[
https://issues.apache.org/jira/browse/YARN-4606?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15108183#comment-15108183
]
Wangda Tan commented on YARN-4606:
----------------------------------
Updated description of the JIRA, originally it is found by [~karams] while
doing fairness ordering policy tests, pasting original test cases here just for
reference:
{code}
Encountered while studying behaviour fairness with UserLimitPercent and
UserLimitFactor during following test:
Ran GridMix with Queue settings: Capacity=10, MaxCap=80, UserLimit=25
UserLimitFactor=32, FairOrderingPolicy only. Encountered a application starving
situation where 33 application (190 apps completed out of 761 apps, queue can
345 containers) are running with total of 45 containers running, and that 12
extra only one app(the app was having around 18000 tasks) , all other apps were
having AM running only no other containers were given any apps. After that app
finished, there were 32 AMs that kept running without any containers for task
being launched
GridMix was run with following settings:
gridmix.client.pending.queue.depth=10, gridmix.job-submission.policy=REPLAY,
gridmix.client.submit.threads=5, gridmix.submit.multiplier=0.0001,
gridmix.job.type=SLEEPJOB, mapreduce.framework.name=yarn,
mapreduce.job.queuename=hive1, mapred.job.queue.name=hive1,
gridmix.sleep.max-map-time=5000, gridmix.sleep.max-reduce-time=5000,
gridmix.user.resolve.class=org.apache.hadoop.mapred.gridmix.RoundRobinUserResolver
With Users file containing 4 users for RoundRobinUserResolver
{code}
> 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
>
> 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
(v6.3.4#6332)
