How do you deploy in 10 seconds?
How do you deploy in 10 seconds?
This post describes my lessons learned after 10 years running production environments in sizes ranging from "just getting started" to a "Series F" company wi...
Hi friends, as promised, I'm back with my second post. I'll be hanging around in the comments for any questions!
In this post, I take a look at a typical deployment process, how long each part of it takes, and then I present a simple alternative that I use which is much faster and perfect for hobbit software.
So it really is that simple: a small bash script, building locally, rsync'ing the changes, and restarting the service. It's just the bare essentials of a deployment. That's how I deploy in 10 seconds.
I'm strongly opposed to local builds on any semi-important or semi-complex production product or system.
Tagged CI release builds give you a lot of important guarantees involved in release concerns.
I'll take the fresh checkout and release build time cost for those consistency and versioned source state guarantees.
I would imagine you could run into an issue like this building off an M1 or newer Mac and deploying to a Linux based env. We've run into a bit of an adjustment with our docker image builds where we need to set the buildarch or else it fails to deploy.
Our build times aren't blazingly fast, typically around 4 minutes for npm/yarn build for frontend apps and loading the data to the image and any other extras like composer installs. Best time saving for us was doing a base image for all the dependency junk that we do a nightly on
This was exactly the problem in my last environment. I was the second dev and two more were onboarded after me, but everyone had issues replicating the original dev’s local environment in order to deploy.
First thing I did was set up a basic gitops pipeline. Worked like a charm.
Your proposed solution to overly complex systems seems to be to ignore the requirements that make them complex in the first place. If that works for you, this is a perfectly fine approach. But most companies with actual signed SLAs won't accept "we'll just have a few seconds of downtime/high latency every time a developer deploys something to production #yolo".
Also, series F but they're only deploying on one server? Try scaling that to a real deployment (200+ servers) with millions of requests going through and see how well that goes.
And also no way their process passes ISO/SOC 2/PCI certifications. CI/CD isn't just "make do things", it's also the process, the logs, all the checks done, mandatory peer reviews. You can't just deploy without the audit logs of who pushed what when and who approved it.
You're not wrong, but not everything needs to scale to 200+ servers (...arguably almost nothing does), and I've actually seen middle managers assume that a product needs that kind of scale when in fact the product was fundamentally not targeting a large enough market for that.
Similarly, not everything needs certifications, but of course if you do need them there's absolutely no getting around it.
I'm not sure I understand the trade offs you're choosing by deploying this way. The benefit of simplicity an speed of deployment seems clear from your write-up. But are those the most important considerations? Why or why not?
this has the same logic as saying npm install takes a while so just don’t use libraries
Also yes.
Deploy very small changes?