Bug: Recent release breaks argocd ha deployment due to Redis HAProxy configuration change

[aurimasniekis]

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
After a recent change adding automountServiceAccountToken: false to the argocd-redis-ha-haproxy deployment, the ArgoCD HA cluster fails to start properly. The pods fail with an error indicating they cannot connect to the Kubernetes API because no configuration is provided.

To Reproduce

1. Deploy ArgoCD in HA mode with the latest version
2. Observe that argocd-redis-ha-haproxy pods fail to start
3. Check logs to see the error about invalid configuration

Expected behavior
The argocd-redis-ha-haproxy pods should start successfully as part of the HA deployment.

Screenshots
N/A

Version

# Replace with your actual ArgoCD version output
argocd version
argocd: v2.14.2+ad27246
  BuildDate: 2025-02-06T02:19:55Z
  GitCommit: ad2724661b66ede607db9b5bd4c3c26491f5be67
  GitTreeState: clean
  GoVersion: go1.23.6
  Compiler: gc
  Platform: darwin/arm64

Logs

{"level":"fatal","msg":"invalid configuration: no configuration has been provided, try setting KUBERNETES_MASTER environment variable","time":"2025-03-06T10:26:55Z"}

Additional context
The issue appears to be caused by the addition of automountServiceAccountToken: false to the argocd-redis-ha-haproxy deployment in this file:
https://github.com/argoproj/argo-cd/blame/5223ce546a677565a24dc37d53a360d544891ba1/manifests/ha/base/redis-ha/chart/upstream.yaml#L1104

When I remove this setting, the cluster starts properly. This suggests that the HAProxy component requires the service account token to be mounted for Kubernetes API communication.

[andrii-korotkov-verkada]

This option has been false for quite a while, so it should be something else.

[crenshaw-dev]

@andrii-korotkov-verkada I let git blame mislead me, that change is new for the argocd-redis-ha-haproxy Deployment.

[andrii-korotkov-verkada]

Most places indeed had automountServiceAccountToken: false for years, but this change was introduced by me https://github.com/argoproj/argo-cd/blame/a8b76f29519bfeddff2be7f30ef1c5e28cdf2a45/manifests/ha/base/redis-ha/chart/upstream.yaml#L1104C7-L1104C42. For now, please use the manifests not from master, but from the version of the tag corresponding to your version 2.14.2

[andrii-korotkov-verkada]

It uses a specialized argocd-redis-ha-haproxy service account and seems to need to access k8s, I think we should just automount the token. I'll raise the PR.

[andrii-korotkov-verkada]

@aurimasniekis, can you try with most recent PR version and let me know if the issue still happens? I've updated the version of haproxy to what's the in the upstream chart and wonder if they tackled the issue.

[andrii-korotkov-verkada]

@crenshaw-dev, tested with 3.0.8, still the same error

{"level":"fatal","msg":"invalid configuration: no configuration has been provided, try setting KUBERNETES_MASTER environment variable","time":"2025-03-11T16:06:29Z"}

Looks like we need that service account token automount.