Krane

Krane is a canary deployment manager for Kubernetes, automatically creating and operating with newly created canary deployments and reporting back to the status objects.

Prerequisites

Krane was tested on the following configuration:

• Kubernetes 1.13.5
• Istio 1.1.4 (contained in the /infra folder)
  • Grafana and Kiali allowed to see visualized results

Installation

Installation consists of multiple steps:

• Build the image using docker build -f ./build/Dockerfile -t <your-name>/<your-repo> ..
• Push the image to the remote registry by running docker push <your-name>/<your-repo>. Note that you may need to use credentials companions to authenticate when using cloud solutions.
• Add your existing image into the ./deploy/operator.yaml to override the default operator (by replacing gcr.io/petomalina/krane with <your-name>/<your-repo>).
• Installing CRDs via kubectl apply -f ./deploy/crds will create Custom Resource Definitions for Canary and CanaryPolicy objects inside the Kubernetes cluster.
• Installing the operator via direct Kubernetes template configuration using kubectl apply -f ./deploy, which will deploy the Operator workload as well as its Role, RoleBinding and ServiceAccount.

Building and Running locally

Performing these steps will ensure the projects can be run locally while being connected to a remote cluster:

• Make sure the kubectl is connected to your target cluster. This approach differs based on the cloud provider or when using on-prem solution (e.g. use gcloud containers in case you are running on GCP with GKE).
• Install dep as a package manager. Note that vgo is not supported in operator-sdk at the moment and using it may result in unexpected results (errors by downloading unsupported or unlocked dependencies that vgo can't resolve).
• Run go run ./cmd/manager/main.go. In case you don't want to use the default namespace, use WATCH_NAMESPACE environment variable to override this value.

Running example

The example uses a single CanaryPolicy object and fully configures the Istio service mesh networking using Gateway and VirtualService. You can override any policy to test the features by editing ./example/00-canary-policy.yaml. Note that the policy will only change for new canaries and never affects the existing.

Two deployments are part of the example to demonstrate usage on a system:

• Aggregator is used as an entrypoint to the system, accepting raw metrics and aggregating them.
• Storage is used for temporary storage of the metrics in memory.

Applying Istio infrastructure after the Canary Policy is done by running:

kubectl apply -f ./01-gw.yaml
kubectl apply -f ./01-vs.yaml

Deploying the application is then done by running:

kubectl apply -f ./02-aggregator-deployment.yaml
kubectl apply -f ./02-storage-deployment.yaml

After all deployments are reporting as stable, run the canary analysis test by applying:

kubectl apply -f ./10-aggregator-canary.yaml

This will run the canary deployment for the Aggregator service, and status can be watched using:

kubectl get canary -n <namespace> my-app-aggregator-deployment-ah87p9-canary -o yaml -w

The -w at the end of the command will provide a stream of updates to the canary. Everything can also be observed from Kiali, Grafana and Prometheus.