kubectl autoscale - Auto-scale a deployment, replica set, stateful set, or replication controller

Options you will use

--max=N

upper limit for the number of pods. Required.

--min=N

lower limit; if not specified the server applies its default (1).

--cpu=70%|500m

target CPU: a percentage is average UTILIZATION of the pods' CPU requests; a quantity is an average VALUE. Without --cpu/--memory the server default is 80% CPU utilization.

--memory=60%|200Mi

target memory, same two forms

--cpu-percent=N

deprecated in 1.34: use --cpu=N%

--name=NAME

name of the HPA (default: the target's name)

--dry-run=client -o yaml

print the autoscaling/v2 HorizontalPodAutoscaler instead of creating it

Examples

$ kubectl autoscale deployment php-apache --cpu=50% --min=1 --max=10

the upstream walkthrough

$ kubectl get hpa -w

watch TARGETS and REPLICAS change

$ kubectl describe hpa php-apache

Conditions and Events explain every decision (or why it cannot decide)

Gotchas

• Utilization is measured against REQUESTS. A container with no cpu request makes the HPA print cpu: <unknown>/50% and the event "missing request for cpu in container ...".
• Scale-up has no stabilization by default; scale-down waits 300s (the highest recommendation of the last 5 minutes wins) to stop flapping.