node-maintenance-operator

Node Maintenance Operator

node-maintenance-operator is an operator generated from the operator-sdk. The purpose of this operator is to watch for new or deleted custom resources called NodeMaintenance which indicate that a node in the cluster should either:

• NodeMaintenance CR created: move node into maintenance, cordon the node - set it as unschedulable and evict the pods (which can be evicted) from that node.
• NodeMaintenance CR deleted: remove pod from maintenance and uncordon the node - set it as schedulable

Note: The current behavior of the operator is to mimic kubectl drain <node name> as performed in Kubevirt - evict all VMs and Pods on a node

Build and run the operator

Before running the operator, the NodeMaintenance CRD and namespace must be registered with the Openshift/Kubernetes apiserver:

$ kubectl create -f deploy/crds/nodemaintenance_crd.yaml
$ kubectl create -f deploy/namespace.yaml

Once this is done, there are two ways to run the operator:

• As a Deployment inside a Openshift/Kubernetes cluster
• As Go program outside a cluster

Deploy operator using OLM

For more information on the Operator Lifecycle Manager and Cluster Service Versions checkout out "Building a Cluster Service Version".

1. Build and push operator and operator-registry image.

./build/make-manifests.sh <VERSION>
make container-build
make container-push

1. Create the node-maintenance-operator Namespace.

oc create -f olm-deploy-manifests/nm-ns.yaml

cat <<EOF | oc create -f -
apiVersion: v1
kind: Namespace
metadata:
  annotations:
  labels:
    kubevirt.io: ""
  name: node-maintenance-operator
EOF

1. Create the operator group.

oc create -f olm-deploy-manifests/nm-op-group.yaml

cat <<EOF | oc create -f -
apiVersion: operators.coreos.com/v1alpha2
kind: OperatorGroup
metadata:
  name: node-maintenance-operator
  namespace: node-maintenance-operator
EOF

1. Using the node-maintenance-operator-registry container image built in step 1, create a CatalogSource. This object tells OLM about the operator.

oc create -f olm-deploy-manifests/nm-catalog-source.yaml

cat <<EOF | oc create -f -
apiVersion: operators.coreos.com/v1alpha1
kind: CatalogSource
metadata:
  name: node-maintenance-operator
  namespace: openshift-operator-lifecycle-manager
spec:
  sourceType: grpc
  image: quay.io/kubevirt/node-maintenance-operator-registry:<VERSION>
  displayName: node-maintenance-operator
  publisher: Red hat
EOF

1. Subscribe to the node-maintenance-operator.

oc create -f olm-deploy-manifests/nm-sub.yaml

cat <<EOF | oc create -f -
apiVersion: operators.coreos.com/v1alpha1
kind: Subscription
metadata:
  name: node-maintenance-operator-subscription
  namespace: node-maintenance-operator
spec:
  channel: beta
  name: node-maintenance-operator
  source: node-maintenance-operator
  sourceNamespace: openshift-operator-lifecycle-manager
  startingCSV: node-maintenance-operator.<VERSION>
EOF

1. Run as a Deployment inside the cluster

The Deployment manifest is generated at deploy/operator.yaml. Be sure to update the deployment image if there are changes as shown here.

Setup RBAC and deploy the node-maintenance-operator:

$ kubectl create -f deploy/service_account.yaml
$ kubectl create -f deploy/role.yaml
$ kubectl create -f deploy/role_binding.yaml
$ kubectl create -f deploy/operator.yaml

Verify that the node-maintenance-operator is up and running:

$ kubectl get deployment
NAME                     DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE
node-maintenance-operator      1         1         1            1           1m

2. Run locally outside the cluster

This method is preferred during development cycle to deploy and test faster.

Set the name of the operator in an environment variable:

export OPERATOR_NAME=node-maintenance-operator

Run the operator locally with the default Kubernetes config file present at $HOME/.kube/config or with specificing kubeconfig via the flag --kubeconfig=<path/to/kubeconfig>:

$ operator-sdk up local --kubeconfig="<path/to/kubeconfig>"

INFO[0000] Running the operator locally.
INFO[0000] Using namespace default.
{"level":"info","ts":1551793839.3308277,"logger":"cmd","msg":"Go Version: go1.11.4"}
{"level":"info","ts":1551793839.3308823,"logger":"cmd","msg":"Go OS/Arch: linux/amd64"}
{"level":"info","ts":1551793839.330899,"logger":"cmd","msg":"Version of operator-sdk: v0.5.0+git"}
...

