You are viewing a plain text version of this content. The canonical link for it is here.

Posted to user@flink.apache.org by aitozi <gj...@gmail.com> on 2019/06/16 14:19:20 UTC

A little doubt about the blog A Deep To Flink's NetworkStack

Hi, community

I read this blog  A Deep To Flink's NetworkStack
<https://flink.apache.org/2019/06/05/flink-network-stack.html>  

In the chapter *Physical Transport*, it says that the /if different subtasks
of the same task are scheduled onto the same TaskManager, their network
connections towards the same TaskManagers will be multiplexed and share a
single TCP channel for reduced resource usage./.

But I check the code I think each taskmanager have a NettyServer and a
NettyClient which work with a channel. So i think the tcp channel is shared
between the taskmanager remote connection no matter there is different
subtasks of the same task are scheduled onto the same TaskManager or not.

Please correct me if my thoughts was wrong.

Thanks
Aitozi



--
Sent from: http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/

Re: A little doubt about the blog A Deep To Flink's NetworkStack

Posted by aitozi <gj...@gmail.com>.

Hi Zhijiang,

Thanks for your explanation, I have understood it. My former thoughts was
wrong, the real TCP channel was established when
createPartitionRequestClient. 

Thanks,
Aitozi



--
Sent from: http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/

Re: A little doubt about the blog A Deep To Flink's NetworkStack

Posted by zhijiang <wa...@aliyun.com>.

Hi Aitozi,

The current connection sharing is only suitbale for the same jobVertex between TaskMangaers via ConnectionIndex. I ever tried to remove this limit by ticket [1] but have not reached an agreement on this.
You could watch this jira if interested.


[1] https://issues.apache.org/jira/browse/FLINK-10462

Best,
Zhijiang
------------------------------------------------------------------
From:aitozi <gj...@gmail.com>
Send Time:2019年6月16日(星期日) 22:19
To:user <us...@flink.apache.org>
Subject:A little doubt about the blog A Deep To Flink's NetworkStack

Hi, community

I read this blog  A Deep To Flink's NetworkStack
<https://flink.apache.org/2019/06/05/flink-network-stack.html>  

In the chapter *Physical Transport*, it says that the /if different subtasks
of the same task are scheduled onto the same TaskManager, their network
connections towards the same TaskManagers will be multiplexed and share a
single TCP channel for reduced resource usage./.

But I check the code I think each taskmanager have a NettyServer and a
NettyClient which work with a channel. So i think the tcp channel is shared
between the taskmanager remote connection no matter there is different
subtasks of the same task are scheduled onto the same TaskManager or not.

Please correct me if my thoughts was wrong.

Thanks
Aitozi



--
Sent from: http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/