Scaling Without Setting Money on Fire: Journeys of a Solo Developer
I recently shared Photo Palettes to Reddit for the first time and the app crashed almost immediately.
How did this happen?
I built a beta version of Photo Palettes and deployed it to Heroku. Personal testing went well. However, the moment a few users were added everything began to fail.
It turns out that libraries that are involved with image processing and manipulation are quite memory intensive. The base plan offered by Heroku, Standard-1X, has a memory limit of 500MB. I was idling at 850MB. A few requests in parallel seemed like enough to cause disaster.
I looked into upgrading. The next tier, Standard-2X, provides two 500 MB instances - essentially two smaller servers rather than one larger one. That meant my single process idling at 850 MB would still crash, so it wouldn’t solve the problem.
The next tier up, Performance-M would solve the issue but 10x the cost. As a single developer working on my own projects, I don't have funds to light on fire.
Then what?
Most of the intensive processes didn’t need to run immediately, which meant they didn’t belong in the server at all. Heroku offers a worker process for running tasks outside of a server and so that's where I decided to move the code.
Implementation took about a day and a half. I replaced the majority of the the intensive processes with database inserts to a queue table. The code was moved from the server to the worker. The worker polls the database, checks for new records, and processes any it finds.
Heroku’s Standard-2X was sufficient for the task which doubled my spend instead of 10x’ing it. I gained a deeper understanding of my codebase and more experience in devops. Memory usage dropped from 800MB down to 300MB. Best of all, my server can now handle much more traffic without a hiccup.
I don’t believe throwing money at a technical problem is always the best answer. In cases like this, it only treats the symptom. If I had gone the money route, I’d eventually be asking, “How do I avoid 100x'ing my spend?” Moments like these are a chance to pause and ask yourself: can this be solved without spending more?
P.S. If you’re wondering about AWS or GCP, I’ve already explained why I avoid them here.
Or