-36556
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Network, Runtime / Task
> Affects Versions: 1.19.1
> Reporter: Piotr Nowojski
> Priority: Major
> Labels: pull-request-available
> Fix For: 2.0.0
>
>
> Starting buffer size is 32KB, so during recovery/startup, before backpressure
> kicks in, Flink job can be flooded with large buffers, completely stalling
> the progress.
> Proposed solution is to make this starting size configurable.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
