[
https://issues.apache.org/jira/browse/YARN-574?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15644745#comment-15644745
]
Naganarasimha G R commented on YARN-574:
----------------------------------------
Thanks [~ajithshetty], for the patch.
As discussed offline, I think existing approach of sending only one resource as
HB response is better as there can be some control at NM end to determine (if
req in future) whether to give multiple resources to be localized or one at a
time(or even none based on the NM's load). Further to it i would to add further
points
# Agree to [~varun_saxena]'s point having fixed 4 threads (core and max pool
size ) is not ideal. But IMHO i would like to keep atleast 2(default max pool
size), as users might not be able make use of the benifit if not.
# I think instead of using {{Executors.newFixedThreadPool(nThreads, tf)}} we
can try to use {{ThreadPoolExecutor}} so that we can we can make use of
{{ThreadPoolExecutor.getActiveCount()}} and compare it with max pool size to
determine whether to request Heartbeat immediately or to wait 1000s.
# If NM sends *LIVE* and if no ResourceLocalizationSpecs are shared across then
no need to check the current load on the executor, we can wait for the defined
poll period and then do the HB.
# TestContainerLocalizer.java, ln no 301-306, Please add some proper message
which can be shown on failure.
# {{TestYarnConfigurationFields.testCompareConfigurationClassAgainstXml}} is
related to the patch
> PrivateLocalizer does not support parallel resource download via
> ContainerLocalizer
> -----------------------------------------------------------------------------------
>
> Key: YARN-574
> URL: https://issues.apache.org/jira/browse/YARN-574
> Project: Hadoop YARN
> Issue Type: Sub-task
> Affects Versions: 2.6.0, 2.8.0, 2.7.1
> Reporter: Omkar Vinit Joshi
> Assignee: Ajith S
> Attachments: YARN-574.03.patch, YARN-574.1.patch, YARN-574.2.patch
>
>
> At present private resources will be downloaded in parallel only if multiple
> containers request the same resource. However otherwise it will be serial.
> The protocol between PrivateLocalizer and ContainerLocalizer supports
> multiple downloads however it is not used and only one resource is sent for
> downloading at a time.
> I think we can increase / assure parallelism (even for single container
> requesting resource) for private/application resources by making multiple
> downloads per ContainerLocalizer.
> Total Parallelism before
> = number of threads allotted for PublicLocalizer [public resource] + number
> of containers[private and application resource]
> Total Parallelism after
> = number of threads allotted for PublicLocalizer [public resource] + number
> of containers * max downloads per container [private and application resource]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
