Ferenc Erdelyi created YARN-11774:
-------------------------------------

             Summary: DominantResourceCalculator - Used Resources Percentage 
Metrics is Incorrect
                 Key: YARN-11774
                 URL: https://issues.apache.org/jira/browse/YARN-11774
             Project: Hadoop YARN
          Issue Type: Improvement
          Components: yarn-service
            Reporter: Ferenc Erdelyi
         Attachments: DominantResourceCalculator_repro1_metrics.png

The issue occurs using Dominant Resource Calculator

Reproduction steps:
- create two queues: root.a and root.b. Submit a vcores-heavy application to 
queue.a and memory-heavy application to queue.b.
Make sure the applications started running, then navigate to YARN UI1 and check 
the Used Resources percentage. We expect to get the percentage based on the 
given resource-heavy values. E.g. if it was vcores, then we get the ratio of 
the used vcores value and the effective vcores value and multiply it by 100 to 
get the %.

In some cases the calculation is incorrect. See the screenshot.
 !DominantResourceCalculator_repro1_metrics.png! 

root.a queue
{code:java}
hadoop jar 
/opt/cloudera/parcels/CDH/lib/hadoop-yarn/hadoop-yarn-applications-distributedshell.jar
 -shell_command 'while true; echo Timestamp: \"\"\$(date +%Y-%m-%d\ 
%H:%M:%S)\"\"; do sleep 3600; done' -jar 
/opt/cloudera/parcels/CDH/lib/hadoop-yarn/hadoop-yarn-applications-distributedshell.jar
 --num_containers 2 --master_memory 1024 –master_vcores 5 --container_memory 
1024 --container_vcores 5 -queue root.a
{code}

root.b queue
{code:java}
hadoop jar 
/opt/cloudera/parcels/CDH/lib/hadoop-yarn/hadoop-yarn-applications-distributedshell.jar
 -shell_command 'while true; echo Timestamp: \"\"\$(date +%Y-%m-%d\ 
%H:%M:%S)\"\"; do sleep 3600; done' -jar 
/opt/cloudera/parcels/CDH/lib/hadoop-yarn/hadoop-yarn-applications-distributedshell.jar
 --num_containers 2 --master_memory 5120 –master_vcores 1 --container_memory 
5120 --container_vcores 1 -queue root.b
Observation:
VCores "Used Capacity" percentage calculation is not intuitive.
Out of 5 vcores queue capacity, we used 11 vcores (over the queue capacity).
Intuitively - based on my understanding - we expect to calculate the percentage 
as 11/5*100=220, but we get a different value - 206.3
{code}


For the memory, the "Used Capacity" calculation, I was not able to confirm the 
issue, however it seems to occur from time-to-time



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-dev-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-dev-h...@hadoop.apache.org

Reply via email to