[
https://issues.apache.org/jira/browse/FLINK-24190?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Anton Kalashnikov updated FLINK-24190:
--------------------------------------
Description:
If the buffer size will be too small(less than record size) due to buffer
debloat it can lead to performance degradation. It looks like it is better to
keep the buffer size greater than the record size(or even greater than it).
Implementation should be easy, we can choose the maximum value between
desirableBufferSize and recordSize during requesting the new
buffer(BufferWritingResultPartition#addToSubpartition)
was:
If the buffer size will be too small(less than record size) due to buffer
debloat it can lead to performance degradation. It looks like it is better to
keep the buffer size greater than the record size(or even greater than it). So
it needs to check how bad it can be and fix it.
Implementation should be easy, we can choose the maximum value between
desirableBufferSize and recordSize during requesting the new
buffer(BufferWritingResultPartition#addToSubpartition)
> Handling large record with buffer debloat
> -----------------------------------------
>
> Key: FLINK-24190
> URL: https://issues.apache.org/jira/browse/FLINK-24190
> Project: Flink
> Issue Type: Sub-task
> Components: Runtime / Network
> Affects Versions: 1.14.0
> Reporter: Anton Kalashnikov
> Priority: Major
> Fix For: 1.15.0
>
>
> If the buffer size will be too small(less than record size) due to buffer
> debloat it can lead to performance degradation. It looks like it is better to
> keep the buffer size greater than the record size(or even greater than it).
> Implementation should be easy, we can choose the maximum value between
> desirableBufferSize and recordSize during requesting the new
> buffer(BufferWritingResultPartition#addToSubpartition)
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
