Rolling Update Deployment Strategy

Rolling Update is a Deployment StrategyDeployment Strategy
Deployment strategies describe how we handle the release of the new code to an environment. Each has it's different use case, pros and cons.

Here is a (non-definitive) list of deployment strategie... (also known as Ramped or Incremental), where we slowly replace version A with version B.

How it works

  • before we start, we have 2 instances of A running and serving traffic (let's call this starting point stage)
  • until all instances of A are replaced with instances of B - repeat this:
    • a new instance of B is created, and once it's ready, it starts accepting traffic (create new stage)
    • once that happens, 1 instance of A is killed (kill old stage)

During this process, the number of instances over time changes like this:

  • A: 2 B: 0 (starting point stage)
  • A: 2 B: 1 (after create new stage)
  • A: 1 B: 1 (after kill old stage)
  • A: 1 B: 2 (after create new stage)
  • A: 0 B: 2 (after kill old stage)

The example above replaces instances one at a time - we could configure this to be done with multiple instances in parallel.

During the deployment, our number of active instances is always between the desired count and (desired count + number of new instances rolling out at the same time).

Tradeoffs

Pros:

  • zero downtime
  • pretty easy and cheap to execute

Cons:

  • rollouts and rollbacks can last a long time (depending on number of instances and startup time)
  • during deployment, some users will be served old version, some new version - we have no control over this aspect

Status: #💡

References: