Machine API Operator

The Machine API Operator manages the lifecycle of specific purpose CRDs, controllers and RBAC objects that extend the Kubernetes API. This allows to convey desired state of machines in a cluster in a declarative fashion.

Architecture

CRDs

• MachineSet
• Machine
• Cluster
• MachineHealthCheck

Controllers

Cluster API controllers

• MachineSet Controller

  • Reconciles desired state for MachineSets by ensuring presence of specified number of replicas and config for a set of machines.

• Machine Controller

  • Reconciles desired state for Machines by ensuring that instances with a desired config exist in a given cloud provider. Currently we support:

  • cluster-api-provider-aws

  • cluster-api-provider-libvirt

  • cluster-api-provider-openstack. Coming soon.

• Node Controller

  • Reconciles desired state of machines by matching IP addresses of machine objects with IP addresses of node objects. Annotating node with a special label containing machine name that the cluster-api node controller interprets and sets corresponding nodeRef field of each relevant machine.

Nodelink controller

• Reconciles desired state of machines by matching IP addresses of machine objects with IP addresses of node objects and annotates nodes with a special machine annotation containing the machine name. The cluster-api node controller interprets the annotation and sets the corresponding nodeRef field of each relevant machine.

• Build:

$ make nodelink-controller

Machine healthcheck controller

• Reconciles desired state for MachineHealthChecks by ensuring that machines targeted by machineHealthCheck objects are healthy or remediated otherwise.

• Build:

  $ make machine-healthcheck
  

• How to test it:

  1. Create a machineset and locate its selector. Assuming the selector corresponds to the following list of match labels:

     machine.openshift.io/cluster-api-cluster: cluster
     machine.openshift.io/cluster-api-machine-role: worker
     machine.openshift.io/cluster-api-machine-type: worker
     machine.openshift.io/cluster-api-machineset: cluster-worker-us-east-1a
     

  2. Define a MachineHealthCheck manifest that will be watching all machines of the machinset based on its match labels:

     apiVersion: healthchecking.openshift.io/v1alpha1
     kind: MachineHealthCheck
     metadata:
       name: example
       namespace: default
     spec:
       selector:
         matchLabels:
           machine.openshift.io/cluster-api-cluster: cluster
           machine.openshift.io/cluster-api-machine-role: worker
           machine.openshift.io/cluster-api-machine-type: worker
           machine.openshift.io/cluster-api-machineset: cluster-worker-us-east-1a

  3. Pick a node that is managed by one of the machineset's machines

  4. SSH into the node, disable and stop the kubelet services:

     # systemctl disable kubelet
     # systemctl stop kubelet
     

  5. After some time the node will transition into NotReady state

  6. Watch the machine-healthcheck controller logs to see how it notices a node in NotReady state and starts to reconcile the node

  7. After some time the current node instance is terminated and new instance is created. Followed by new node joining the cluster and turning in Ready state.

Dev

• Build:

  $ make build

• Run:

  $ ./bin/machine-api-operator start --kubeconfig ${HOME}/.kube/config --images-json=pkg/operator/fixtures/images.json

• Image:

  $ make image
  

The Machine API Operator is designed to work in conjunction with the Cluster Version Operator. You can see it in action by running an OpenShift Cluster deployed by the Installer.

However you can run it in a vanilla Kubernetes cluster by precreating some assets:

• Create a openshift-machine-api-operator namespace
• Create a CRD Status definition
• Create a CRD Machine definition
• Create a CRD MachineSet definition
• Create a CRD MachineDeployment definition
• Create a CRD Cluster definition
• Create a Installer config
• Then you can run it as a deployment
• You should then be able to deploy a cluster and a machineSet object

Machine API operator with Kubemark over Kubernetes

INFO: For development and testing purposes only

1. Deploy MAO over Kubernetes:

 $ kustomize build | kubectl apply -f -

2. Deploy Kubemark actuator prerequisities:

   $ kustomize build config | kubectl apply -f -

3. Create cluster-config-v1 configmap to tell the MAO to deploy kubemark provider:

   apiVersion: v1
   kind: ConfigMap
   metadata:
     name: cluster-config-v1
     namespace: kube-system
   data:
     install-config: |-
       platform:
         kubemark: {}

   The file is already present under config/kubemark-install-config.yaml so it's sufficient to run:

   $ kubectl apply -f config/kubemark-install-config.yaml

CI & tests

Run unit test:

$ make test

Run e2e-aws-operator tests. This tests assume that a cluster deployed by the Installer is up and running and a KUBECONFIG environment variable is set:

$ make test-e2e

Tests are located under machine-api-operator repository and executed in prow CI system. A link to failing tests is published as a comment in PR by @openshift-ci-robot. Current test status for all OpenShift components can be found in https://deck-ci.svc.ci.openshift.org.

CI configuration is stored under openshift/release repository and is split into 4 files:

• cluster/ci/config/prow/plugins.yaml - says which prow plugins are available and where job config is stored
• ci-operator/config/openshift/machine-api-operator/master.yaml - configuration for machine-api-operator component repository
• ci-operator/jobs/openshift/machine-api-operator/openshift-machine-api-operator-master-presubmits.yaml - prow jobs configuration for presubmits
• ci-operator/jobs/openshift/machine-api-operator/openshift-machine-api-operator-master-postsubmits.yaml - prow jobs configuration for postsubmits

More information about those files can be found in ci-operator onboarding file.