[MESOS-2653] Slave should act on correction events from QoS controller

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.23.0
• Labels: mesosphere
• Target Version/s: 0.23.0
• Epic Link: Oversubscription
• Sprint: Mesosphere Sprint 12
• Story Points: 8
• Assignee: Niklas Quarfot Nielsen
• Reporter: Vinod Kone
• Shepherd: Jie Yu
• Created: 22/Apr/15 22:26
• Updated: 26/Nov/18 12:20
• Resolved: 26/Nov/18 12:20

Description

Slave might want to kill revocable tasks based on correction events from the QoS controller.

The QoS controller communicates corrections through a stream (or process::Queue) to the slave which corrections it needs to carry out, in order to mitigate interference with production tasks.

The correction is communicated through a message:
[code]
message QoSCorrection {
enum CorrectionType

{ KillExecutor = 1 // KillTask = 2 // Resize, throttle task }

optional string reason = X;
optional ExecutorID executor_id = X;
// optional TaskID task_id = X;
}
[/code]

And the slave will setup a handler to process these events. Initially, only executor termination is supported and cause the slave to issue 'containerizer->destroy()'.

Issue Links

• blocks: MESOS-2704 Add tests for QoS controller corrections