[ 
https://issues.apache.org/jira/browse/FLINK-9076?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442750#comment-16442750
 ] 

ASF GitHub Bot commented on FLINK-9076:
---------------------------------------

GitHub user NicoK opened a pull request:

    https://github.com/apache/flink/pull/5874

    [FLINK-9076][network] reduce minimum number of floating buffers to 0

    ## What is the purpose of the change
    
    Currently, floating buffers (per gate) are always required in case 
credit-based flow control is enabled. This, however, increases our minimum 
number of required network buffers.
    
    Instead, we should also work with a minimum of zero floating buffers and 
set the max to the configured value. This way, if there are not enough buffers, 
all `LocalBufferPool`s will at least share the available ones and not fail the 
job.
    
    ## Brief change log
    
    - make credit-based input gate buffer pools flexible (0 to 
`extraNetworkBuffersPerGate` buffers, depending on availability) - this is 
similar to the non-credit based code path
    - clarify the documentation of 
`taskmanager.network.memory.buffers-per-channel` in this regard
    
    ## Verifying this change
    
    This change added tests and can be verified as follows:
    
    - added unit tests inside `NetworkEnvironmentTest` to cover minimum and 
insufficient buffers scenarios for both credit-based and non-credit based
    - add a network end-to-end test to `StreamNetworkThroughputBenchmarkTest` 
covering minimum and insufficient buffers scenarios for default, i.e. 
credit-based, flow control
    
    ## Does this pull request potentially affect one of the following parts:
    
      - Dependencies (does it add or upgrade a dependency): **no**
      - The public API, i.e., is any changed class annotated with 
`@Public(Evolving)`: **no**
      - The serializers: **no**
      - The runtime per-record code paths (performance sensitive): **no** (but 
per-buffer - and too few buffers may degrade performance)
      - Anything that affects deployment or recovery: JobManager (and its 
components), Checkpointing, Yarn/Mesos, ZooKeeper: **no**
      - The S3 file system connector: **no**
    
    ## Documentation
    
      - Does this pull request introduce a new feature? **no**
      - If yes, how is the feature documented? **docs**


You can merge this pull request into a Git repository by running:

    $ git pull https://github.com/NicoK/flink flink-9076

Alternatively you can review and apply these changes as the patch at:

    https://github.com/apache/flink/pull/5874.patch

To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:

    This closes #5874
    
----
commit 77896ab92114ecce6cf4549700dccb158e32d9f1
Author: Nico Kruber <nico@...>
Date:   2018-04-18T15:04:38Z

    [FLINK-9208][tests] fix naming of StreamNetworkThroughputBenchmarkTest

commit c359f70db93508c25e1ceb880f858635f9b4d705
Author: Nico Kruber <nico@...>
Date:   2018-04-06T09:15:10Z

    [FLINK-9076][network] reduce minimum number of floating buffers to 0

commit 60792fc1192260a6c443906f3ea6a175c8900787
Author: Nico Kruber <nico@...>
Date:   2018-04-18T15:44:22Z

    [FLINK-9076][network] add network end-to-end test with minimum buffers

----


> Make credit-based floating buffers optional
> -------------------------------------------
>
>                 Key: FLINK-9076
>                 URL: https://issues.apache.org/jira/browse/FLINK-9076
>             Project: Flink
>          Issue Type: Sub-task
>          Components: Network
>    Affects Versions: 1.5.0, 1.6.0
>            Reporter: Nico Kruber
>            Assignee: Nico Kruber
>            Priority: Blocker
>             Fix For: 1.5.0
>
>
> Currently, floating buffers (per gate) are always required in case 
> credit-based flow control is enabled. This, however, increases our minimum 
> number of required network buffers.
> Instead, without changing too much, we could already work with a minimum of 
> one or zero floating buffers and set the max to the configured value. This 
> way, if there are not enough buffers, allĀ {{LocalBufferPool}}s will at least 
> share the available ones.



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

Reply via email to