Hacker News new | past | comments | ask | show | jobs | submit login

The problem is that docker compose starts 20 containers and the fans go full bore just because you wanted to try a new wiki or notes app. The complexity of relatively simple software is getting insane.



I am one of the 10 users[1] in the world that uses Docker Swarm (container orchestrator like Kubernetes or Nomad), and I disagree with that statement. I have over 25 containers running (including Jellyfin, Nextcloud, Gitea, TeamSpeak, ...) and it barely uses 10GB of RAM (Jellyfin is eating up more than half of this).

Most of my Compose files contain 2 services (1 for app + 1 for database), but some contain 3 and some contain 1. It's incredibly easy to install new software and incredibly easy to shut it down temporarily if you don't need to use it all day.

I'd even argue that some companies would benefit more from using Swarm than Kubernetes. There is a lot of things to take into account when using Kubernetes (let alone setting it up for GitOps), but Docker Swarm can be managed by 1 person.

[1]: A joke, obviously, but it really isn't popular at all


I think k8s can be managed by one person too, if you are only using it like docker swarm. Especially if you use something like k3s with SQL database.

I found setting up gitops via flux quite easy, apart from order of operations, like installing controllers and custom resource definitions before resources that need those CRDs etc.

What were you thinking of things to take into account for k8s over swarm?

The main difference for me is k8s needs a hell of a lot more boilerplate yaml for doing basically anything.


> Especially if you use something like k3s with SQL database

I'll admit I've never used K8s outside of work very much, so I can't really argue on that!

> What were you thinking of things to take into account for k8s over swarm? The main difference for me is k8s needs a hell of a lot more boilerplate yaml for doing basically anything.

I think that's a big one, yes. Stateful services (i.e. volumes) are also much easier to setup and understand with Docker Swarm - which is the same as Compose. The routing mesh[1] is also lovely. I didn't use the Kubernetes routing mesh at work because the infrastructure department didn't allow us to, why is one a reason I was arguing against it; we used a very powerful and complex system without profiting from one of its most powerful feature.

[1]: https://docs.docker.com/engine/swarm/ingress/


We are talking about normal users.


Starting one wiki/note-taking software is not going to make your fans go crazy more than running your favorite JetBrains IDE. I had a non-Arm MacBook Pro and sometimes the fans would go crazy for no reasons, so all my Arm-based colleagues who had an M1 were laughing.


I do agree we are exposing way too many low level details to users these days. Probably because we expect an expert to be setting up these network services. The dream would be to have some low power appliance people can just plug in to provide a data persistence service for applications. Then applications just use that (discovered via zeroconf/avahi) for their "sharing" needs. Everything else should be bundled into the app and invisible to users just like it used to be.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: