Combination of replicas definition and HPA keeps scaling pods between same nums.

[nakkoh]

Checklist:

• I've searched in the docs and FAQ for my answer: https://bit.ly/argocd-faq.
• I've included steps to reproduce the bug.
• I've pasted the output of argocd version.

Describe the bug

I push the following manifest files into Git repository.

• Deployment with the following definition

  spec:
    replicas: 1

• HPA with the following definition

  spec:
    minReplicas: 2
    maxReplicas: 10

I created the Application to create the above manifest.
After that, pods deployed, and kept beeing scaled to 1 or 2 again and agian.

This time the target of HPA was unknown.

$ k -n test-hpa get hpa nginx 
NAME    REFERENCE          TARGETS       MINPODS   MAXPODS   REPLICAS   AGE
nginx   Deployment/nginx   <unknown>/1   2         10        1          17h

But HPA can get target metrics if the app was disabled auto-sync of Argo CD.

[How to work around]

Set the following definition into Application definition.

spec:
  ignoreDifferences:
  - group: apps
    kind: Deployment
    name: nginx
    namespace: test-hpa
    jsonPointers:
    - /spec/replicas

To Reproduce

Create the Application by the following manifests.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx
  namespace: test-hpa
  labels:
    app: nginx
spec:
  replicas: 1
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
    spec:
      containers:
      - name: nginx
        image: nginx:1.14.2
        ports:
        - containerPort: 80
---
apiVersion: autoscaling/v2beta1
kind: HorizontalPodAutoscaler
metadata:
  name: nginx
  namespace: test-hpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: nginx
  minReplicas: 2
  maxReplicas: 10
  metrics:
  - type: Resource
    resource:
      name: cpu
      targetAverageValue: 1

Expected behavior

The replicas defined in Deployment manifest is ignored and the pod num is decided by HPA.

Screenshots

Version

argocd: v1.6.2+3d1f37b
  BuildDate: 2020-07-31T23:43:46Z
  GitCommit: 3d1f37b0c53f4c75864dc7339e2831c6e6a947e0
  GitTreeState: clean
  GoVersion: go1.14.1
  Compiler: gc
  Platform: linux/amd64
argocd-server: v1.8.1+c2547dc
  BuildDate: 2020-12-10T02:59:21Z
  GitCommit: c2547dca95437fdbb4d1e984b0592e6b9110d37f
  GitTreeState: clean
  GoVersion: go1.14.12
  Compiler: gc
  Platform: linux/amd64
  Ksonnet Version: v0.13.1
  Kustomize Version: v3.8.1 2020-07-16T00:58:46Z
  Helm Version: v3.4.1+gc4e7485
  Kubectl Version: v1.17.8

Logs

[jessesuen]

This behavior is expected. Your git desired state conflicts with a controller adjusting the replicas so it's expected that they will fight over the correct value. You should omit replicas from your deployment if you wish it to be controlled by HPA. See:

https://argoproj.github.io/argo-cd/user-guide/best_practices/#leaving-room-for-imperativeness

[nakkoh]

Thank you for your reply.
I understood.