How to do zero-downtime releases

Learn how to release new versions without any downtime using Istio

The purpose of the zero-downtime release is to release a new version of the application without affecting its users. If you have a website running, this means that you can release a new version without taking the website down. It means that you can make continuous requests to that application while releasing a new application, and the application users will never get that dreaded 504 Service Unavailable response.

You might wonder why we would use Istio to do rolling updates if the Kubernetes option is much simpler. Yes, you can get zero-downtime deployments whether you use Istio or Kubernetes. However, with Istio, you get more features and control over traffic routing. You can use weight-based routing, mirroring the traffic to a different version, or you can route the traffic based on the request properties (URI, scheme, method, etc.).

Prerequisites

You can follow the prerequisites for instructions on how to install and setup Istio.

Kubernetes Deployments need to be versioned

Each deployment of the service needs to be versioned - you need a label called version: v1 (or release: prod or anything similar to that), as well as name the deployment, so it's clear which version it represents (e.g. helloworld-v1). Usually, you'd have at minimum these two labels set on each deployment:

labels:
    app: helloworld
    version: v1

You could also include many other labels if it makes sense, but you should have a label that identifies your component and its version.

Kubernetes Service needs to be generic

There's no need to put a version label in the Kubernetes Service selector. The label with the app/component name is enough. Also, keep the following in mind:

1. Start with a destination rule that contains versions you are currently running, and make sure you keep it in sync. There's no need to end up with a destination rule with many unused or obsolete subsets.

2. If you are using matching and conditions, always define a "fallback" route in the VirtualService resource. If you don't, any requests not matching the conditions will end up in digital heaven and won't get served.

Let's take the following example:

apiVersion: networking.istio.io/v1alpha3
kind: VirtualService
metadata:
    name: my-service
spec:
    hosts:
      - my-service.default.svc.cluster.local
    http:
      - match:
        - headers:
            my-header:
              regex: '.*debug.*'
        route:
          - destination:
              host: my-service.default.svc.cluster.local
            port:
              number: 3000
            subset: debug

The above VirtualService is missing a "fallback" route. In case the request doesn't match (i.e., missing my-header: debug, for example), Istio won't know where to route the traffic to. To fix this, always define a route that applies if none of the matches evaluates to true. Here's the same VirtualService, with a fallback route to the subset called prod.

apiVersion: networking.istio.io/v1alpha3
kind: VirtualService
metadata:
    name: my-service
spec:
    hosts:
      - my-service.default.svc.cluster.local
    http:
      - match:
        - headers:
            my-header:
              regex: '.*debug.*'
        route:
          - destination:
              host: my-service.default.svc.cluster.local
            port:
              number: 3000
            subset: debug
      - route:
        - destination:
            host: my-service.default.svc.cluster.local
            port:
                number: 3000
            subset: prod

With these guidelines in mind, here's a rough process of doing a zero-downtime deployment using Istio. We are starting with Kubernetes deployment called helloworld-v1, a destination rule with one subset (v1) and a VirtualService resource that routes all traffic to the v1 subset. Here's how the DestinationRule resource looks like:

apiVersion: networking.istio.io/v1alpha3
kind: DestinationRule
metadata:
  name: helloworld
spec:
  host: helloworld.default.svc.cluster.local
  subsets:
    - name: v1
      labels:
        version: v1

And the corresponding virtual service:

apiVersion: networking.istio.io/v1alpha3
kind: VirtualService
metadata:
  name: helloworld
spec:
  hosts:
    - helloworld
  http:
    - route:
      - destination:
          host: helloworld
          port:
            number: 3000
          subset: v1
        weight: 100

Once you deployed these two resources, all traffic is routed to the v1 subset.

Rolling out the second version

Before you deploy the second version, the first thing you need to do is to modify the DestinationRule and add a subset that represents the second version.

1. Deploy the modified destination rule that adds the new subset:

   apiVersion: networking.istio.io/v1alpha3
   kind: DestinationRule
   metadata:
     name: helloworld
   spec:
     host: helloworld.default.svc.cluster.local
     subsets:
       - name: v1
         labels:
           version: v1
       - name: v2
         labels:
           version: v2

2. Create/deploy the helloworld-v2 Kubernetes deployment.

3. Update the virtual service and re-deploy it. In the virtual service, you can configure a percentage of the traffic to the subset v1 and a percentage of the traffic to the new subset v2.

There are multiple ways you can do this - you can gradually route more traffic to v2 (e.g., in 10% increments, for example), or you can do a straight 50/50 split between versions, or even route 100% of the traffic to the new v2 subset.

Finally, once you routed all traffic to the newest/latest subset, you can follow the steps in this order to remove the previous v1 deployment and subset:

1. Remove the v1 subset from the VirtualService and re-deploy it. This will cause all traffic to go to v2 subset.
2. Remove the v1 subset from the DestinationRule and re-deploy it.
3. Finally, you can now safely delete the v1 Kubernetes deployment, as no traffic is being sent to it anymore.

If you got to this part, all traffic is now flowing to the v2 subset, and you don't have any v1 artifacts running anymore.