Blue-Green Deployments | Upgrading Clusters

Overview

This topic serves as an alternative approach for node host upgrades to the in-place upgrade method.

The blue-green deployment upgrade method follows a similar flow to the in-place method: masters and etcd servers are still upgraded first, however a parallel environment is created for new node hosts instead of upgrading them in-place.

This method allows administrators to switch traffic from the old set of node hosts (e.g., the blue deployment) to the new set (e.g., the green deployment) after the new deployment has been verified. If a problem is detected, it is also then easy to rollback to the old deployment quickly.

While blue-green is a proven and valid strategy for deploying just about any software, there are always trade-offs. Not all environments have the same uptime requirements or the resources to properly perform blue-green deployments.

In an OpenShift Origin environment, the most suitable candidate for blue-green deployments are the node hosts. All user processes run on these systems and even critical pieces of OpenShift Origin infrastructure are self-hosted on these resources. Uptime is most important for these workloads and the additional complexity of blue-green deployments can be justified.

The exact implementation of this approach varies based on your requirements. Often the main challenge is having the excess capacity to facilitate such an approach.

Figure 1. Blue-Green Deployment

Preparing for a Blue-Green Upgrade

After you have upgraded your master and etcd hosts using method described for In-place Upgrades, use the following sections to prepare your environment for a blue-green upgrade of the remaining node hosts.

Labeling Blue Nodes

You must ensure that your current node hosts in production are labeled either blue or green. In this example, the current production environment will be blue and the new environment will be green.

Get the current list of node names known to the cluster:
```
$ oc get nodes
```
Ensure that all hosts have appropriate node labels. Consider that by default, all master hosts are also configured as unschedulable node hosts (so that they are joined to the pod network). Additional labels will assist in management of clusters, specifically labeling hosts as their types (e.g., type=master or type=node) should assist management.

For example, to label node hosts that are also masters as type=master, run the following for each relevant <node_name>:
```
$ oc label node <node_name> type=master
```
To label non-master node hosts as type=node, run the following for each relevant <node_name>:
```
$ oc label node <node_name> type=node
```
Alternatively, if you have already finished labeling certain nodes with type=master and just want to label all remaining nodes as type=node, you can use the --all option and any hosts that already had a type= set will not be overwritten:
```
$ oc label node --all type=node
```
Label all non-master node hosts in your current production environment to color=blue. For example, using the labels described in the previous step:
```
$ oc label node -l type=node color=blue
```
In the above command, the -l flag is used to match a subset of the environment using the selector type=node, and all matches are labeled with color=blue.

Creating and Labeling Green Nodes

Create the new green environment for any node hosts that are to be replaced by adding an equal number of new node hosts to the existing cluster. You can use either the quick installer or advanced install method as described in Adding Hosts to an Existing Cluster.

When adding these new nodes, use the following Ansible variables:

Apply the color=green label automatically during the installation of these hosts by setting the openshift_node_labels variable for each node host. You can always adjust the labels after installation as well, if needed, using the oc label node command.
In order to delay workload scheduling until the nodes are deemed healthy (which you will verify in later steps), set the openshift_schedulable=false variable for each node host to ensure they are unschedulable initially.

Example new_nodes Host Group

Add the following to your existing inventory. Everything that was in your inventory previously should remain.

[new_nodes]
node4.example.com openshift_node_labels="{'region': 'primary', 'color':'green'}" openshift_schedulable=false
node5.example.com openshift_node_labels="{'region': 'primary', 'color':'green'}" openshift_schedulable=false
node6.example.com openshift_node_labels="{'region': 'primary', 'color':'green'}" openshift_schedulable=false
infra-node3.example.com openshift_node_labels="{'region': 'infra', 'color':'green'}" openshift_schedulable=false
infra-node4.example.com openshift_node_labels="{'region': 'infra', 'color':'green'}" openshift_schedulable=false

Verifying Green Nodes

Verify that your new green nodes are in a healthy state. Perform the following checklist:

Verify that new nodes are detected in the cluster and are in Ready state:

$ oc get nodes

ip-172-31-49-10.ec2.internal    Ready                      3d

Verify that the green nodes have proper labels:

$ oc get nodes -o wide --show-labels

ip-172-31-49-10.ec2.internal    Ready                      4d        beta.kubernetes.io/arch=amd64,beta.kubernetes.io/instance-type=m4.large,beta.kubernetes.io/os=linux,color=green,failure-domain.beta.kubernetes.io/region=us-east-1,failure-domain.beta.kubernetes.io/zone=us-east-1c,hostname=openshift-cluster-1d005,kubernetes.io/hostname=ip-172-31-49-10.ec2.internal,region=us-east-1,type=infra

Perform a diagnostic check for the cluster:

$ oc adm diagnostics

[Note] Determining if client configuration exists for client/cluster diagnostics
Info:  Successfully read a client config file at '/root/.kube/config'
Info:  Using context for cluster-admin access: 'default/internal-api-upgradetest-openshift-com:443/system:admin'
[Note] Performing systemd discovery

[Note] Running diagnostic: ConfigContexts[default/api-upgradetest-openshift-com:443/system:admin]
       Description: Validate client config context is complete and has connectivity
...
         [Note] Running diagnostic: CheckExternalNetwork
              Description: Check that external network is accessible within a pod

       [Note] Running diagnostic: CheckNodeNetwork
              Description: Check that pods in the cluster can access its own node.

