A Survival Guide for Containerizing your Infrastructure — Part 1: Why switch?

Read: how we cut server costs by 75% and greatly simplified our Continuous Integration and Deployment (CI/CD) by switching from Amazon Web Services (AWS) to Docker containers on Google Cloud Platform (GCP) — in only 7 days.

Why you might want to read this post:

  1. You want to be convinced of the merits of using/switching to Docker/Kubernetes for your infrastructure.
Skeptical? You should be! I was too at first…

DISCLAIMER: I’m in no way a devops expert. What follows are my experiences as a full-stack developer diving head-first into new territory, specifically Docker, Kubernetes and Google Container Engine (GKE) on GCP. My main reason for writing this post is to showcase how accessible these new tools are, and how quickly I was able to migrate our production architecture. I welcome any corrections, edits and suggestions!

Why Containerize?

Microservices are all the rage (at least this year) and containers are playing a big role in making them feasible. Read about the pros and cons of microservices, or listen to this great a16z podcast if you’re new to the topic. But even if you don’t want to break apart your monolithic application into microservices, utilizing containers for the major architectural components of your application (your API, front-end web application, asynchronous task workers, cache, etc) can still offer up some big wins. So what exactly is a container?

container (n): a hermetically sealed package of software that can be launched quickly, run in a standardized way, and is essentially stateless.

Containers are easy to distribute because they encapsulate all of their dependencies and can be brought up / shut down quickly. Unlike a virtual machine, they aren’t actually a full OS (in fact, multiple containers can run on the same OS). They avoid network conflicts because each container has its own IP address, and they avoid version conflicts because each container packages only the library dependencies it needs.

Example: If you are familiar with python you know that virtualenvs are critical to running multiple python apps on the same system to avoid library version conflicts. With containers you can say goodbye to virtualenvs because you would create a different container for each python application, each installing its own dependencies. The container IS the virtualenv.

In my opinion though, the real advantages lie in Kubernetes, an open source system created to orchestrate the creation, lifecycles and communication between your shiny new containers.

Too many words…show me pictures!

Before we dive into Kubernetes, lets’ back up a second and look at a setup you might be familiar with. Below is an example setup on AWS with a Django API behind NGINX on an EC2 instance, another instance accepting asynchronous tasks (Rabbit broker + Celery workers) and both having access to a Postgres database on an Amazon Relational Database Service (RDS).

Simple setup on AWS

If you’re a small startup with little devops experience, your thought process for setting up the API component would probably look something like this:

  • Step 1: Log into the AWS web console and launch a new EC2 instance by choosing a base linux image and specifying an instance size. NOTE: Your monthly bill will largely be determined by the instance size you select.

%$&#! We have traffic and need to scale!

  • Step 4: Right-click on your instance and ‘Launch more instances like this’, then repeat Step 2. Manually create an Elastic Load Balancer and add your API instances to it.

If you are ultimately successful, you may end up with something more like the following. This sort of setup is NOT simple, and is definitely overkill if built before you actually need it:

Desired state on AWS

At this point you are likely throwing away hard earned cash thanks to underutilized EC2 instances…especially if, like us, you made conservatively large-sized instances so you wouldn’t have to go through the hassle of vertical scaling later.

Underutilizing your EC2 instances

Sound familiar? Personally I learned far more about Chef, Ansible and the AWS toolset than than I ever wanted to, and it left me feeling like this:

How it feels configuring EC2 instances on AWS

A Better Way

The good news is that there is hope! In only 7 days (and long nights) I was able to both (a) replicate our existing deployment architecture and (b) drastically improve it by containerizing our setup. Granted, my startup was recently accepted into the GCP for Startups program, so we had a strong incentive to switch (free GCP credits!). In hindsight though, it would have been worth it no matter what.

Below is the same setup using Kubernetes on Google Container Engine (GKE):

  • Scalability is the default. Networking, load balancing and replication are first class citizens on Kubernetes. Deployments specify the Docker image to replicate, the number of replications, and define the endpoint for testing readiness of the container. Services handle load balancing for a deployment, in addition to defining which ports to expose (internally or externally).

Want to have 6 API containers online instead of 3? Patch your Deployment configuration with “replicas: 6” and go get yourself a cold brew. There’s even support for auto-scaling.

  • Resource Efficiency. You can see that each major component has been converted to a container. Rather than guessing at the resource requirements of each component, the entire cluster (everything you see above) is treated as a single, logical unit defined in a simple, declarative format. You then decide how many resources to grant the entire cluster (# and size of vCPUs) and the resources are distributed evenly throughout the cluster to your various containers. This fundamental shift encompassed the majority of our cost savings.

Are you a Heroku user thinking “But I pay Heroku to manage all this for me!” Yes, Heroku offers automagical scalability and seamless deployment pipelines. But as I’m sure you are aware, you’re paying a premium for them. I don’t have the numbers to back it up, but I’d venture to guess you’d see similar (if not greater) cost savings by converting to containers, and you’d be left with significantly more versatility in the end.

Remember, the above example is only converting major components to containers…the logical next step is to go full microservices for each component. But once you’re on GKE, that becomes more of an application layer challenge.

Of course the devil’s in the details and learning a new system is never easy, but once you have a decent handle on it you start feeling like this:

How it feels configuring Kubernetes on GKE

There’s always a downside. Eliminating state is one of the largest reasons to containerize, but sometimes your hand is forced. In our case, we had previously been using a specially configured Postgres database with PostGIS installed for our geospatial requirements. Google’s support team recommended a Docker container running Postgres with a Persistent Disk attached as a Persistent Volume (Kubernetes terminology) to the container. Philosophically, it just feels wrong! Of course we could attempt to switch to a service like Google Cloud SQL but our goal was to make the switch with as little application refactoring as possible (thus saving time and money). So far it’s been effective for us.

In fairness, I should note that AWS has a Container Service as well, and you can read about some differences between it and GKE here. But that’s the beautiful thing about containers…if AWS offers a strong enough incentive for us to switch, it should be relatively straightforward.

RECAP: Main Advantages of Containerizing on GKE

  1. Portable. Any system running the Docker Engine can run your containers the exact same way, whether that’s in GKE, AWS or on your personal laptop. You can even replicate your entire Kubernetes orchestration on your laptop with something called minikube…it’s pretty remarkable. That means (a) increased confidence in your deployments and (b) negligible switching cost to any other infrastructure provider that supports containers.

Deploying software can be one of the most intimidating parts of shipping a product. Your goal is to eliminate surprises when new features go live, whether you’re deploying a minor copy change or launching an entirely new feature. Similarly, you want to be prepared to scale seamlessly with your userbase. Having shipped products for my own startups multiple times, I can say from personal experience that containerization is a giant leap forward, even for if you’re just starting with a brand new project.

Ready to make the switch? Check back soon for Part 2: Tips, Tricks and Pitfall.

If you liked this, click the💚 below so other people will see this here on Medium.



A collection of technical articles and blogs published or curated by Google Cloud Developer Advocates. The views expressed are those of the authors and don't necessarily reflect those of Google.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Austin Cooley

Founder @Resource_io — making recruiting human. Previously founded @Tripstr, @Quilt. Kauffman Labs for Ed Tech Ventures.