Kamal: Reason behind choosing Docker instead micro VMs (like firecracker)

Micro VMs are not lower overhead than Docker! Where are people getting this idea that micro VMs are faster? Micro VMs provide greater security isolation, resource isolation, and allow the guest to run a different kernel/OS than the host, but they aren’t “faster”. OCI runtimes can spawn a container orders of magnitude faster than Firecracker can spawn a VM, even though Firecracker is quite fast and light compared to traditional VM runners.

If you’re a hosting provider like Fly or AWS, micro VMs are an important primitive for those reasons. But if you’re a company building out a cluster to run your own apps, you are probably fine to share a Linux kernel version and reign in or relocate noisy neighbors. After all you will probably use the same base Docker container as a starting point for your similar, homogenous apps. If you’re sharing base layers, why not share the kernel at runtime?

Skip this. It's just some guy asking on GitHub in 2023, and 2-3 nobodies speculating. Why this was posted to HN, I have no idea.

I'm more excited to learn about Kamal here. [0] Need to dig into the differences between this and Dokku. At first glance, Kamal seems more config file focused while Dokku is command centric. Kamal also seems less mature, lacking the Dokku's plugin ecosystem and detailed documentation.

[0]: https://kamal-deploy.org/

I find it funny that by now Ignite, the tool mentioned in the thread has already been abandoned in favor of another project, Flintlock, which also seems pretty abandoned (only Dependabot activity). I guess most companies choose containers as they’re mature and have good tooling available, whereas MicroVMs are still pretty experimental especially long-running ones as you have many restrictions and memory management is painful. I think for your own stuff Docker makes most sense, if you’re hosting untrusted code then yeah MicroVMs are the way to go.

Well color me disappointed, I was hoping this was going to be a technical discussion about the decision/tradeoffs. Instead it's just an old question from the community that has no official answer from the Basecamp folks and no recent activity.

As long doing conjecture, I would guess it's because the team at Basecamp/37 Signals already had significant experience with Docker, it's much more widely used and understood in the industry, and it was a simpler approach.

Here's a technical blog from 37 Signals themselves: https://dev.37signals.com/bringing-our-apps-back-home/