       [Note] Running diagnostic: CheckPodNetwork
              Description: Check pod to pod communication in the cluster. In case of ovs-subnet network plugin, all pods
should be able to communicate with each other and in case of multitenant network plugin, pods in non-global projects
should be isolated and pods in global projects should be able to access any pod in the cluster and vice versa.

       [Note] Running diagnostic: CheckServiceNetwork
              Description: Check pod to service communication in the cluster. In case of ovs-subnet network plugin, all
pods should be able to communicate with all services and in case of multitenant network plugin, services in non-global
projects should be isolated and pods in global projects should be able to access any service in the cluster.
...

Registry and Router Canary Deployments

A common practice is to scale the registry and router pods until they are migrated to new (green) infrastructure node hosts. For these pods, a canary deployment approach is commonly used.

Scaling these pods up will make them immediately active on the new infrastructure nodes. Pointing their deployment configuration to the new image initiates a rolling update. However, because of node anti-affinity, and the fact that the blue nodes are still unschedulable, the deployments to the old nodes will fail.

At this point, the registry and router deployments can be scaled down to the original number of pods. At any given point, the original number of pods is still available so no capacity is lost and downtime should be avoided.

Warming the Green Nodes

In order for pods to be migrated from the blue environment to the green, the required container images must be pulled. Network latency and load on the registry can cause delays if there is not sufficient capacity built in to the environment.

Often, the best way to minimize impact to the running system is to trigger new pod deployments that will land on the new nodes. T Accomplish this by importing new image streams.

Major releases of OpenShift Origin (and sometimes asynchronous errata updates) introduce new image streams for builder images for users of Source-to-Image (S2I). Upon import, any builds or deployments configured with image change triggers are automatically created.

Another benefit of triggering the builds is that it does a fairly good job of fetching the majority of the ancillary images to all node hosts such as the various builder images, the pod infrastructure image, and deployers. The green nodes are then considered warmed (that is, ready for the expected load increase), and everything else can be migrated over using node evacuation in a later step, proceeding more quickly as a result.

When you are ready to continue with the upgrade process, follow these steps to warm the green nodes:

Disable the blue nodes so that no new pods are run on them by setting them unschedulable:
```
$ oc adm manage-node --schedulable=false --selector=color=blue
```
Set the green nodes to schedulable so that new pods only land on them:
```
$ oc adm manage-node --schedulable=true --selector=color=green
```
Update the default image streams and templates as described in Manual In-place Upgrades.
Import the latest images as described in Manual In-place Upgrades.

It is important to realize that this process can trigger a large number of builds. The good news is that the builds are performed on the green nodes and, therefore, do not impact any traffic on the blue deployment.
To monitor build progress across all namespaces (projects) in the cluster:
```
$ oc get events -w --all-namespaces
```
In large environments, builds rarely completely stop. However, you should see a large increase and decrease caused by the administrative image import.

Evacuating and Decommissioning Blue Nodes

For larger deployments, it is possible to have other labels that help determine how evacuation can be coordinated. The most conservative approach for avoiding downtime is to evacuate one node host at a time.

If services are composed of pods using zone anti-affinity, then an entire zone can be evacuated at once. It is important to ensure that the storage volumes used are available in the new zone as this detail can vary among cloud providers.

In OpenShift Origin 1.2 and later, a node host evacuation is triggered whenever the node service is stopped. Node labeling is very important and can cause issues if nodes are mislabled or commands are run on nodes with generalized labels. Exercise caution if master hosts are also labeled with color=blue.

When you are ready to continue with the upgrade process, follow these steps.

Evacuate and delete all blue nodes by following one of the following options:

Option A Manually evacuate then delete the color=blue nodes with the following commands:

$ oc adm manage-node --selector=color=blue --evacuate
$ oc delete node --selector=color=blue

Option B Filter out the masters before running the delete command:

Verify the list of blue node hosts are as expected by running:

$ oc get nodes -o go-template='{{ range .items }}{{ if and (eq .metadata.labels.foo "bar") \
    (ne .metadata.labels.type "master") }}{{ .metadata.name }}{{ "\n" }}{{end}}{{ end }}';

After the list is determined to be of the blue nodes, run:

$ for i in $(oc get nodes -o \
    go-template='{{ range .items }}{{ if and (eq .metadata.labels.color "blue") \
    (ne .metadata.labels.type "master") }}{{ .metadata.name }}{{ "\n" }}{{end}}{{ end }}'); \
do
    oc delete node $i
done

After the blue node hosts no longer contain pods and have been removed from OpenShift Origin they are safe to power off. As a safety precaution, leaving the hosts around for a short period of time can prove beneficial if the upgrade has issues.
Ensure that any desired scripts or files are captured before terminating these hosts. After a determined time period and capacity is not an issue, remove these hosts.