Develop on a Cloud Machine

I'm a full stack growth engineer at Credit Karma. My team has recently made some changes to how we work, so I wanted to share how positive the results have been.

We’ve found that doing our development work on AWS machines is much faster than working locally, especially for projects that involve our monolithic PHP application.

For us, developing on AWS machines is better than developing in a local VM, but it depends on the size of your company and how much memory/CPU your app needs.

There are a few reasons for this, at least at Credit Karma:

How do you get your changes to the AWS machine?

There are a number of ways, including vi over ssh, sshfs, VS Code extensions, and rsync.

I found rsync to be the most straightforward. Back at Voxer, CJ Silverio, now the CTO of npm, wrote a script for this. Her script checks the working directory for changes with git status and syncs only what has changed. I added awareness of our repo’s stable branch, so that you’re free to commit as you like, and your changes will still end up on the AWS machine.

Using our script, which we call pushuat, we are able to develop at least 3x faster than in a local VM. We even see 10x+ development speed improvements for some parts of the codebase, where it is practically impossible to make changes because the app is so resource intensive.

I’m planning to open source the script and I hope to link to it soon. The script runs rsync every second, which is very low effort compared to learning vim or committing and pushing and pulling every iterative change on your feature branch. Not to mention that you probably have development specific config changes that you don’t want to commit.

Here are a few of the symptoms you and your team should watch out for, to tip you towards working on cloud machines:

How expensive is it to develop on an AWS machine?

At Credit Karma we run two machines in concert to create one development environment. That comes out to about $500/mo.

I personally only need one development environment at any given moment. When it comes to my happiness and productivity as an engineer, I find it very worth it.

“$500/mo, that’s so privileged”

Just like one of my coworkers, you’re thinking “to say you can save yourself however many minutes a day by spending $500/mo sounds very privileged. It’s a rare company that would not find that an onerous expense.”

Tech companies in the bay area pay >$100k/y base for a junior engineer, with a fully loaded cost of >$140k/y. A multiplicative impact on productivity for only 4% of the fully loaded cost is a great deal for a business person. If you’d like to tweak the numbers to fit your company, drag the numbers around on (which I made just for you :).

Based on my meetings with the team that maintains the build process for the AWS machines, it is worth it for them too. After my talk at our engineering summit, we had so many new machines spun up that we ran out of our IP range. Thankfully, they opened up more IPs for us :)

From a developer experience perspective, they’ve made the process of requesting a new machine very smooth. You visit a shortlink, enter a few parameters into a Jenkins build, and then wait for an email saying your machine is ready. Thank you Hok Tong and Daniel Martinez, you’ve done a great job on this.

That’s all for now—I highly recommend you try working on a cloud machine. Email if you have questions :)

David Trejo
Email me ✉️

David Trejo

Growth Engineer at Credit Karma & consultant. Past clients include Aconex, Triplebyte, Neo, Brown Computer Science Department, Voxer, Cloudera, and the Veteran's Benefits Administration.

Want to be a better Full Stack Growth Engineer?

You'll get articles on everything I learn as a growth engineer at Credit Karma, a profitable private company with >80 million members, >$4 billion valuation, and >900 employees.