Kueue Scheduler
Kueue Scheduler manages the scheduling policies that assign workloads to queues and allocate cluster resources according to priority.
Scheduling Policy Overview
The Kueue scheduler processes Workloads in the following order:
- Quota check: Verify the Workload can run within the Cluster Queue's nominalQuota
- Cohort Borrowing: Re-check feasibility by borrowing unused quota from other Cluster Queues in the same Cohort
- Preemption: Preempt lower-priority Workloads to admit higher-priority ones
Priority Configuration
| Setting | Description |
|---|---|
| WorkloadPriorityClass | Assign a priority class directly to a Workload |
| PriorityClass (k8s) | Reference a Kubernetes PriorityClass |
Cohort Configuration
A Cohort is a group of Cluster Queues that share quota:
- nominalQuota: Baseline guaranteed resource amount
- borrowingLimit: Maximum resources that can be borrowed from other Queues
- lendingLimit: Maximum resources that can be lent to other Queues
Preemption Policies
| Policy | Description |
|---|---|
Never | Preemption disabled |
LowerPriority | Preempt only lower-priority Workloads |
LowerOrNewerEqualPriority | Preempt lower or equally-prioritized newer Workloads |
WARNING
Changing preemption policies can affect running Workloads. Coordinate with your team before making changes.