How to Manage pod resources

There are 3 different levels how pod resources can be managed. Each level takes precedence over the previous one.

Global defaults
resourceRequirementsTemplate field from a Schedule object
resources fields in a Backup, Prune, Check, Archive or Restore spec within a Schedule.

Global defaults are only applied through Schedules. If you create your own Backup object for example, neither global default nor any specs from resourceRequirementsTemplate from any Schedule will be applied.

The computed resources resulting from the merge described above are not persisted in the Schedules. This allows cluster administrator to reconfigure global resource defaults without having to touch each Schedule after a change.

Global default

Let’s assume the cluster administrator or a resource quota requires to set memory limit on a all pods spawned by K8up. In that case K8up needs to be started with the following environment variable:

BACKUP_GLOBAL_MEMORY_LIMIT=200M

See Operator Config reference to see which environment variables are supported.

Following example is a simple Schedule example without further resource specs.

apiVersion: k8up.io/v1
kind: Schedule
metadata:
  name: schedule-test
spec:
  ...
  prune:
    schedule: '*/5 * * * *'

This will create a Prune object with the resource default applied:

apiVersion: k8up.io/v1
kind: Prune
metadata:
  name: schedule-test-prune-qbq2k
spec:
  ...
  resources:
    limits:
      memory: 200M

This object in turn spawns a Job with the given resource spec.

Resources from a template

Following example is a Schedule example that also sets CPU request for all jobs.

apiVersion: k8up.io/v1
kind: Schedule
metadata:
  name: schedule-test
spec:
  resourceRequirementsTemplate:
    requests:
      cpu: "50m"
  ...
  check:
    resources:
      requests:
        memory: "64Mi"
        cpu: "250m"
    schedule: '@hourly-random'
  prune:
    schedule: '*/2 * * * *'

Combined with the global memory limit, the Prune object will have following resources:

apiVersion: k8up.io/v1
kind: Prune
metadata:
  name: schedule-test-prune-qbq2k
spec:
  ...
  resources:
    requests:
      cpu: "50m"
    limits:
      memory: 200M

Whereas, the Check object will overwrite the CPU resource request template with its own specification:

apiVersion: k8up.io/v1
kind: Check
metadata:
  name: schedule-test-check-q5kae
spec:
  ...
  resources:
    requests:
      memory: "64Mi"
      cpu: "250m"
    limits:
      memory: 200M