[ 
https://issues.apache.org/jira/browse/YARN-7792?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16342363#comment-16342363
 ] 

Arun Suresh commented on YARN-7792:
-----------------------------------

bq. BTW, RejectedSchedulingRequest#reason right now is not set, which makes AM 
no way to know why it's rejected..
It is set by the PlacementProcessor in two situations.
# when the algorithm cannot 
[place|https://github.com/apache/hadoop/blob/YARN-6592/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/constraint/processor/PlacementProcessor.java#L273-L275]
# when a placed requests cannot be 
[scheduled|https://github.com/apache/hadoop/blob/YARN-6592/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/constraint/processor/PlacementProcessor.java#L286-L288]
 by the Scheduler - this has a retry of around 5 by default.

bq. , previously, there isn't such a notion of rejecting resource requests due 
to transient states - requests that aren't satisfied due to user-limit, 
capacity etc. are just getting pending in RM. AM doesn't need to keep on 
retrying those requests.
Yup - Out take currently is to let the AM decide what to do if placement / 
scheduling was not successful. As we mentioned in the doc, currently we assume 
constraints are *hard*. If placement is not possible, we don't loosen the 
constraints. Out plan is to introduce the following in future work (we will 
open JIRAs in YARN-7812):
# Support *soft* constraints: If retry is expired, send it down to the 
Scheduler as an ANY request.
# Support for DELAYED_OR composite constraint: If the placement was not 
possible in x attempts, try the next Constraint.




> Merge work for YARN-6592
> ------------------------
>
>                 Key: YARN-7792
>                 URL: https://issues.apache.org/jira/browse/YARN-7792
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>            Reporter: Sunil G
>            Priority: Blocker
>         Attachments: YARN-6592.001.patch, YARN-7792.002.patch, 
> YARN-7792.003.patch
>
>
> This Jira is to run aggregated YARN-6592 branch patch against trunk and check 
> for any jenkins issues.



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

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to