Canary Deployments

Overview

The Varnish Gateway supports weighted traffic splitting via the standard Gateway API weight field in BackendRef. Use this for canary deployments and A/B testing.

Weight Calculation

Weights are not percentages Traffic distribution:

backend_probability = backend_weight / sum(all_weights)

Example: weights 90/10 → 90% and 10% traffic split

When a service has multiple pods, each pod inherits the same weight. With 2 stable pods (weight 90) and 2 canary pods (weight 10), you get total weights of 180/20 = 90%/10% split. The ratio is preserved.

Example

yaml apiVersion: gateway.networking.k8s.io/v1 kind: HTTPRoute metadata: name: canary-route spec: parentRefs: - name: varnish-gateway hostnames: - "app.example.com" rules: - backendRefs: - name: app-stable port: 8080 weight: 90 - name: app-canary port: 8080 weight: 10

This routes 90% of requests to app-stable service and 10% to app-canary.

Progressive Rollout

Gradually increase canary weight:

1. Start: weight: 1 (1% traffic)
2. If stable: weight: 10 (10% traffic)
3. If stable: weight: 25 (25% traffic)
4. If stable: weight: 50 (50% traffic)
5. Complete: Remove stable backend or flip weights

Testing

```bash

Make 100 requests and count responses

for i in {1..100}; do curl -H "Host: app.example.com" http://GATEWAY_IP/; done | sort | uniq -c ```

Monitoring

```bash

Check backend health and request counts

kubectl exec -it deploy/varnish-gateway -- varnishadm backend.list -p ```

Complete working example

The manifest below deploys a Gateway, two backend Deployments (app-stable running hashicorp/http-echo with text stable-v1.0, and app-canary with text canary-v2.0), their Services, and an HTTPRoute that splits traffic 90/10.

```yaml

Gateway (if not already deployed)

apiVersion: gateway.networking.k8s.io/v1 kind: Gateway metadata: name: varnish-gateway namespace: default spec: gatewayClassName: varnish listeners: - name: http protocol: HTTP port: 80

---

Stable backend deployment (version 1.0)

apiVersion: apps/v1 kind: Deployment metadata: name: app-stable namespace: default spec: replicas: 2 selector: matchLabels: app: myapp version: stable template: metadata: labels: app: myapp version: stable spec: containers: - name: app image: hashicorp/http-echo args: - "-text=stable-v1.0" - "-listen=:8080" ports: - containerPort: 8080

---

Canary backend deployment (version 2.0)

apiVersion: apps/v1 kind: Deployment metadata: name: app-canary namespace: default spec: replicas: 2 selector: matchLabels: app: myapp version: canary template: metadata: labels: app: myapp version: canary spec: containers: - name: app image: hashicorp/http-echo args: - "-text=canary-v2.0" - "-listen=:8080" ports: - containerPort: 8080

---

Stable service

apiVersion: v1 kind: Service metadata: name: app-stable namespace: default spec: selector: app: myapp version: stable ports: - port: 8080 targetPort: 8080

---

Canary service

apiVersion: v1 kind: Service metadata: name: app-canary namespace: default spec: selector: app: myapp version: canary ports: - port: 8080 targetPort: 8080

---

HTTPRoute with weighted traffic split (90% stable, 10% canary)

apiVersion: gateway.networking.k8s.io/v1 kind: HTTPRoute metadata: name: canary-route namespace: default spec: parentRefs: - name: varnish-gateway namespace: default hostnames: - "app.example.com" rules: - backendRefs: - name: app-stable port: 8080 weight: 90 - name: app-canary port: 8080 weight: 10 ```

Save as canary.yaml and apply with kubectl apply -f canary.yaml. Verify the split:

```bash

Get the gateway service IP

kubectl get svc -n default varnish-gateway

Make 100 requests and count responses

for i in {1..100}; do curl -H "Host: app.example.com" http:///; done | sort | uniq -c

Expect approximately 90 stable-v1.0 responses and 10 canary-v2.0 responses

```