This repository contains a Kubernetes operator which manages the lifecycle of a Calico or Calico Enterprise installation on Kubernetes or OpenShift. Its goal is to make installation, upgrades, and ongoing lifecycle management of Calico and Calico Enterprise as simple and reliable as possible.
This operator is built using the operator-sdk, so you should be familiar with how that works before getting started.
There are a few important areas to be aware of:
- Operator API definitions exist in
pkg/apis/operator/v1
- Rendering code for generating Kubernetes resources is in
pkg/render
- Control/reconcile loops for each component can be found in
pkg/controller/<component>
- Status reporting is in
pkg/controller/status
Tests:
- Tests for file
X.go
can be found inX_test.go
. - FV tests which run against a local cluster can be found in
test/*.go
.
When developing in the operator, there are a few design principles to be aware of.
- API changes should be rare occurences, and the API should contain as little as possible. Use auto-detection or automation wherever possible to reduce the API surface.
- Each "component" should receive its own CRD, namespace, controller, and status manager. e.g., compliance, networking, apiserver.
- Controllers interact with each other through the Kubernetes API. For example, by updating status on relevant objects.
New APIs are added using the operator-sdk
tool.
operator-sdk add api --api-version=operator.tigera.io/v1 --kind=<Kind>
When modifying or adding CRDs, you will need to run make gen-files
to update the auto-generated files. The tool
might change the scope of existing resources to "Namespaced", so make sure to set them back to their desired state.
New controllers are also added using the operator-sdk
tool.
operator-sdk add controller --api-version=operator.tigera.io/v1 --kind=<Kind>
You will need to modify the auto-generated controller's Add
function to accept additional arguments
in order to match the AddToManagerFuncs defined in pkg/controller/controller.go.
You can create a local k3d cluster with the Makefile:
make cluster-create
Then, run the operator against the local cluster:
KUBECONFIG=./kubeconfig.yaml go run ./cmd/manager
To launch Calico, install the default custom resource:
kubectl create -f ./deploy/crds/operator_v1_installation_cr.yaml
To tear down the cluster:
make cluster-destroy
To install Calico Enterprise instead of Calico, you need to install an image pull secret, as well as modify the Installation CR.
Create the pull secret in the tigera-operator namespace:
kubectl create secret -n tigera-operator generic tigera-pull-secret \
--from-file=.dockerconfigjson=<PATH/TO/PULL/SECRET> \
--type=kubernetes.io/dockerconfigjson
Then, modify the installation CR (e.g., with kubectl edit installations
) to include the following:
spec:
variant: TigeraSecureEnterprise
imagePullSecrets:
- name: tigera-pull-secret
You can then install additional Calico Enterprise components by creating their CRs from within
the ./deploy/crds/
directory.
To run all the unit tests, run:
make test
To run a specific test or set of tests, use the GINKGO_FOCUS
argument.
make test GINKGO_FOCUS="component function tests"
The operator creates and manages resources and will reconcile them to be in the desired state. Due to the reconciliation it does, if a user makes direct changes to a resource the operator will revert those changes. To enable the user to make temporary changes, an annotation can be added to any resource directly managed by the operator which will cause the operator to no longer update the resource. Adding the following as an annotation to any resource will prevent the operator from making any future updates to the annotated resource:
Do not use this unless you are a developer working on the operator. If you add this annotation, you must remove it before the operator can manage the resource again.
unsupported.operator.tigera.io/ignore: "true"
Notice that the annotation is added in the top level metadata (not in the spec.template.metadata). (note the below is not a valid manifest but just an example)
kind: DaemonSet
apiVersion: apps/v1
metadata:
name: calico-node
namespace: calico-system
labels:
k8s-app: calico-node
annotations:
# You should NOT use this unless you want to block the operator from doing its job managing this resource.
unsupported.operator.tigera.io/ignore: "true"
spec:
template:
metadata:
labels:
k8s-app: calico-node
annotations:
scheduler.alpha.kubernetes.io/critical-pod: ''
spec:
containers:
- name: calico-node
image: calico/node:my-special-tag