gistfile1.txt

# k3d cluster delete prophet && k3d cluster create prophet -p “8000:31111@server:0” -p “8080:31112@server:0"
# helm upgrade -i kedify-agent oci://ghcr.io/kedify/charts/kedify-agent --version=v0.3.0 --create-namespace -nkeda -f dev.values
# helm upgrade -i kedify-agent kedifykeda/kedify-agent --version=v0.3.2 --create-namespace -nkeda -f dev.values
# dev.values:
clusterName: demo-predictor
keda:
  enabled: true
  image:
    keda:
      tag: v2.18.0-1
  env:
  # expose gRPC api for obtaining the raw metrics from KEDA
  - name: RAW_METRICS_GRPC_PROTOCOL
    value: enabled
keda-add-ons-http:
  enabled: true
  scaler:
    replicas: 0
  interceptor:
    scaledObject:
      cpuTrigger:
        enabled: false
      memoryTrigger:
        enabled: false
kedify-predictor:
  enabled: true
  image:
    tag:
      v0.0.2-dev
  kedaPredictionController:
    retrainUsingJobs: false
  service:
    type: NodePort
    nodePort: 31111
  settings:
    logs:
      logLvl: debug
agent:
  orgId: ***
  apiKey: kfy_***
  kedifyServer: kedify-proxy.api.dev.kedify.io:443
  extraPostInstallObjects:
  # Deployment & Service that feeds the metrics-api scaler
  - apiVersion: apps/v1
    kind: Deployment
    metadata:
      name: mm
      namespace: default
    spec:
      replicas: 1
      selector:
        matchLabels:
          app: mm
      template:
        metadata:
          labels:
            app: mm
        spec:
          containers:
            - name: main
              image: ghcr.io/kedify/sample-minute-metrics
              ports:
                - containerPort: 8080
              env:
                - name: INTERPOLATE_VALUES
                  value: “true”
                - name: TIME_RELATIVE_TO_START
                  value: “false”
                - name: CYCLE_MINUTES
                  value: “1440”
                - name: SCHEDULE
                  # simulating two traffic peaks during a day (this produces a similar graph as the csv file on which the model was trained)
                  value: “0:2,60:4,120:5,180:13,240:38,300:109,360:329,420:715,480:1408,540:2001,600:2393,660:2111,720:1450,780:888,840:682,900:744,960:1053,1020:1485,1080:1777,1140:1876,1200:1655,1260:1209,1320:836,1380:445,1410:100"
  - apiVersion: v1
    kind: Service
    metadata:
      name: mm
      namespace: default
    spec:
      selector:
        app: mm
      ports:
        - protocol: TCP
          port: 80
          targetPort: 8080
  # ScaledObject that has two triggers and scaling modifiers that takes the average from the two:
  # - one that obtains the metric from metric-api and at the same time feeds this metric to kedify-predictor (where we create a model for this timeseries)
  # - second that takes forecasted value from the trained model
  - apiVersion: keda.sh/v1alpha1
    kind: ScaledObject
    metadata:
      name: mm
      namespace: default
    spec:
      scaleTargetRef:
        name: demo
      pollingInterval: 10
      cooldownPeriod: 5
      minReplicaCount: 0
      maxReplicaCount: 6
      advanced:
        scalingModifiers:
          formula: “(mm + mmInTenMinutes)/2”
          target: “400"
          metricType: “AverageValue”
        horizontalPodAutoscalerConfig:
          behavior:
            scaleUp:
              stabilizationWindowSeconds: 0
            scaleDown:
              stabilizationWindowSeconds: 0
      triggers:
        - type: metrics-api
          name: mm
          metadata:
            url: “http://mm.default.svc.cluster.local/api/v1/minutemetrics”
            valueLocation: “value”
            targetValue: “400"
        - type: kedify-predictive
          name: mmInTenMinutes
          metadata:
            modelName: mm-in-ten
            targetValue: “400”
  # Simple demo app that is subject of scaling (Deployment, Service & two ConfigMaps)
  - apiVersion: apps/v1
    kind: Deployment
    metadata:
      name: demo
      namespace: default
    spec:
      replicas: 1
      selector:
        matchLabels:
          app: demo
      strategy:
        type: Recreate
      template:
        metadata:
          labels:
            app: demo
        spec:
          containers:
            - name: nginx
              image: nginx
              ports:
                - containerPort: 8080
              volumeMounts:
                - name: nginx-config
                  mountPath: /etc/nginx/conf.d/default.conf
                  subPath: nginx.conf
                - name: nginx-index
                  mountPath: /usr/share/nginx/html/
          volumes:
            - name: nginx-index
              configMap:
                name: demo-index-html
                defaultMode: 420
            - name: nginx-config
              configMap:
                name: demo-config
                defaultMode: 420
  - apiVersion: v1
    kind: Service
    metadata:
      name: demo
      namespace: default
    spec:
      type: NodePort
      ports:
        - port: 8080
          name: http
          nodePort: 31112
      selector:
        app: demo
  - apiVersion: v1
    kind: ConfigMap
    metadata:
      name: demo-config
      namespace: default
    data:
      nginx.conf: |
        server {
          listen 8080;
          root /usr/share/nginx/html/;
          location /status {
            stub_status on;
            access_log  on;
            allow all;
          }
          location / {
          }
        }
  - apiVersion: v1
    kind: ConfigMap
    metadata:
      name: demo-index-html
      namespace: default
    data:
      index.html: |
        <html>
          <body style=“align-content: center; background: cadetblue; width: 20%; margin: auto;“>
            <h1 style=“font-family: monospace; color: white; text-shadow: 2px 2px 6px #444444;“>
                Hello KEDA Prediction Controller!
            </h1>
            <br/><br/><br/>
            <h2 style=“font-family: monospace; color: white; text-shadow: 2px 2px 6px #444444;“>
            Model
            </h2>
            <img src=“http://localhost:8000/models/mm-in-ten/graph”/>
          </body>
        </html>
  # CR for kedify-predictor where we request the metric subscription on trigger called ‘mm’ belonging to SO ‘mm’
  # so in other words it will create a Prophet model that will be retrained each day on the live incoming data
  # visualization of prediction will be eventually available at: http://kedify-predictor.keda.svc/models/mm-in-ten/graph
  # note: to speed up the initial phase, there is also oneShotCsv source where we take data from CSV and ingest the
  # metrics for the model (it would work also w/o it, but one would have to wait a day or two)
  - kind: MetricPredictor
    apiVersion: keda.kedify.io/v1alpha1
    metadata:
      name: mm-in-ten
      namespace: default
    spec:
      source:
        keda:
          name: mm
          triggerName: mm
        oneShotCsv:
          url: https://storage.googleapis.com/kedify-predictor/website_traffic_4weeks_30s.csv
          addTimestamps: true
          timestampPeriod: 30s
      model:
        type: Prophet
        name: mm-in-ten
        defaultHorizon: 10m
        retrainInterval: 1d