How to kill a YARN application

Killing a YARN application stops its ApplicationMaster and containers when a job is stuck, misqueued, or consuming resources that must be returned. The operation is destructive for the running job, so identify the application ID and owner before killing it.

The kill request goes to the ResourceManager, which checks the caller's access before acting. With yarn.acl.enable set to true, only the application owner, a YARN administrator, a user in the application's modify ACL, or a queue administrator can kill the application, while a cluster with ACLs off accepts the request from any user.

Prefer an application-level cancel command from the processing engine when it preserves metadata or cleanup better than a YARN kill. Use the YARN command when the cluster scheduler must terminate the application directly.

1. List running YARN applications.

   $ yarn application -list -appStates RUNNING
   Total number of applications (application-types: [] and states: [RUNNING]):1
   Application-Id                  Application-Name User  Queue    State
   application_1720000000000_0042  daily-etl        alice analytics RUNNING

   Related: How to list YARN applications

2. Check the application owner and queue.

   $ yarn application -status application_1720000000000_0042
   Application-Id : application_1720000000000_0042
   Application-Name : daily-etl
   User : alice
   Queue : analytics
   State : RUNNING

3. Kill the exact application ID.

   $ yarn application -kill application_1720000000000_0042
   Killing application application_1720000000000_0042

4. Verify the application final state.

   $ yarn application -status application_1720000000000_0042
   State : KILLED
   Final-State : KILLED

5. Check that the queue no longer shows the killed job as running.

   $ yarn application -list -appStates RUNNING
   Total number of applications (application-types: [] and states: [RUNNING]):0