KEDA Autoscaling example

This notebook shows how you can scale Seldon Deployments based on Prometheus metrics via KEDA.

KEDA is a Kubernetes-based Event Driven Autoscaler. With KEDA, you can drive the scaling of any container in Kubernetes based on the number of events needing to be processed.

With the support of KEDA in Seldon, you can scale your seldon deployments with any scalers listed here. In this example we will scale the seldon deployment with Prometheus metrics as an example.

Install Seldon Core

Install Seldon Core as described in docs

Make sure add --set keda.enabled=true

Install Prometheus

!kubectl create namespace seldon-monitoring
!helm upgrade --install seldon-monitoring kube-prometheus-stack \
    --version 44.4.1 \
    --set fullnameOverride=seldon-monitoring \
    --namespace seldon-monitoring \
    --repo https://prometheus-community.github.io/helm-charts/ \
    --wait

Error from server (AlreadyExists): namespaces "seldon-monitoring" already exists
Release "seldon-monitoring" has been upgraded. Happy Helming!
NAME: seldon-monitoring
LAST DEPLOYED: Thu Dec  4 11:20:53 2025
NAMESPACE: seldon-monitoring
STATUS: deployed
REVISION: 2
NOTES:
kube-prometheus-stack has been installed. Check its status by running:
  kubectl --namespace seldon-monitoring get pods -l "release=seldon-monitoring"

Visit https://github.com/prometheus-operator/kube-prometheus for instructions on how to create & configure Alertmanager and Prometheus instances using the Operator.

!kubectl rollout status -n seldon-monitoring statefulsets/prometheus-seldon-monitoring-prometheus

statefulset rolling update complete 1 pods at revision prometheus-seldon-monitoring-prometheus-58fb79649...

!cat pod-monitor.yaml

apiVersion: monitoring.coreos.com/v1
kind: PodMonitor
metadata:
  name: seldon-podmonitor
  namespace: seldon-monitoring
  labels:
    release: seldon-monitoring
spec:
  namespaceSelector:
    matchNames:
      - seldon
  selector:
    matchLabels:
      seldon.io/model: "true"
  podMetricsEndpoints:
    - port: metrics
      path: /prometheus

!kubectl apply -f pod-monitor.yaml

podmonitor.monitoring.coreos.com/seldon-podmonitor unchanged

Install KEDA

Follow the docs for KEDA to install.

Create model with KEDA

To create a model with KEDA autoscaling you just need to add a KEDA spec referring in the Deployment, e.g.:

kedaSpec:
  pollingInterval: 15                                # Optional. Default: 30 seconds
  minReplicaCount: 1                                 # Optional. Default: 0
  maxReplicaCount: 5                                 # Optional. Default: 100
  triggers:
  - type: prometheus
          metadata:
            # Required
            serverAddress: http://seldon-monitoring-prometheus.seldon-monitoring.svc.cluster.local:9090
            metricName: access_frequency
            threshold: '10'
            query: rate(seldon_api_executor_client_requests_seconds_count{model_name="classifier"}[1m])

The full SeldonDeployment spec is shown below.

VERSION = !cat ../../version.txt
VERSION = VERSION[0]
VERSION

'1.19.0-dev'

%%writefile model_with_keda_prom.yaml
apiVersion: machinelearning.seldon.io/v1
kind: SeldonDeployment
metadata:
  name: seldon-model
spec:
  name: test-deployment
  predictors:
  - componentSpecs:
    - spec:
        containers:
        - image: seldonio/mock_classifier:1.19.0-dev
          imagePullPolicy: IfNotPresent
          name: classifier
          resources:
            requests:
              cpu: '0.5'
      kedaSpec:
        pollingInterval: 15                                # Optional. Default: 30 seconds
        minReplicaCount: 1                                 # Optional. Default: 0
        maxReplicaCount: 5                                 # Optional. Default: 100
        triggers:
        - type: prometheus
          metadata:
            # Required
            serverAddress: http://seldon-monitoring-prometheus.seldon-monitoring.svc.cluster.local:9090
            metricName: access_frequency
            threshold: '10'
            query: rate(seldon_api_executor_client_requests_seconds_count{model_name="classifier"}[1m])
    graph:
      children: []
      endpoint:
        type: REST
      name: classifier
      type: MODEL
    name: example

Overwriting model_with_keda_prom.yaml

!kubectl apply -f model_with_keda_prom.yaml -n seldon

seldondeployment.machinelearning.seldon.io/seldon-model created

!kubectl wait sdep/seldon-model \
  --for=condition=ready \
  --timeout=120s \
  -n seldon

seldondeployment.machinelearning.seldon.io/seldon-model condition met

Create Load

We label some nodes for the loadtester. We attempt the first two as for Kind the first node shown will be the master.

!kubectl label nodes $(kubectl get nodes -o jsonpath='{.items[0].metadata.name}') role=locust
!kubectl label nodes $(kubectl get nodes -o jsonpath='{.items[1].metadata.name}') role=locust

node/kind-control-plane labeled
node/kind-worker labeled

Before add loads to the model, there is only one replica

!kubectl get deployment seldon-model-example-0-classifier -n seldon

NAME                                READY   UP-TO-DATE   AVAILABLE   AGE
seldon-model-example-0-classifier   1/1     1            1           41s

!helm install seldon-core-loadtesting seldon-core-loadtesting -n seldon --repo https://storage.googleapis.com/seldon-charts \
    --set locust.host=http://seldon-model-example:8000 \
    --set oauth.enabled=false \
    --set locust.hatchRate=1 \
    --set locust.clients=1 \
    --set loadtest.sendFeedback=0 \
    --set locust.minWait=0 \
    --set locust.maxWait=0 \
    --set replicaCount=1

NAME: seldon-core-loadtesting
LAST DEPLOYED: Thu Dec  4 11:23:04 2025
NAMESPACE: seldon
STATUS: deployed
REVISION: 1
TEST SUITE: None

After a few mins you should see the deployment scaled to 5 replicas

import json
import time


def getNumberPods():
    dp = !kubectl get deployment -n seldon seldon-model-example-0-classifier -o json
    dp = json.loads("".join(dp))
    return dp["status"]["replicas"]


scaled = False
for i in range(60):
    pods = getNumberPods()
    print(pods)
    if pods > 1:
        scaled = True
        break
    time.sleep(5)
assert scaled

1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
4

!kubectl get deployment/seldon-model-example-0-classifier -n seldon scaledobject/seldon-model-example-0-classifier

NAME                                                READY   UP-TO-DATE   AVAILABLE   AGE
deployment.apps/seldon-model-example-0-classifier   5/5     5            5           3m9s

NAME                                                     SCALETARGETKIND      SCALETARGETNAME                     MIN   MAX   TRIGGERS     AUTHENTICATION   READY   ACTIVE   FALLBACK   PAUSED    AGE
scaledobject.keda.sh/seldon-model-example-0-classifier   apps/v1.Deployment   seldon-model-example-0-classifier   1     5     prometheus                    True    True     False      Unknown   3m9s

Remove Load

!helm delete seldon-core-loadtesting -n seldon

release "seldon-core-loadtesting" uninstalled

After 5-10 mins you should see the deployment replica number decrease to 1

Cleanup

!kubectl delete -f model_with_keda_prom.yaml -n seldon

seldondeployment.machinelearning.seldon.io "seldon-model" deleted

!helm delete seldon-monitoring -n seldon-monitoring

release "seldon-monitoring" uninstalled

!kubectl delete namespace seldon-monitoring

namespace "seldon-monitoring" deleted