Skip to content

Canary release of a KServe ML model

This tutorial shows how Iter8 can be used to release ML models hosted in a KServe environment using a canary release strategy.

In a canary release, inference requests that match a particular pattern, for example those that have a particular header, are directed to the candidate version of the model. The remaining requests go to the primary, or initial, version of the model.

Canary release

The user declaratively describes the desired application state at any given moment. An Iter8 release chart assists users who describe the application state at any given moment. The chart provides the configuration needed for Iter8 to automatically deploy application versions and configure the routing to implement the canary release strategy.


Before you begin
  1. Ensure that you have the kubectl and helm CLIs installed.
  2. Have access to a cluster running KServe. You can create a KServe Quickstart environment as follows:
    curl -s "https://raw.githubusercontent.com/kserve/kserve/release-0.11/hack/quick_install.sh" | bash
    
    If using a local cluster (for example, Kind or Minikube), we recommend providing the cluster with at least 16GB of memory.

Install the Iter8 controller

Iter8 can be installed and configured to watch resources either in a single namespace (namespace-scoped) or in the whole cluster (cluster-scoped).

helm install --repo https://iter8-tools.github.io/iter8 --version 1.1 iter8 controller
helm install --repo https://iter8-tools.github.io/iter8 --version 1.1 iter8 controller \
--set clusterScoped=true

For additional install options, see Iter8 Installation.

Deploy initial version

Deploy the initial version of the model using the Iter8 release chart by identifying the environment into which it should be deployed, a list of the versions to be deployed (only one here), and the release strategy to be used:

cat <<EOF | helm upgrade --install wisdom --repo https://iter8-tools.github.io/iter8 release --version 1.1 -f -
environment: kserve
application: 
  metadata:
    labels:
      app.kubernetes.io/name: wisdom
  modelFormat: sklearn
  runtime: kserve-mlserver
  versions:
  - metadata:
      labels:
        app.kubernetes.io/version: v0
    storageUri: "gs://seldon-models/sklearn/mms/lr_model"
  strategy: canary
EOF

Wait for the backend model to be ready:

kubectl wait --for condition=ready isvc/wisdom-0 --timeout=600s
What happens?
  • Because environment is set to kserve, an InferenceService object is created.
  • The namespace default is inherited from the Helm release namespace since it is not specified in the version or in application.metadata.
  • The name wisdom-0 is derived from the Helm release name since it is not specified in the version or in application.metadata. The name is derived by appending the index of the version in the list of versions; -0 in this case.
  • Alternatively, an inferenceServiceSpecification could have been provided.

To support routing, a Service (of type ExternalName) named default/wisdom pointing at the KNative gateway, knative-local-gateway.istio-system, is deployed. The name is the Helm release name since it not specified in application.metadata. Further, an Iter8 routemap is created.

Once the InferenceService is ready, the Iter8 controller automatically configures the routing by creating an Istio VirtualService. It is configured to route all inference requests to the only deployed version, wisdom-0.

Verify routing

You can send verify the routing configuration by inspecting the VirtualService:

kubectl get virtualservice wisdom -o yaml

You can also send inference requests from a pod within the cluster:

  1. Create a sleep pod in the cluster from which requests can be made:

    curl -s https://raw.githubusercontent.com/iter8-tools/docs/v0.18.4/samples/kserve-serving/sleep.sh | sh -
    

  2. Exec into the sleep pod:

    kubectl exec --stdin --tty "$(kubectl get pod --sort-by={metadata.creationTimestamp} -l app=sleep -o jsonpath={.items..metadata.name} | rev | cut -d' ' -f 1 | rev)" -c sleep -- /bin/sh
    

  3. To send requests without the header traffic:

    curl -H 'Content-Type: application/json' \
    http://wisdom.default -d @input.json -s -D - \
    | grep -e HTTP -e app-version
    

  4. Requests can also be sent with the header traffic: test. When a candidate is deployed, requests with this header will be routed to the candidate. When no candidate is deployed, all requests will be routed to the primary version.

