Skip to content

PATTERN Cited by 1 source

Cohort-based quota hierarchy

Model a multi-level tenant hierarchy using Cohorts as parent groups and ClusterQueues as leaf consumers, enabling intra-cohort resource borrowing and preemption-based reclamation.

Context

You have a tree of organizational tenants with capacity allocations at each level. You need subtrees to share idle capacity while preserving guarantees for each node.

Solution

Map the hierarchy onto Kueue's Cohort model:

  • Internal/parent tenants → Cohorts (grouping mechanism enabling borrowing/lending)
  • Leaf tenants → ClusterQueues with nominal quotas (resource flavors define heterogeneous capacity)
  • Lending semantics — idle quota in one ClusterQueue is available to siblings within the same Cohort
  • Reclamation — when the owner queue needs capacity back, it preempts borrowed workloads (reclaimWithinCohort: Any)

At Netflix

Netflix mapped their existing CMB tenant hierarchy directly:

CMB Kueue
Internal tenant Cohort
Leaf tenant ClusterQueue + LocalQueue
Reserved capacity Nominal quotas per resource flavor
Shared capacity Cohort-level borrowing

(Source: sources/2026-06-22-netflix-how-netflix-simplified-batch-compute-with-kueue)

Seen in

Last updated · 559 distilled / 1,651 read