Setting Node Maintenance

Set Maintenance on - Create a NodeMaintenance CR

To set maintenance on a node a NodeMaintenance CR should be created. A NodeMaintenance CR contains:

• Name: The name of the node which will be put into maintenance
• Reason: the reason for the node maintenance

Create the example NodeMaintenance CR found at deploy/crds/nodemaintenance_cr.yaml:

$ cat deploy/crds/nodemaintenance_cr.yaml

apiVersion: kubevirt.io/v1alpha1
kind: NodeMaintenance
metadata:
  name: nodemaintenance-xyz
spec:
  nodeName: node02
  reason: "Test node maintenance"

$ kubectl apply -f deploy/crds/nodemaintenance_cr.yaml
{"level":"info","ts":1551794418.6742408,"logger":"controller_nodemaintenance","msg":"Reconciling NodeMaintenance","Request.Namespace":"default","Request.Name":"node02"}
{"level":"info","ts":1551794418.674294,"logger":"controller_nodemaintenance","msg":"Applying Maintenance mode on Node: node02 with Reason: Test node maintenance","Request.Namespace":"default","Request.Name":"node02"}
{"level":"info","ts":1551783365.7430992,"logger":"controller_nodemaintenance","msg":"WARNING: ignoring DaemonSet-managed Pods: default/local-volume-provisioner-5xft8, kubevirt/disks-images-provider-bxpc5, kubevirt/virt-handler-52kpr, openshift-monitoring/node-exporter-4c9jt, openshift-node/sync-8w5x8, openshift-sdn/ovs-kvz9w, openshift-sdn/sdn-qnjdz\n"}
{"level":"info","ts":1551783365.7471824,"logger":"controller_nodemaintenance","msg":"evicting pod \"virt-operator-5559b7d86f-2wsnz\"\n"}
{"level":"info","ts":1551783365.7472217,"logger":"controller_nodemaintenance","msg":"evicting pod \"cdi-operator-55b47b74b5-9v25c\"\n"}
{"level":"info","ts":1551783365.747241,"logger":"controller_nodemaintenance","msg":"evicting pod \"virt-api-7fcd86776d-652tv\"\n"}
{"level":"info","ts":1551783365.747243,"logger":"controller_nodemaintenance","msg":"evicting pod \"simple-deployment-1-m5qv9\"\n"}
{"level":"info","ts":1551783365.7472336,"logger":"controller_nodemaintenance","msg":"evicting pod \"virt-controller-8987cffb8-29w26\"\n"}
...

Set Maintenance off - Delete the NodeMaintenance CR

To remove maintenance from a node a NodeMaintenance CR with the node's name should be deleted.

$ cat deploy/crds/nodemaintenance_cr.yaml

apiVersion: kubevirt.io/v1alpha1
kind: NodeMaintenance
metadata:
  name: nodemaintenance-xyz
spec:
  nodeName: node02
  reason: "Test node maintenance"

$ kubectl delete -f deploy/crds/nodemaintenance_cr.yaml

{"level":"info","ts":1551794725.0018933,"logger":"controller_nodemaintenance","msg":"Reconciling NodeMaintenance","Request.Namespace":"default","Request.Name":"node02"}
{"level":"info","ts":1551794725.0021605,"logger":"controller_nodemaintenance","msg":"NodeMaintenance Object: default/node02 Deleted ","Request.Namespace":"default","Request.Name":"node02"}
{"level":"info","ts":1551794725.0022023,"logger":"controller_nodemaintenance","msg":"uncordon Node: node02"}

NodeMaintenance Status

The NodeMaintenance CR can contain the following status fields:

apiVersion: kubevirt.io/v1alpha1
kind: NodeMaintenance
metadata:
  name: nodemaintenance-xyz
spec:
  nodeName: node02
  reason: "Test node maintenance"
status:
  phase: "Running"
  lastError: "Last failure message"
  pendingPods: [pod-A,pod-B,pod-C]
  totalPods: 5
  evictionPods: 3

phase is the representation of the maintenance progress and can hold a string value of: Running|Succeeded. The phase is updated for each processing attempt on the CR.

lastError represents the latest error if any for the latest reconciliation.

pendingPods PendingPods is a list of pending pods for eviction.

totalPods is the total number of all pods on the node from the start.

evictionPods is the total number of pods up for eviction from the start.

Tests

Running e2e tests:

Local

Before running the tests, the NodeMaintenance CRD and namespace must be registered with the Openshift/Kubernetes apiserver:

$ kubectl create -f deploy/crds/nodemaintenance_crd.yaml
$ kubectl create -f deploy/crds/namespace.yaml

Chang the <IMAGE_VERSION> under deploy/operator.yaml image link to the desired version:

image: quay.io/kubevirt/node-maintenance-operator:<IMAGE_VERSION>

Run the operator tests locally with the default Kubernetes config file present at $HOME/.kube/config or with specificing kubeconfig via the flag --kubeconfig=<path/to/kubeconfig> and a namespace --namespace=<namespace>:

operator-sdk test local ./test/e2e --kubeconfig=<path/to/kubeconfig> --namespace="node-maintenance-operator"

Next Steps

• Handle unremoved pods and daemonsets
• Check where should the operator be deployed (infra pods?)
• Check behavior for storage pods
• Fencing
• Versioning
• Enhance error handling
• Operator integration and packaging