Q: Your enterprise GitOps infrastructure has grown to 550 EKS/GKE clusters and 12,000 Argo CD Applications. SREs report severe performance degradation: the Argo CD UI takes 45 seconds to load, reconciliation loops lag by over 30 minutes, and the single `argocd-repo-server` frequently encounters OOMKilled crashes during Git manifest generation. You must re-architect Argo CD for high availability and enterprise scale, implementing controller sharding, repo-server horizontal autoscaling with Redis caching, and resource exclusion rules.

The `argocd-application-controller` is stateful. Scale it horizontally by configuring dynamic cluster sharding using the `ARGOCD_CONTROLLER_REPLICAS` environment variable or Argo CD operator sharding algorithms.

apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: argocd-application-controller
  namespace: argocd
spec:
  replicas: 5
  template:
    spec:
      containers:
        - name: argocd-application-controller
          env:
            - name: ARGOCD_CONTROLLER_REPLICAS
              value: '5'
          args:
            - /usr/local/bin/argocd-application-controller
            - --status-processors
            - '50'
            - --operation-processors
            - '25'
            - --repo-server-timeout-seconds
            - '120'

Deploy Redis Sentinel HA for manifest and OIDC session caching. Scale `argocd-repo-server` to 10 replicas with horizontal pod autoscaling (HPA) and configure memory limits to prevent fork/exec Helm memory exhaustion.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: argocd-repo-server
  namespace: argocd
spec:
  replicas: 10
  template:
    spec:
      containers:
        - name: argocd-repo-server
          resources:
            requests:
              cpu: 1000m
              memory: 2Gi
            limits:
              cpu: 4000m
              memory: 8Gi
          env:
            - name: ARGOCD_REPO_SERVER_PARALLELISM_LIMIT
              value: '10'
            - name: REDIS_SERVER
              value: 'argocd-redis-ha-haproxy:6379'

Prevent the application controller from monitoring high-churn Kubernetes resources (e.g., EndpointSlices, PodMetrics, Helm temporary secrets) by tuning `resource.exclusions` in the `argocd-cm` ConfigMap.

apiVersion: v1
kind: ConfigMap
metadata:
  name: argocd-cm
  namespace: argocd
data:
  resource.exclusions: |
    - apiGroups:
        - discovery.k8s.io
      kinds:
        - EndpointSlice
    - apiGroups:
        - metrics.k8s.io
      kinds:
        - PodMetrics
    - apiGroups:
        - events.k8s.io
      kinds:
        - Event

Expose and monitor Argo CD Prometheus metrics: `argocd_app_reconcile_bucket`, `argocd_cluster_api_resource_objects`, and `argocd_git_request_total`. Configure alerts when workqueue depth exceeds 100 items for > 5 minutes.