Minor Description
The client exposes a kill subcommand to kill jobs, which currently applies to cron and non-cron jobs. This is translated into a killTasks RPC, and the scheduler internally inspects the query to determine if this is a request to kill a whole job. Acting on this RPC for a job means that the scheduler attempts to transition a slew of tasks to KILLING state. This is benign, but inefficient and can result in lots of log noise.
In the short term, the scheduler should restrict non-cron job-scoped kills to only active tasks. Longer-term, we should expose a more explicit way for the client to ask for a cron job to be unscheduled.