[
https://issues.apache.org/jira/browse/FLINK-9756?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Nico Kruber closed FLINK-9756.
------------------------------
Resolution: Invalid
Fix Version/s: (was: 1.5.2)
(was: 1.6.0)
Actually, this only happens to the {{BufferListener#notifyBufferAvailable()}}
implementation of {{RemoteInputChannel}} which does not recycle the given
{{Buffer}} in case of errors. Let's solve this with FLINK-9755.
The implementation in {{PartitionRequestClientHandler.BufferListenerTask}}
already recycles the buffer and therefore gets back into
{{LocalBufferPool#recycle()}}.
> Exceptions in BufferListener#notifyBufferAvailable do not trigger further
> listeners in LocalBufferPool#recycle()
> ----------------------------------------------------------------------------------------------------------------
>
> Key: FLINK-9756
> URL: https://issues.apache.org/jira/browse/FLINK-9756
> Project: Flink
> Issue Type: Bug
> Components: Network
> Affects Versions: 1.5.0
> Reporter: Nico Kruber
> Assignee: Nico Kruber
> Priority: Major
>
> Any {{Exception}} thrown in {{BufferListener#notifyBufferAvailable}} will
> currently not trigger calling further listeners in
> {{LocalBufferPool#recycle()}} and only add the given memory segment to the
> queue of available ones.
> Usually this will not be the last call to {{recycle()}} and future calls may
> call the listeners but this also introduces further delay in configurations
> with tight numbers of buffers in the local pool and listeners waiting on them
> when one task's listener was failing.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
