[jira] [Commented] (FLINK-10367) Avoid recursion stack overflow during releasing SingleInputGate

["Nico Kruber (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/FLINK-10367?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16627115#comment-16627115 ] 

Nico Kruber commented on FLINK-10367:
-------------------------------------

After taking a brief look at the code, I'd also prefer option 2. However, it may lead to spurious exceptions during release (I'm not 100% sure here, please check): part of releasing the channel is to close the partition request client or remove it from the connection manager. I guess, until this is done, other parts of the network _could_ (really?) receive further data and rely on the buffer pool being available.
In case of a graceful shutdown after completing a task, this shouldn't happen, but in case of an error and the following channel-release, we don't want the error from the release to suppress the original one.

If that really is a possibility, there is also this option: (I'm not saying this is the one to go with)

3. to let {{notifyBufferAvailable}} also return whether it has actually taken over the buffer and not recycle inside these implementations but instead trying to re-distribute the {{Buffer}} to other listeners in {{LocalBufferPool#recycle}}

> Avoid recursion stack overflow during releasing SingleInputGate
> ---------------------------------------------------------------
>
>                 Key: FLINK-10367
>                 URL: https://issues.apache.org/jira/browse/FLINK-10367
>             Project: Flink
>          Issue Type: Improvement
>          Components: Network
>    Affects Versions: 1.5.0, 1.5.1, 1.5.2, 1.5.3, 1.6.0
>            Reporter: zhijiang
>            Assignee: zhijiang
>            Priority: Minor
>
> For task failure or canceling, the {{SingleInputGate#releaseAllResources}} will be invoked before task exits.
> In the process of {{SingleInputGate#releaseAllResources}}, we first loop to release all the input channels, then destroy the {{BufferPool}}.  For {{RemoteInputChannel#releaseAllResources}}, it will return floating buffers to the {{BufferPool}} {{which assigns this recycled buffer to the other listeners(RemoteInputChannel}}). 
> It may exist recursive call in this process. If the listener is already released before, it will directly recycle this buffer to the {{BufferPool}} which takes another listener to notify available buffer. The above process may be invoked repeatedly in recursive way.
> If there are many input channels as listeners in the {{BufferPool}}, it will cause {{StackOverflow}} error because of recursion. And in our testing job, the scale of 10,000 input channels ever caused this error.
> I think of two ways for solving this potential problem:
>  # When the input channel is released, it should notify the {{BufferPool}} of unregistering this listener, otherwise it is inconsistent between them.
>  # {{SingleInputGate}} should destroy the {{BufferPool}} first, then loop to release all the internal input channels. To do so, all the listeners in {{BufferPool}} will be removed during destroying, and the input channel will not have further interactions during {{RemoteInputChannel#releaseAllResources}}.
> I prefer the second way to solve this problem, because we do not want to expand another interface method for removing buffer listener, further currently the internal data structure in {{BufferPool}} can not support remove a listener directly.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)