[
https://issues.apache.org/jira/browse/FLINK-10431?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16712632#comment-16712632
]
ASF GitHub Bot commented on FLINK-10431:
----------------------------------------
StefanRRichter commented on issue #6898: [FLINK-10431] Extraction of
scheduling-related code from SlotPool into preliminary Scheduler
URL: https://github.com/apache/flink/pull/6898#issuecomment-445192951
@Clarkkkkk background is that it will make things easier and otherwise you
have concurrency between two components that want to interact in transactional
ways: if the scheduler runs in a different thread than the slot pool there can
be concurent modifications to the slot pool (e.g. slots added/removed) between
the scheduler asking for the available slots and the scheduler requesting the
available slot. All of this has to be resolved and it becomes harder to
understand and reason about the code. This can be avoided if scheduler and slot
pool run in the same thread, and we are also aiming at having all modifications
to the execution graph in the same thread as well. The threading model would
then be that blocking or expensive operations run in their own thread so that
the main thread is never blocked, but the results are always synced back to a
main thread to runs all the modifications in scheduler, slot pool, execution
graph, etc.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
> Extract scheduling-related code from SlotPool
> ---------------------------------------------
>
> Key: FLINK-10431
> URL: https://issues.apache.org/jira/browse/FLINK-10431
> Project: Flink
> Issue Type: Sub-task
> Components: Distributed Coordination
> Affects Versions: 1.7.0
> Reporter: Stefan Richter
> Assignee: Stefan Richter
> Priority: Major
> Labels: pull-request-available
>
> The other half of the current scheduling logic is the management of slot
> sharing and is located in the SlotPool. We need to extract this logic into
> our new Scheduler component from the previous step. This leaves us with a
> simpler SlotPool that mainly cares about obtaining, holding, and releasing
> slots in interaction with a ResourceManager. The new Scheduler can now
> identify slot sharing groups and interacts with the SlotPool.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