    curl -H 'Content-Type: application/json' \
    -H 'traffic: test' \
    http://wisdom.default -d @input.json -s -D - \
    | grep -e HTTP -e app-version
    

The output includes the success of the request (the HTTP return code) and the version of the application that responded (in the app-version response header). In this example:

HTTP/1.1 200 OK
app-version: wisdom-0
To send requests from outside the cluster

To configure the release for traffic from outside the cluster:

(a) In a separate terminal, port-forward the Istio ingress gateway:

kubectl -n istio-system port-forward svc/istio-ingressgateway 8080:80
(b) Download the sample input:
curl -sO https://raw.githubusercontent.com/iter8-tools/docs/v0.17.3/samples/kserve-serving/input.json
(c) To send requests without the header traffic:
curl -H 'Content-Type: application/json' \
-H 'Host: wisdom.default' \
localhost:8080 -d @input.json -s -D - \
| grep -e '^HTTP' -e app-version
(d) To send requests with the header traffic: test:
curl -H 'Content-Type: application/json' \
-H 'Host: wisdom.default' \
-H 'traffic: test' \
localhost:8080 -d @input.json -s -D - \
| grep -e HTTP -e app-version

Deploy candidate

A candidate version of the model can be deployed simply by adding a second version to the list of versions comprising the application:

cat <<EOF | helm upgrade --install wisdom --repo https://iter8-tools.github.io/iter8 release --version 1.1 -f -
environment: kserve
application: 
  metadata:
    labels:
      app.kubernetes.io/name: wisdom
  modelFormat: sklearn
  runtime: kserve-mlserver
  versions:
  - metadata:
      labels:
        app.kubernetes.io/version: v0
    storageUri: "gs://seldon-models/sklearn/mms/lr_model"
  - metadata:
      labels:
        app.kubernetes.io/version: v1
    storageUri: "gs://seldon-models/sklearn/mms/lr_model"
  strategy: canary
EOF
About the candidate

In this tutorial, the model source (field storageUri) for the candidate version is the same as for the primary version of the model. In a real example, this would be different. The version label (app.kubernetes.io/version) can be used to distinguish between versions.

When the candidate version is ready, the Iter8 controller will Iter8 will automatically reconfigure the routing so that inference requests with the header traffic set to test will be sent to the candidate model:

HTTP/1.1 200 OK
app-version: wisdom-1

All other requests will be sent to the primary model (wisdom-0):

HTTP/1.1 200 OK
app-version: wisdom-0

Verify routing

You can verify the routing configuration by inspecting the VirtualService and/or by sending requests as described above. Those with header traffic set to test will be handled by the candidate model (wisdom-1) while all others will be handled by the primary version.

Promote candidate

The candidate model can be promoted by redefining the primary version and removing the candidate:

cat <<EOF | helm upgrade --install wisdom --repo https://iter8-tools.github.io/iter8 release --version 1.1 -f -
environment: kserve
application: 
  metadata:
    labels:
      app.kubernetes.io/name: wisdom
  modelFormat: sklearn
  runtime: kserve-mlserver
  versions:
  - metadata:
      labels:
        app.kubernetes.io/version: v1
    storageUri: "gs://seldon-models/sklearn/mms/lr_model"
  strategy: canary
EOF
What is different?

The version label (app.kubernetes.io/version) of the primary version was updated. In a real world example, storageUri would also be updated (with that from the candidate version).

Once the (reconfigured) primary InferenceService ready, the Iter8 controller will automatically reconfigure the routing to send all requests to it.

Verify Routing

You can verify the routing configuration by inspecting the VirtualService and/or by sending requests as described above. They will be handled by the primary version. Output will be something like:

HTTP/1.1 200 OK
app-version: wisdom-0

Cleanup

Delete the models and their routing:

helm delete wisdom

If you used the sleep pod to generate load, remove it:

kubectl delete deploy sleep

Uninstall Iter8 controller:

helm delete iter8

For additional uninstall options, see Iter8 Uninstall.