DrainSafe

Azure has a scheduled events feature which lets safely drain the workload based on planned/unplanned maintenance events.

Table of Contents

• Design
  • Events
  • Node Annotations
  • Scheduled Events Controller
  • Safe drain Controller
  • Sequence
  • Deploy

Design

Events

Following events are defined based on which controllers perform certain actions

• MaintenanceScheduled - Maintenance is scheduled on virtual machine
• NodeCordoning - Node is queued for cordoning
• NodeCordoned - Scheduling is disabled on virtual machine
• NodeDraining - Workload is queued to be drained on virtual machine
• NodeDrained - Workload is drained on virtual machine
• MaintenanceStarted - Maintenance is started on virtual machine
• NodeRunning - Maintenance is completed on virtual machine
• NodeUncordoned - Scheduling is enabled on virtual machine

Node Annotations

We use above event values and annotate the node with current state of actions drainsafe.azure.com/maintenancestate

Scheduled Events Controller

• Runs as a daemonset which watches scheduled events for virtual machine its running on.
• Annotates the node with MaintenanceScheduled when a maintenance is scheduled.
• Annotates the node with MaintenanceStarted Event when a maintenance is started.
• Annotates the node with NodeRunning Event when there are no scheduled events at daemonset startup.

Safe drain Controller

• Runs as a controller watches pre defined events as annotations on kubernetes node.
• Annotates the node with NodeCordoned when node has been corded based on MaintenanceScheduled.
• Annotates the node with NodeDrained when a node has been drained based on NodeCordoned.
• Annotates the node with NodeUncordoned when node has been uncordened based on NodeRunning.

Sequence

Deploy

kubectl apply -f https://raw.githubusercontent.com/awesomenix/drainsafe/master/config/deployment/drainsafe-deployment.yaml