capacity-checker-deploy.yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: capacity-checker-deploy
spec:
  replicas: 1  
  selector:
    matchLabels:
      app: capacity-checker
  template:
    metadata:
      labels:
        app: capacity-checker
    spec:
      serviceAccountName: appsimulator
      nodeSelector:
        alpha.eksctl.io/nodegroup-name: kub316-ng
      containers:
      - name: python-script
        image: 891377065549.dkr.ecr.us-west-2.amazonaws.com/stablediffusion:amd64-neuron-assets
        imagePullPolicy: Always
        command:
          - /bin/bash
          - -c
          - |
            found_capacity_issues=0
            while true; do
              python /scripts/script.py
              insuf_count=$(find / -name "amd-*" -type f 2>/dev/null | wc -l)
              if [ "$insuf_count" -gt 0 ]; then
                echo "Found capacity issues: going to optimize for capacity"
                curl -sL https://raw.githubusercontent.com/aws-samples/scalable-hw-agnostic-inference/master/sd21-equal-routing-ing.yaml | kubectl apply -f -
                curl -sL https://raw.githubusercontent.com/aws-samples/scalable-hw-agnostic-inference/master/sd21-scaledobject-equal-routing.yaml | kubectl apply -f -
                rm /amd-*
                found_capacity_issues=1
              fi
              if [ "$found_capacity_issues" -eq 1 ]; then 
                load_cycle=$(kubectl get deploy -n load load -o jsonpath='{.status.readyReplicas}') 
                if [ "$load_cycle" -ge 1 ] && [ "$load_cycle" -le 5 ]; then
                  echo "Started a new cycle so going back to weighted-routing to optimize for cost"
                  curl -sL https://raw.githubusercontent.com/aws-samples/scalable-hw-agnostic-inference/master/sd21-scaledobject-weighted-routing.yaml | kubectl apply -f -
                  curl -sL https://raw.githubusercontent.com/aws-samples/scalable-hw-agnostic-inference/master/sd21-weighted-routing-ing.yaml | kubectl apply -f -
                  found_capacity_issues=0
                else
                  echo "In a load cycle; keep looking...."
                fi
              fi
              sleep 300
            done
        volumeMounts:
        - name: script-volume
          mountPath: /scripts
      volumes:
      - name: script-volume
        configMap:
          name: capacity-checker-config