AWS

Continuous delivery in AWS – tools overview

Continuous Delivery (CD) is a term that is used for a collection of practices that strive for enabling an organisation to provide both speed and quality in their software delivery process. It is a complex topic and in this article we will focus on one aspect, which is selecting tools for CD pipelines when deploying software in AWS.

Before we dive into various tooling options for continuous delivery though, let us define some scope, terminology and also talk a bit why we would bother with this in the first place.

Scope

Our scope for this overview is for delivering solutions that runs in AWS. Source code lives in a version control system and we assume that it is a hosted solution (not on-premise) and that it is Git. Git is currently the most common version control system in use. Some services mentioned may also work with other version control systems, such as Subversion, for example.

Continuous delivery tools can either be a software-as-a-service (SaaS) solution, or we can manage it ourselves on servers – in the cloud or on-premise. Our scope here is only for the SaaS solutions.

If you have a version control system that is hosted on-premise or on your own servers in some cloud, that typically works with continuous delivery software that you can host yourself – either on-premise or in the cloud. These options we will not cover here.

Terminology

First of all, there are a few terms and concepts to look at:

  • Pipeline – this generally refers to the process that starts with changing the code to the release and deployment of the updated software. This would be a mostly or even completely automated process in some cases.
  • Continuous integration (CI) – the first part of the pipeline, in which developers can perform code updates in a consistent and safe way with fast feedback loops. The idea is to do this often and that it should be quick, so any errors in changes can be caught and corrected quickly. Doing it often means that there are only small changes each time, which makes it easier to pinpoint and correct any errors. For CI to work well, it needs a version control system and a good suite of automated tests that can be executed when updates someone commits updates to a version control system.
  • Continuous Delivery (CD) – This refers to the whole process from code changes and CI to the point where a software release is ready for deployment. This includes everything in the continuous integration part and other steps that may be needed to make the software ready for release. Ideally this is also fully automated, although may include manual steps. Again, the goal is that this process is quick, consistent and safe, so that it would be possible to make a deployment “at the click of a button”, or similar simple procedure. But the deployment is not part of continuous delivery.
  • Continuous Deployment (CD) – Unfortunately the abbreviation is the same as for continuous delivery, but it is not the same. This is continuous delivery plus automated deployment. In practice, this is applicable for some solutions but not all. With serverless solutions it is generally easier to do this technically, but in many cases it is not a technology decision, but a business decision.

Why continuous delivery?

Speed and safety for the business/organisation – that is essentially what it boils down to. To be able to adapt and change based on market and changing business requirements and to do this in a way that minimises disruption of the business.

Depending on which stakeholders you look at, there are typically different aspects of this process that are of interest:

  • Business people’s interests are in speed and predictability of delivery of changed business requirements and that services continues to work satisfyingly for customers.
  • Operations people’s interests are in safety, simplicity and predictability of updates and that disruptions can be avoided.
  • Developers’ interest is in fast feedback on the work they do and that they can do changes without fear of messing things up for themselves and their colleagues. Plus that they can focus on solving problems and building useful or cool solutions.

It is a long process to reach continuous delivery Nirvana, and the world of IT a mess to various degrees – we are never done. A sane choice of tooling for continuous delivery can at least get us part of the way.

Continuous delivery tools

If we want a continuous delivery tool which targets AWS, uses git and runs as a SaaS solution, we have a few categories:

  • Services provided by AWS
  • Services provided by the managed version control system solution
  • Third party continuous delivery SaaS tools

Services provided by AWS

AWS has a number of services that is related to continuous delivery, which all have names that start with “Code” in them. This includes:

  • AWS CodeCommit
  • AWS CodePipeline
  • AWS CodeBuild
  • AWS CodeDeploy
  • AWS CodeGuru
  • AWS CodeStar

A key advantage with using AWS services is that credentials and access is the regular identity and access management (IAM) in AWS and encryption with key management service (KMS). There is no AWS secrets information that has to be stored elsewhere outside of AWS, since it all lives in AWS – assuming your CI/CD workflow goes all-in on AWS – or to a large extent at least.

A downside with these AWS services is that they are not the most user-friendly, plus there are a number of them. They can be used together to set up elaborate CI/CD workflows, but it requires a fair amount of effort to do so. CodeStar is a service here that was an attempt to set up an end-to-end development workflow with CI/CD.
I like the idea behind CodeStar and for some use cases it may be just fine. But it has not received so much love from AWS since it was launched.

