Zero Downtime deployment is a technique used in software deployment that aims to reduce or eliminate the time the system is unavailable to users during a software update or deployment.
Zero downtime deployments can be achieved through careful planning and execution.
This article will discuss why it is important to have zero downtime deployments, how to implement them in your application, and the potential risks and benefits of implementing them.
What is Zero Downtime deployment?
A successful zero downtime deployment is considered a release of a new version of your application without your end users having to wait for the update to complete and even notice that the application is unavailable.
With an actively developed project that might trigger your CI/CD pipeline multiple times per day, your project might often have to be deployed to production. Imagine if your application was taken down for maintenance each time you deployed it, then your users would have to wait for the application to be back up before they could use it, which is unacceptable. This is where Zero Downtime deployment comes into play.
Why is it important?
There are several benefits to using zero-downtime deployments:
- Increased system uptime
- Reduced downtime for users
- Reduced cost of downtime
- Improved reliability
- Consistent deployments
- Continuous integration and deployment
Every time your application is down, your business will be negatively impacted. For reference, Facebook experienced an approximate of 6-hour outage on October 4th which could roughly be estimated to $60 million in lost revenue.
What are some of the challenges associated with zero downtime deployments?
There are some drawbacks to zero downtime deployments that you should be considered:
- Increased complexity and maintenance
- Increased cost of deployment and infrastructure
- Backward compatibility requirements
- Data compatibility issues
As we mentioned earlier, a successful zero downtime deployment is achieved by carefully planning, so next, let's review a few approaches that you can consider for implementing zero-downtime deployments.
Zero downtime deployment strategies
As with any part of the software development process, these challenges can be overcome by well-planned and executed deployment strategies.
Let's review a few deployment strategies that you can consider for implementing zero-downtime deployments.
Blue/Green deployment
There are many ways to achieve zero downtime, but the most common technique is to deploy new software in parallel with the old software.
The process goes like this:
- The old instance of the application is kept up and running while the new software is deployed.
- You then start the new instance of your application and wait for it to be ready.
- Once the new instance is ready, you can switch the traffic to the new model. Depending on your setup, this could be achieved by changing the DNS record of the old instance to point to the new instance or switching your load balancer rules to forward the traffic to the new model.
- Finally, once your old instance is no longer needed, you can shut it down.
Here is a quick diagram of the process:
Canary deployment
Canary deployments are used to test new software with a limited number of users before it is released to production.
You can ensure that the new software is ready for production before releasing it to all users.
So rather than releasing a new version of your application to all users, you can release a new version to a small number of users. This gives you the ability to test the latest version of your application, gather feedback, fix any unexpected issues, and only then release it to all users.
Some companies allow users to opt-in to testing new beta features before they are released to the public.
The main issue with canary deployments is that you need to anticipate data compatibility issues between the old and new versions of your application as they are being deployed simultaneously.
Here is a quick diagram of the process:
Rolling updates
Rolling updates are deployments used to update the software on a rolling basis meaning that current running instances are replaced with new cases in a controlled manner. That way, the old instance is replaced only when a new instance is ready.
An essential part of rolling updates is that you need to be able to move back to the previous version of your application if the new version fails to deploy.
This is a common deployment strategy for web applications and is widely used with Kubernetes and microservices.
For more information on rolling updates with Kubernetes, see this article.
Here is a quick diagram of the process:
Conclusion
As you can see, zero downtime deployments are a great way to improve the reliability of your application. Despite the drawbacks, the benefits of zero-downtime deployments are vital to your application success and are worth the effort required to achieve them.
No matter which deployment strategy you choose, you should make sure that you adopt test-driven development (TDD) early on to test your code as you develop it and ensure that you are not deploying a faulty version of your code.
With all that being said, note that downtime is inevitable. It is often caused by human errors, edge cases, unexpected issues, or unexpected events. Your goal should be to minimize downtime and ensure that your application is always available to your users. You just need to make sure that you learn from your mistakes and develop processes that will minimize them.
