[jira] [Commented] (FLINK-14872) Potential deadlock for task reading from blocking ResultPartition.

["Yingjie Cao (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/FLINK-14872?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16979037#comment-16979037 ] 

Yingjie Cao commented on FLINK-14872:
-------------------------------------

Only limit the number of Buffers can be allocated can only reduce the possibility of deadlock and gives users a chance to increase the number of network buffers by configuration. To totally solve the problem, we may also need to request core buffers directly from global NetworkBufferPool like the exclusive buffers for InputGate, we may add a method to local buffer pool like reserveRequiredBuffers and call it at result partition setup. Currently, core buffer request of the result partition is lazy, the buffer is not allocated from global pool until emitting records. Like the exclusive buffer request for InputGate, we can add a request time out and let the users know that the number of network buffer should be increased.

Another possible option is letting the InputGate release Buffer by spilling to disk, though the implementation is more complicated.

Letting the ResultPartition reserve core buffer at setup may lead to waste of Buffer, but I think that is not a big problem, most of the task will emit records sooner or latter.

> Potential deadlock for task reading from blocking ResultPartition.
> ------------------------------------------------------------------
>
>                 Key: FLINK-14872
>                 URL: https://issues.apache.org/jira/browse/FLINK-14872
>             Project: Flink
>          Issue Type: Bug
>            Reporter: Yingjie Cao
>            Priority: Major
>             Fix For: 1.10.0
>
>
> Currently, the buffer pool size of InputGate reading from blocking ResultPartition is unbounded which have a potential of using too many buffers and may lead to ResultPartition of the same task can not acquire enough core buffers and finally lead to deadlock.
> Considers the following case:
> Core buffers are reserved for InputGate and ResultPartition -> InputGate consumes lots of Buffer (not including the buffer reserved for ResultPartition) -> Other tasks acquire exclusive buffer for InputGate and trigger redistribute of Buffers (Buffers taken by previous InputGate can not be released) -> The first task of which InputGate uses lots of buffers begin to emit records but can not acquire enough core Buffers (Some operators may not emit records out immediately or there is just nothing to emit) -> Deadlock.
>  
> I think we can fix this problem by limit the number of Buffers can be allocated by a InputGate which reads from blocking ResultPartition.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)