Configure preemption in Yarn Fair Scheduler

[#1]

Hi All, Is there any reference doc for configuring pre-emption is YARN Fair Scheduler?

[#2]

HDP dont support Fair Scheduler, it support only capacity scheduler, you can check the following documents to configure pre-emption in capacity scheduler

http://hortonworks.com/blog/better-slas-via-resource-preemption-in-yarns-capacityscheduler/

[#3]

@Rajkumar Singh, does that mean that setting the class to org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.FairScheduler does not take into effect?

[#4]

@Arun A K it does take effect after setting yarn.resourcemanager.scheduler.class value to org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.FairScheduler using ambari but there is no property in ambari with the name of yarn.scheduler.fair.allocation.file so you need to define the custom property with the name of yarn.scheduler.fair.allocation.file and set it to fair-scheduler.xml.

[#5]

@Rajkumar Singh, okay got it. One more related question - On which node should I be placing the fair allocation xml file?

[#6]

on RM node and restart the RM.

[#7 · Accepted solution]

Hi Arun, FairScheduler is not HDP recommended/supported resource scheduler, so we don't have document to cover it. Please refer apache one: https://hadoop.apache.org/docs/r2.7.3/hadoop-yarn/hadoop-yarn-site/FairScheduler.html.

I would like to add more background here: preemption feature is firstly added into CapacityScheduler which is quit mature and production ready. For FairScheduler, I am not exactly sure status (alpha or GA) but just notice several fixes are going on in community: https://issues.apache.org/jira/browse/YARN-4752

Any special reason to use FairScheduler? If not, you can also try preemption for CapacityScheduler. Here is doc: https://docs.hortonworks.com/HDPDocuments/HDP2/HDP-2.4.2/bk_yarn_resource_mgt/content/preemption.htm...

[#8]

Thanks @Junping Du

[#9]

Hi Arun - This post describes how you can use preemption with a fair policy in the capacity scheduler. It should give you a similar behavior to the fair scheduler but using a YARN capacity queue.

https://community.hortonworks.com/articles/44079/yarn-pre-emption-with-spark-using-a-fair-policy.htm...

[#10]

Thanks Chris. I will have a look.