scale-from-zero

Test automatic scale from zero

This example shows that NSEs can be created on the fly on NSC requests. This allows effective scaling for endpoints. The requested endpoint will be automatically spawned on the same node as NSC (see step 12), allowing the best performance for connectivity.

Here we are using an endpoint that automatically shuts down when it has no active connection for specified time. We are using very short timeout for the purpose of the test: 15 seconds.

We are only using one client in this test, so removing it (see step 13) will cause the NSE to shut down.

Supplier watches for endpoints it created and clears endpoints that finished their work, thus saving cluster resources (see step 14).

Run

1. Create test namespace:

NAMESPACE=($(kubectl create -f https://raw.githubusercontent.com/networkservicemesh/deployments-k8s/c89573ec0f82ee02c425c43da72b026397cb52fc/examples/features/namespace.yaml)[0])
NAMESPACE=${NAMESPACE:10}

2. Select nodes to deploy NSC and supplier:

NODES=($(kubectl get nodes -o go-template='{{range .items}}{{ if not .spec.taints }}{{ .metadata.name }} {{end}}{{end}}'))
NSC_NODE=${NODES[0]}
SUPPLIER_NODE=${NODES[1]}
if [ "$SUPPLIER_NODE" == "" ]; then SUPPLIER_NODE=$NSC_NODE; echo "Only 1 node found, testing that pod is created on the same node is useless"; fi

3. Create patch for NSC:

cat > patch-nsc.yaml <<EOF
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: nsc-kernel
spec:
  template:
    spec:
      nodeName: $NSC_NODE
      containers:
        - name: nsc
          env:
            - name: NSM_NETWORK_SERVICES
              value: kernel://autoscale-icmp-responder/nsm-1
            - name: NSM_REQUEST_TIMEOUT
              value: 30s
EOF

4. Create patch for supplier:

cat > patch-supplier.yaml <<EOF
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: nse-supplier-k8s
spec:
  template:
    spec:
      nodeName: $SUPPLIER_NODE
      containers:
        - name: nse-supplier
          env:
            - name: NSE_SERVICE_NAME
              value: autoscale-icmp-responder
            - name: NSE_LABELS
              value: app:icmp-responder-supplier
            - name: NSE_NAMESPACE
              valueFrom:
                fieldRef:
                  fieldPath: metadata.namespace
            - name: NSE_POD_DESCRIPTION_FILE
              value: /run/supplier/pod-template.yaml
          volumeMounts:
            - name: pod-file
              mountPath: /run/supplier
              readOnly: true
      volumes:
        - name: pod-file
          configMap:
            name: supplier-pod-template-configmap
EOF

5. Create customization file:

cat > kustomization.yaml <<EOF
---
apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization

namespace: $NAMESPACE

bases:
- https://github.com/networkservicemesh/deployments-k8s/apps/nse-supplier-k8s?ref=c89573ec0f82ee02c425c43da72b026397cb52fc
- https://github.com/networkservicemesh/deployments-k8s/apps/nsc-kernel?ref=c89573ec0f82ee02c425c43da72b026397cb52fc

patchesStrategicMerge:
- patch-nsc.yaml
- patch-supplier.yaml

configMapGenerator:
  - name: supplier-pod-template-configmap
    files:
      - https://raw.githubusercontent.com/networkservicemesh/deployments-k8s/c89573ec0f82ee02c425c43da72b026397cb52fc/examples/features/scale-from-zero/pod-template.yaml
EOF

6. Register network service:

kubectl apply -f https://raw.githubusercontent.com/networkservicemesh/deployments-k8s/c89573ec0f82ee02c425c43da72b026397cb52fc/examples/features/scale-from-zero/autoscale-netsvc.yaml

7. Deploy NSC and supplier:

kubectl apply -k .

8. Wait for applications ready:

kubectl wait -n $NAMESPACE --for=condition=ready --timeout=1m pod -l app=nse-supplier-k8s

kubectl wait -n $NAMESPACE --for=condition=ready --timeout=1m pod -l app=nsc-kernel

kubectl wait -n $NAMESPACE --for=condition=ready --timeout=1m pod -l app=nse-icmp-responder

9. Find NSC and NSE pods by labels:

NSC=$(kubectl get pod -n $NAMESPACE --template '{{range .items}}{{.metadata.name}}{{"\n"}}{{end}}' -l app=nsc-kernel)
NSE=$(kubectl get pod -n $NAMESPACE --template '{{range .items}}{{.metadata.name}}{{"\n"}}{{end}}' -l app=nse-icmp-responder)

10. Check connectivity:

kubectl exec $NSC -n $NAMESPACE -- ping -c 4 169.254.0.0

kubectl exec $NSE -n $NAMESPACE -- ping -c 4 169.254.0.1

11. Check that the NSE spawned on the same node as NSC:

NSE_NODE=$(kubectl get pod -n $NAMESPACE --template '{{range .items}}{{.spec.nodeName}}{{"\n"}}{{end}}' -l app=nse-icmp-responder)

if [ $NSC_NODE == $NSE_NODE ]; then echo "OK"; else echo "different nodes"; false; fi

12. Remove NSC:

kubectl scale -n $NAMESPACE deployment nsc-kernel --replicas=0

13. Wait for the NSE pod to be deleted:

kubectl wait -n $NAMESPACE --for=delete --timeout=1m pod -l app=nse-icmp-responder

Cleanup

Delete namespace:

kubectl delete ns ${NAMESPACE}

Delete network service:

kubectl delete -n nsm-system networkservices.networkservicemesh.io autoscale-icmp-responder