[
https://issues.apache.org/jira/browse/FLINK-4545?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15923554#comment-15923554
]
ASF GitHub Bot commented on FLINK-4545:
---------------------------------------
Github user zhijiangW commented on the issue:
https://github.com/apache/flink/pull/3467
@NicoK ，thank you for explanation, and I already trace the code in your
local branch. Wish your further change commit in global pool.
@StephanEwen , thanks for further elaboration. From my understanding, each
task can decide the core number of buffers in `LocalBufferPool` based on input,
output channels and configuration, the maximum number of buffers based on
`ResultPartitionType`. And all the `LocalBufferPool`s make effect on the total
number of buffers in `NetworkBufferPool`, may need consider maximum memory
usages.
And my concern is to consider the memory usages in `NetworkBufferPool`
before starts the `TaskManager`, and this part of memory should be added into
the total resource of `TaskManager`.
I am willing to do that as a part of my current work in [Fine-grained
Resource Configuration](https://issues.apache.org/jira/browse/FLINK-5131) after
this feature completes.
> Flink automatically manages TM network buffer
> ---------------------------------------------
>
> Key: FLINK-4545
> URL: https://issues.apache.org/jira/browse/FLINK-4545
> Project: Flink
> Issue Type: Wish
> Components: Network
> Reporter: Zhenzhong Xu
>
> Currently, the number of network buffer per task manager is preconfigured and
> the memory is pre-allocated through taskmanager.network.numberOfBuffers
> config. In a Job DAG with shuffle phase, this number can go up very high
> depends on the TM cluster size. The formula for calculating the buffer count
> is documented here
> (https://ci.apache.org/projects/flink/flink-docs-master/setup/config.html#configuring-the-network-buffers).
>
> #slots-per-TM^2 * #TMs * 4
> In a standalone deployment, we may need to control the task manager cluster
> size dynamically and then leverage the up-coming Flink feature to support
> scaling job parallelism/rescaling at runtime.
> If the buffer count config is static at runtime and cannot be changed without
> restarting task manager process, this may add latency and complexity for
> scaling process. I am wondering if there is already any discussion around
> whether the network buffer should be automatically managed by Flink or at
> least expose some API to allow it to be reconfigured. Let me know if there is
> any existing JIRA that I should follow.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
