Headroom Overview

Headroom will allow you to keep a little bit of extra capacity on top of your cluster to allow rapid scheduling of new PODs (usually triggered by the HPA or VPA) and it’ll also assist with ensuring that enough capacity is available when a Node interruption/preemption will occur.

Headroom can be configured automatically and also manually

Automatic Headroom

Automatic refers to headroom that is dynamic and designed to accommodate the next scale up of services in the cluster. It is configured on the Ocean level as a percentage of the entire cluster resources. The automatic headroom is being calculated hourly and will accommodate according to the required amount of resources in the cluster.

Manual Headroom

Manual headroom refers to headroom requirements that are specified by the user, the manual headroom is configured on either the Ocean and/or VNG level and is specified as headroom units that consist of the following components:

  • Number of units
  • vCPU per unit
  • Memory per unit
  • GPU per unit

As of recently Ocean supports configuring both automatic and manual headroom on the same cluster (one on top of the other)