You do not necessarily need all of these services to set up a CI/CD workflow – in its simplest form you just need a supported source code repository (CodeCommit/Github/Bitbucket) and CodeBuild. But things can quickly get more complicated, in particular once the number of repositories, developers and/or AWS accounts involved starts to grow. One project that tries to alleviate that pain is the AWS Deployment Framework.

Services provided by the managed version control system solution

Three of the more prominent version control system hosting services are Github, Gitlab and Bitbucket. They all have CI/CD services bundled with their hosted service offering. Both Bitbucket and Gitlab also provide on-premise/self-hosted versions of their source code repository software as well as continuous delivery tooling and other tools for the software lifecycle. The on-premise continuous delivery tooling for Bitbucket is Bamboo, while the hosted (cloud) version is Bitbucket Pipelines. For Gitlab the software is the same for both hosted and on-premise. We only cover the cloud options here.

On the surface the continuous delivery tooling is similar for all these three – a file in each repository which describes the CI/CD workflow(s) for that particular repository. They are all based on running Docker containers to execute steps in the workflow and can handle multiple branches and pipelines. They all have some kind of organisational and team handling capabilities.

Beyond the continuous delivery basics they start to deviate a bit in their capabilities and priorities. Bitbucket, being an Atlassian product/service, focus on good integration with Jira in particular, but also some 3rd party solutions. Gitlab prides itself on providing a one-stop solution/application for the whole software lifecycle – what features are enabled depends on which edition of the software that is used. Github, being perhaps the most well-known source code repository provider, has a well-established ecosystem for integration with various tools into their toolchain, provided by 3rd parties and community – more so than the other providers.

Github and Gitlab have the concept of runners that allow you to set up your own machines to run tasks in the pipelines.

So if you are already using other Atlassian products, Bitbucket and Bitbucket Pipelines might be a good fit. If you want an all-in-one solution then Gitlab can suite well. For a best-of-breed approach to pick different components, then Github is likely a good fit.

Third party continuous delivery SaaS tools

There are many providers which provide hosted continuous delivery tooling. Some of these providers have been in this space for a reasonably long time, before the managed version control system providers added their own continuous delivery tooling.

In this segment there may be providers that support specific use cases better, or are able to set up faster and/or parallel pipelines easily. They also tend to support multiple managed version control system solutions and multiple cloud provider targets. Some of them also provide self-hosted/on-premise versions of their software solutions. Thus this category of providers may be interesting for integrating with a diverse portfolio of existing solutions.

Some of the more popular SaaS providers in this space include:

Pricing models

Regardless of category, pretty much all the different providers mentioned here provide some kind of free tier and then one or more on-demand paid tiers.

For example: Github Actions, Bitbucket Pipelines, Gitlab CI/CD and AWS CodeBuild provide a number of free build minutes per month. This is however limited to certain machine sizes used in executing the tasks in the pipelines.

A simple price model of just counting build minutes is easy to grasp, but will also not allow flexibility in machine sizes, since larger machine will require more capacity from the provider. In AWS case with CodeBuild, you can select a number of different machine sizes – but you need to pay for anything larger than the smaller machines from the first minute.

The third party continuous delivery providers have slightly different free tier models, I believe partially in order to distinguish them from the offerings of the managed version control system providers. For example, CircleCI provides a number of free “credits” per week. Depending on machine capacity and feature, pipeline execution will cost different amounts of credits.

The number of parallel pipeline executions is typically also a factor for all the different providers – free tiers tend to have 1 pipeline that can execute at any time, while more parallel execution will cost more.

Many pricing models also a restriction on the number of users and there may be a price tag attached to each active user also. All in all, you pay for compute capacity, to save time on pipeline execution and to have more people utilize the continuous delivery pipelines.

AWS, with a number of services fulfilling various parts of the continuous delivery solution, may be a bit more complex to grasp initially what things will actually cost. Also, the machine sizes may not be identical across the different services either, so a build minute for one service may not necessarily be one build minute at another provider.

Trying to calculate the exact amount the continuous delivery solution will cost may be counterproductive at an early stage though. Look at features needed first and their importance, then consider pricing after that.

End notes

Selecting continuous delivery tooling can be a complex topic. The bottom line is that it is intended to deliver software faster, more secure and more consistently, with fewer problems – and with good insight into the workflow for different stakeholders. Do not loose sight of that goal and what your requirements are – beyond the simple cases. Most alternatives will be ok for the very simple cases. Do not be afraid to try out some of them, but time box the effort.

If you wish to discuss anything of the above, please feel free to contact me at erik.lundevall@tiqqe.com