Github user tgravescs commented on the issue: https://github.com/apache/spark/pull/19267 for the admin type command we would need to have an RPC between the client command and the driver. There is a jira for that already (https://issues.apache.org/jira/browse/SPARK-21737) but its not done yet. I can see the admin command useful as it could be used in any cluster - yarn, mesos, standalone. It also doesn't appear the yarn support will be done any time soon unless we push on it so the only option would be that admin type command. So overall I'm fine with that approach it just needs to have more details about how that would work. The admin command on yarn wouldn't necessarily need to look at all apps. If you are decommissioning a single NM it could go to the rest api and get the list of apps running on it and just go to those. I'm curious, Are you running into a particular case of this and if so what env?
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org