DEV Community

Cover image for Tips and tricks for scalable CI/CD flow
Alex Barashkov
Alex Barashkov

Posted on

Tips and tricks for scalable CI/CD flow

When you plan a project, you don’t always know how big it will be by the end. It’s crucial, therefore, to follow guidelines that make the development process as smooth as possible. One especially useful tool for this is a Continuous Integration/Delivery flow, or CI/CD flow.

At the beginning of a typical development process, before the project is ready for publication, your CI flow looks something like this:

During the development phase, more steps are added to the process until it looks something like the image below. However, you may see many more steps than this in some projects. Original image.

In this article, I will demonstrate a typical CI/CD flow and provide tips to help you scale this up for bigger projects. Note that these recommendations are made on the assumption that you have developed your infrastructure setup using Docker and orchestration tools, such as Kubernetes.

Define Your VCS Flow

All modern CI tools rely on version control systems, so to maintain your CI flow at a high level you should have a proper version control flow in place. I recommend Git Flow or Github Flow if you’re looking for ideas. Pick one and adapt your pipeline according to your needs.


  • Run tests on each push in any branch. This helps your developers find bugs faster and speed up the code review process.
  • Define branches that you will use to trigger Docker image build processes and deploy them as shown in the image above.
  • If your releases have a complicated manual approval process, use tags to trigger production releases.

Specify Your Environments

Once you have a git flow in place, integrate it with your builds’ logic and adapt it for the number of environments you have. Initially, you will likely deploy everything to production after your local environment testing. Later on, you may decide to do development, UAT, and production.


We often use a master branch to initiate deployment to the development environment. Then we merge the master branch into the release branch. This triggers a deployment to UAT, where we allow a small number of users to test the new version. If everything works correctly and the QA team confirms the release, we create a tag from the release branch, which initiates deployment to production.

Set Up Notifications

You should keep track of all CI flow processes via the notification channel you prefer. We use Slack for our projects.

  • Create a system-notification channel and enable it to send notifications about all CI and infrastructure processes.
  • Set up notifications about success or failure statuses and to alert you when the build starts.

Keep Docker Image Simple

It is best practice not to pass to Docker image env variables. All env variables should be passed through the Docker run command or on your orchestration tool level (for example, Kubernetes). This will also help keep your CI configuration files simple.

Unify Your Pipeline

Companies today increasingly use microservices architecture; a unified pipeline helps you benefit from this by scaling faster to deployment and configuring your CI in new microservices. In Drone CI or Travis CI, there are plenty of CI environment variables you could use during your pipeline steps (find them at Travis CI env variables and Drone CI env variables). Here are a few examples of CI env variables in action:

  • To tag a Docker image, you can use a concatenation of a build branch and build number. As a result, you will have something like “master-6.” Alternatively, you can use your git tag version. tags: "${DRONE_COMMIT_BRANCH}-${DRONE_BUILD_NUMBER}"

  • Use a repository name for deployments in your orchestration tools. Example of Drone CI deployment to Kubernetes cluster with env variables.

   image: peloton/drone-k8s-deployment
   deployment_names: "${DRONE_REPO_NAME}"
   container_names: "${DRONE_REPO_NAME}"
   namespaces: microservices
   secrets: [kubernetes_url, kubernetes_token]
Enter fullscreen mode Exit fullscreen mode

Define Docker Image Names Convention

You should define a strategy for naming Docker images based on the specifics and importance of your environment. For example, in development and UAT environments where you don’t need to have a rollback functionality, you can use branch names as a tag so that all new builds override the previous image. For production, it’s better to tag images with the version number of the release (v1.0.0), which makes your registry more structured and readable by reducing its size.

Use Builder Pattern

If you can compile and build something on the CI level, it can be helpful to do this as an extra step before putting it in the final image. Take a look at this video to get a better understanding.

DB Migrations as a Separate Step

Do not do your migration as a part of your Dockerfile; instead, do it in a separate step. Extra commands, such as compilation of assets or migrations, should be run as separate steps, not executed at Docker start. You should also try to get your Docker images up and running in the shortest amount of time possible.

Avoid Private Libraries and Submodules

Any extra dependencies, such as private npm modules or submodules, can cause problems for you during the setup process of your CI configuration. For example, if you have a submodule and by default your CI only has access to your repositories through http and an authorization token, your configuration for the local machine through git will not work. You will need to override your submodule and not all CI tools provide that feature.

Drone CI Submodule Override Example

You will face similar issues if you use private npm modules, because you will need passkeys or already downloaded modules inside the Docker Image.


When you configure a CI for a new project, spend some time thinking about and defining name conventions, potential methods of growing and unification of your build steps. This helps you to deploy new services faster, use those practices as a groundwork for new projects and onboard employees in the pipeline process faster.

Top comments (10)

anortef profile image
Adrián Norte

How can it be Continuous Integration if it uses branches?

the DB migration thing should be written on the sky :P

swehunter2000 profile image

why would branches and CI be mutually exclusive? Branches might be the most important part when developing a system with multiple developers.

anortef profile image
Adrián Norte

CI means Continuous (keyword here) Integration it means that every change made to the code is integrated the moment it is committed therefore branching cannot be CI at most it can be Very Often Integration but not CI.

This guy explains it way better than me.

Thread Thread
risingsunstudios profile image
Rising Sun Studios

Did you read all the comments in reply to that blog post? Branching does not prevent CI. Failing to integrate changes regularly prevents CI.

Thread Thread
anortef profile image
Adrián Norte

I think those comments are wrong and didn't got the idea.

Continuous integration is continuous not often, continuous.
That means that every change, no matter how little, is integrated immediately. Branches on the other hand only integrate when, in best case, the developer does rebase often(and is still a half integration cuz you are doing things on your bubble without active feedback) or in worst case, when the branch is merged therefore the integration is not continuous.

Thread Thread
neskhodovskiy profile image
Serhiy Neskhodovskiy

I think you misunderstand CI.

Unless you are the only developer on the project (and very often even then) you will have branches in your VCS.

Having branches does not prevent you from doing CI. However, not every branch is purposed for CI of course. You don't want to integrate everything without control, do you? Otherwise a small mistake may take down entire project and you'll have to rollback (and lose dollars meanwhile).

So CI is ONLY done on SOME branches. 1. A change is made on non-CI branch. 2. Change passes quality control. 3. Change is merged into a CI branch. 4. Deployment is triggered automatically on EVERY change in CI branch, that's why it's still CONTINUOUS integration (related to that controlled branch).

Thread Thread
anortef profile image
Adrián Norte

No, it isn´t and I explained it in my comment the why.

Branches are totally incompatible with CI is literally impossible to do CI having a branching strategy because you wouldn´t be integrating every little change in a continuous manner and your comment already says so with the "You don't want to integrate everything without control, do you?". Yes, I want to integrate everything and then run tests because that is CI.

The only way to do CI is to use trunk based development if not what you are doing is promiscuous integration that is also not bad, not everyone needs or is ready for CI.

Thread Thread
neskhodovskiy profile image
Serhiy Neskhodovskiy

Okay, Adrián, can you please reference any widely accepted definitions that prove your point?

Thread Thread
anortef profile image
Adrián Norte • Edited
Thread Thread
neskhodovskiy profile image
Serhiy Neskhodovskiy