The yaml I used is shown below
apiVersion: v1
kind: Service
metadata:
name: xxx-svc
labels:
app: xxxxxx
spec:
type: NodePort
ports:
- port: 8080
selector:
app: xxxxxx
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: my-xxx
labels:
app: xxxxxx
spec:
selector:
matchLabels:
app: xxxxxx
template:
metadata:
labels:
app: xxxxxx
spec:
containers:
- name: xxxxxx
image: yyy/xxxxxx:latest
ports:
- containerPort: 8080
resources:
requests:
cpu: "100m"
memory: "504Mi"
limits:
cpu: "100m"
memory: "504Mi"
---
apiVersion: autoscaling/v2beta2
kind: HorizontalPodAutoscaler
metadata:
name: xxxxxx
spec:
scaleTargetRef:
apiVersion: apps/v1
kind: Deployment
name: my-xxx
minReplicas: 1
maxReplicas: 3
metrics:
- type: Resource
resource:
name: cpu
target:
type: Utilization
averageUtilization: 50
- type: Resource
resource:
name: memory
target:
type: Value
averageValue: 500MiService, HPA, Deployment everything deployed successfully, but when I check hpa (kubectl get hpa) I am getting below result
NAME REFERENCE TARGETS MINPODS
MAXPODS REPLICAS AGE
xxxxxx Deployment/my-xxx unknown/500Mi, 1%/50% 1 3 3 69mThe reason I got (kubectl describe hpa) is
Warning FailedComputeMetricsReplicas 21m (x4 over 22m) horizontal-pod-autoscaler failed to get memory utilization: missing request for memory
What might be the reason that memory is Unknown but CPU is working
Kubernetes auto scaling for memory is not applied yet. You can write a script which tops on the desired pods and get the memory value and divide it to the desired memory to get percentage and scale up or scale down pods. Just run the script in a cron job which keeps on checking the stats in a frequent interval.
The reason for this:
Warning FailedComputeMetricsReplicas 21m (x4 over 22m) horizontal-pod-autoscaler failed to get memory utilization: missing request for memory
Kubernetes HPA does not work by default with memory you need to create custom metric for memory and then use it. I found some additional information here how people try to solve same issue.
Pod Memory Based AutoScaling
In this section, we are discussing how you can deploy autoscaling on the basis of memory that pods are consuming. We have used the command “kubectl top pod” to get the utilized pod memory and applied the logic.
- Get the average pod memory of the running pods: Execute the script as follows: