This is a preview release. Do not use it in production environments.

Download the release bundle from Download Edge-Native Defender (Preview). New to Edge-Native Defender? Start with Edge-Native Defender overview.

Running Envoy yourself, outside Kubernetes? See Defender for Envoy.

Watch the install on Tetrate Enterprise Gateway (3½ minutes): downloading the release, the Secret, helm install, pointing the Gateway at Defender, allowed and blocked requests, and the gateway Envoy's own counters. Istio and kgateway follow the same steps with their own setting in step 3.

At a glance

Deployment architecture

Defender runs on the Kubernetes gateways built on Envoy, including those from Tetrate and Solo.io. It joins the Gateway you choose, in every one of its pods, so it grows and shrinks with the gateway's own replicas.

  BEFORE

    client ---> gateway pod [ Envoy ] ------------------------> your backends


  AFTER: Envoy Gateway and Istio

    client ---> gateway pod [ Envoy + Defender filter ] ------> your backends
                                     |
                                     +---> your Cequence tenant: policies in,
                                           analytics out, in the background

  AFTER: kgateway and Gloo Gateway

    client ---> gateway pod [ Envoy ] ------------------------> your backends
                              |  localhost
                              v
                            [ defender-extproc ]
                              |
                              +---> your Cequence tenant: policies in,
                                    analytics out, in the background

What stays the same. Your routes, backends, listeners and other policies. Defender is added through the Gateway's own extension resources and one parametersRef on the Gateway; remove that reference to undo it.

What Defender does with a request. It decides each request from a policy set already held in the gateway pod, so no request waits on Cequence. A blocked request is answered by the gateway and never reaches your backend. An allowed request is forwarded byte for byte, and your client receives the response byte for byte: Defender inspects a copy of each body, up to your tenant's capture settings, and never changes what it forwards.

What is sent to Cequence. For each request: its headers and URL, the response status and headers, and the inspected body copies, masked first with your tenant's Sensitive Data rules. Your traffic filters decide whether a request is reported at all. Your client secret is read from a Kubernetes Secret and never appears in Defender's configuration, the gateway's Deployment or its logs.

Installation

Prerequisites

• One of these gateways, at the version Defender is tested with:
• Envoy Gateway 1.9 (Tetrate Enterprise Gateway).
• Istio 1.31, with Gateway API gateways (Gloo Mesh, Tetrate Service Bridge).
• kgateway 2.4 (Gloo Gateway).
• Helm 3, if you install with Helm. No chart repository is needed: the chart is in the bundle, at charts/defender-for-envoy.
• Outbound HTTPS from the gateway pods to edge.<tenant>.cequence.cloud and auth.<tenant>.cequence.cloud. On Envoy Gateway and Istio, also to downloads.cequence.ai, where the gateway pods download the filter at start and check it against its sha256.
• Your tenant's name (such as acme.1), client ID and client secret.
• On kgateway, credentials from Cequence to pull the defender-extproc image from Cequence's registry.

The bundle's defender.conf.example names the two release-specific values:

• ENVOY_WASM_PATH: the filter's download URL, for Envoy Gateway and Istio. Its file name is the filter's sha256, so the URL ends in /<sha256>.wasm.
• ENVOY_EXTPROC_IMAGE: the defender-extproc image, for kgateway.

1. Create the Secret

In the namespace where the gateway pods run: for Envoy Gateway, by default envoy-gateway-system; for Istio and kgateway, the Gateway's own namespace.

kubectl -n <namespace> create secret generic defender-client --from-literal=client-secret="$CEQUENCE_CLIENT_SECRET"

If your Envoy Gateway runs gateway pods in the Gateway's namespace, create the Secret there, and set proxyNamespace (Helm) or --proxy-namespace (render-k8s) to it.

2. Install Defender

With Helm. From the bundle folder, in the Gateway's namespace:

helm install defender-for-envoy ./charts/defender-for-envoy -n <Gateway's namespace> \
  --set platform=<envoy-gateway, istio or kgateway> \
  --set gateway.name=<gateway name> \
  --set tenant.slug=<tenant name, such as acme.1> \
  --set tenant.clientId=<client id> \
  --set filter.url=<ENVOY_WASM_PATH> --set filter.sha256=<the sha256 in that URL>

The chart derives your tenant's hosts from tenant.slug: for acme.1, https://edge.acme.1.cequence.cloud and https://auth.acme.1.cequence.cloud/auth/realms/cequence/protocol/openid-connect/token. Set tenant.edgeUrl and tenant.authUrl only if Cequence gives you other addresses.

On kgateway, set extproc.image=<ENVOY_EXTPROC_IMAGE> instead of the two filter settings, and name your image pull secret with --set "extproc.imagePullSecrets[0].name=<pull secret>".

If Cequence has given you access to its chart repository, you can install from there instead: helm repo add cequence <chart repository URL> --username <token user> --password <token>, then use cequence/defender-for-envoy in place of ./charts/defender-for-envoy.

Without Helm. Copy the bundle's defender.conf.example to defender.conf, fill in CEQUENCE_TENANT and CEQUENCE_CLIENT_ID, then write the manifests and apply them. No client secret is needed here; the gateway pods read it from the Secret.

./bin/defenderctl-envoy render-k8s --platform <envoy-gateway, istio or kgateway> --gateway <namespace>/<gateway name>
kubectl apply -f defender-<platform>/

On kgateway, add your image pull secret to the gateway pods (podTemplate.imagePullSecrets in the GatewayParameters it writes).

Both ways produce the same resources.

3. Point your Gateway at Defender

Both ways of installing print the setting to add to your Gateway:

The platform restarts the gateway pods with Defender added. If your Gateway already names parameters of its own, merge Defender's into them instead; the bundle's KUBERNETES.md says what each one adds.

4. Check it before you go live

1. Install with diagnostics on (--set diagnostics=true, or CEQUENCE_DIAGNOSTICS=yes in defender.conf) on a test Gateway, and request /__end/status through it. policies_loaded shows your tenant's policy count.
2. Send a request that one of your policies blocks, and check it is blocked. On Envoy Gateway, kubectl -n envoy-gateway-system rollout status deploy -l gateway.envoyproxy.io/owning-gateway-name=<gateway name> says when the gateway pods have restarted with Defender, and Defender's EnvoyExtensionPolicy reports Accepted: kubectl -n <Gateway's namespace> get envoyextensionpolicy defender -o jsonpath='{.status.ancestors[0].conditions[?(@.type=="Accepted")].status}'. The gateway Envoy's own counters show both kinds of request: port-forward a gateway pod's admin port (kubectl -n envoy-gateway-system port-forward pod/<pod> 19000:19000) and open http://127.0.0.1:19000/stats?filter=downstream_rq_[24]xx. Blocks count as 4xx on the Gateway's listener.
3. Turn diagnostics off again (helm upgrade ... --reuse-values --set diagnostics=false, or render and apply again without it). The status page is readable by anyone who can reach the Gateway.

Upgrades and rollback

Each release has its own filter URL and defender-extproc image, in the new bundle's defender.conf.example. With Helm, keep your other settings and change only those:

helm upgrade defender-for-envoy ./charts/defender-for-envoy -n <Gateway's namespace> --reuse-values \
  --set filter.url=<new ENVOY_WASM_PATH> --set filter.sha256=<the sha256 in that URL>

On kgateway, use --set extproc.image=<new ENVOY_EXTPROC_IMAGE> instead.

To roll back, helm history defender-for-envoy -n <Gateway's namespace> lists the revisions, and helm rollback defender-for-envoy <revision> -n <Gateway's namespace> returns to one.

Without Helm, set ENVOY_WASM_PATH (or ENVOY_EXTPROC_IMAGE) in defender.conf from the new bundle, run the new bundle's render-k8s and apply. To roll back, do the same with the previous values.

During the upgrade, the platform replaces the gateway pods, as it does for any change to the Gateway's parameters. Run two or more gateway replicas so the Gateway keeps serving while they roll. Each new pod forwards requests until it has loaded your policies, and analytics not yet sent from an old pod are lost. Policy changes never need an upgrade: Defender picks them up while it runs.

Performance and sizing

Measured on a 4 vCPU host at 1,000 requests a second, with a policy set of about 6,000 policies, analytics and masking on, and 1 KB responses; the Envoy Gateway and Istio column is the same filter measured in Envoy. Cost scales with request rate; more policies, larger bodies or more Sensitive Data rules raise it.

On kgateway the Defender container requests 250m CPU and 256 MiB by default (extproc.resources in the chart). Do not limit its CPU below what your traffic needs: a throttled Defender delays every request.

Cost

Cequence charges nothing for the infrastructure. Defender runs in your existing gateway pods, using the CPU and memory in "Performance and sizing" above; there is no separate service, load balancer or licence to add on your side.

Availability

• If Cequence is slow or unreachable, your traffic is unaffected. Defender keeps enforcing the policies it last received, and checks for new ones every 10 seconds. It holds a bounded backlog of analytics and sends it once Cequence is reachable again; a long outage can drop some.
• If Defender itself cannot run, the gateway forwards requests without it. On Envoy Gateway and Istio, the filter is installed fail-open: if it cannot load, the gateway's Envoy skips it. On kgateway, if the Defender container does not answer within 1 second, the request is forwarded uninspected.
• If you need Defender to stop acting at once, change the policies in your Cequence tenant to a non-blocking action; gateway pods pick it up within about 10 seconds and keep reporting. To take Defender off the path entirely, remove the parametersRef from your Gateway; the pods restart without it.
• A fresh install blocks nothing until your tenant's first policy publish arrives.

Known limits

• block and add-header policies are enforced. challenge, rate-limit and redirect policies are reported but not applied: the request is forwarded.
• On Envoy Gateway and Istio, a policy that matches on the request body can block the request but cannot add a header to it. On kgateway it can do both.
• On kgateway, request bodies reach your backend with Transfer-Encoding: chunked and no Content-Length; HTTP/1.1 backends must accept chunked bodies.
• Envoy Gateway allows one EnvoyExtensionPolicy per Gateway. If yours has one, Defender's filter must be added to it; KUBERNETES.md in the bundle says how.
• Istio's classic ingress gateway and sidecars are not covered yet.
• Defender sees the request as the gateway has normalised it. Envoy Gateway merges duplicate slashes in the path by default, so a policy written for // does not match; set path.disableMergeSlashes: true in a ClientTrafficPolicy if you need the slashes kept.
• Changing Defender's settings or version restarts the gateway pods; policy changes do not.
• Defender is tested on the open-source Envoy Gateway, Istio and kgateway these products are built on.

If something is wrong

Requests keep flowing in every case below: Defender forwards when it cannot decide.

defenderctl-envoy status --platform <platform> --gateway <namespace>/<gateway name> lists Defender's resources on the Gateway. If you contact support, send the output of /__end/status, which contains no credentials. Full troubleshooting and every setting are in the bundle's KUBERNETES.md.

Removing it

1. Remove the parametersRef from your Gateway, or point it back at what it named before. The pods restart without Defender.
2. With Helm, helm uninstall defender-for-envoy -n <Gateway's namespace>. Without Helm, defenderctl-envoy render-k8s --delete --platform <platform> --gateway <namespace>/<gateway name> prints the kubectl delete commands for what it added; it runs nothing itself.
3. Delete the defender-client Secret, and ask Cequence to disable the client credential or rotate its secret.

Your routes and backends were never changed, so nothing else needs undoing.