[
https://issues.apache.org/jira/browse/BEAM-5713?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16647630#comment-16647630
]
Maximilian Michels commented on BEAM-5713:
------------------------------------------
{quote}
I added more tasks and they are all squeezed into the same slots (only 8 out of
144 task slots are used).
{quote}
Tasks which directly depend on each other share the same task slot. That is how
pipelining in Flink works. AFAIK pipelines can get arbitrarily long.
{quote}
The scheduling of all tasks to the same slot is consistent, distribution over
hosts isn't. With parallelism 4, different result (multiple hosts).
{quote}
Is that consistent behavior for parallelism of 4? I find that it depends on
what the iterator returns from the task slot HashMap. This depends on a number
of factors, e.g. what jobs you ran before, how TaskManager registered.
> Flink portable runner schedules all tasks of streaming job on same task
> manager
> -------------------------------------------------------------------------------
>
> Key: BEAM-5713
> URL: https://issues.apache.org/jira/browse/BEAM-5713
> Project: Beam
> Issue Type: Bug
> Components: runner-flink
> Affects Versions: 2.8.0
> Reporter: Thomas Weise
> Assignee: Maximilian Michels
> Priority: Major
> Labels: portability, portability-flink
> Attachments: Different SlotSharingGroup.png, With
> RichParallelSourceFunction and parallelism 5.png,
> image-2018-10-11-11-43-50-333.png, image-2018-10-11-16-20-45-221.png
>
> Time Spent: 0.5h
> Remaining Estimate: 0h
>
> The cluster has 9 task managers and 144 task slots total. A simple streaming
> pipeline with parallelism of 8 will get all tasks scheduled on the same task
> manager, causing the host to be fully booked and the remaining cluster idle.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
