[ 
https://issues.apache.org/jira/browse/YARN-4606?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16391802#comment-16391802
 ] 

Eric Payne commented on YARN-4606:
----------------------------------

[[email protected]], thank you for the patch. The overall approach looks 
fine, but I have a couple of concerns.
 - The behavior of assigning resources to schedulable applications has changed. 
With this patch, in the following use case, resources are not assigned to the 
second app when they should be. I have not analyzed the behavior closely enough 
to debug the issue, but I wish to document the behavior:
 -- Queue1 total resources: 40
 -- Queue1 Max Application Master Resources: 2
 -- Container sizes are all 1 resource
|*User Name*|*Applicatiton ID*|*Used AM resources*|*Total Used 
Resources*|*Pending Resources*|
|User1|App1|1|39|20|
|User2|App2|0|0|1 (waiting for AM)|

 -- In this scenario, User2 wants to start App2 but User1 is consuming all 
resources in the queue with App1. When App1 releases a resource, however, it is 
not given to App2. The resource is given back to App1, which brings its Pending 
value down to 19. This is incorrect behavior since Queue1 has room for 2 AMs.

 - I think the {{TestRMHA}} unit test needs to be modified to adjust to this 
patch:
{code:java}
TestRMHA
TestRMHA.testFailoverAndTransitions:219->verifyClusterMetrics:754 Incorrect 
value for metric activeApplications expected:<1> but was:<0>
TestRMHA.testFailoverClearsRMContext:550->verifyClusterMetrics:754 Incorrect 
value for metric activeApplications expected:<1> but was:<0>
{code}

 - A couple of minor things:
 -- IIUC, the value stored in {{activeUsersOfPendingApps}} represents the 
number of suers that do not have any active applications. Is that correct? If 
so, I think it would be more clear if it were called 
{{usersWithOnlyPendingApps}}.
 -- In {{AbstractUsersManager}} and {{ActiveUsersManager}}, *atleast* should be 
"at least*.

> CapacityScheduler: applications could get starved because computation of 
> #activeUsers considers pending apps 
> -------------------------------------------------------------------------------------------------------------
>
>                 Key: YARN-4606
>                 URL: https://issues.apache.org/jira/browse/YARN-4606
>             Project: Hadoop YARN
>          Issue Type: Bug
>          Components: capacity scheduler, capacityscheduler
>    Affects Versions: 2.8.0, 2.7.1
>            Reporter: Karam Singh
>            Assignee: Wangda Tan
>            Priority: Critical
>         Attachments: YARN-4606.1.poc.patch, YARN-4606.POC.patch
>
>
> Currently, if all applications belong to same user in LeafQueue are pending 
> (caused by max-am-percent, etc.), ActiveUsersManager still considers the user 
> is an active user. This could lead to starvation of active applications, for 
> example:
> - App1(belongs to user1)/app2(belongs to user2) are active, app3(belongs to 
> user3)/app4(belongs to user4) are pending
> - ActiveUsersManager returns #active-users=4
> - However, there're only two users (user1/user2) are able to allocate new 
> resources. So computed user-limit-resource could be lower than expected.



--
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