Hi! Im new to self hosting. Currently i am running a Jellyfin server on an old laptop. I am very curious to host other things in the future like immich or other services. I see a lot of mention of a program called docker.
search this on The internet I am still Not very clear what it does.
Could someone explain this to me like im stupid? What does it do and why would I need it?
Also what are other services that might be interesting to self host in The future?
Many thanks!
EDIT: Wow! thanks for all the detailed and super quick replies! I’ve been reading all the comments here and am concluding that (even though I am currently running only one service) it might be interesting to start using Docker to run all (future) services seperately on the server!
A program isn’t just a program: in order to work properly, the context in which it runs — system libraries, configuration files, other programs it might need to help it such as databases or web servers, etc. — needs to be correct. Getting that stuff figured out well enough that end users can easily get it working on random different Linux distributions with arbitrary other software installed is hard, so developers eventually resorted to getting it working on their one (virtual) machine and then just (virtually) shipping that whole machine.
Docker is not a virtual machine, it’s a fancy wrapper around chroot
I’m aware of that, but OP requested “explain like I’m stupid” so I omitted that detail.
No, chroot is kind of its own thing
It is just a kernel namespace
Yes, technically chroot and jails are wrappers around kernel namespaces / cgroups and so is docker.
But containers were born in a post chroot era as an attempt at making the same functionality much more user friendly and focused more on bundling cgroups and namespaces into a single superset, where chroot on its own is only namespaces. This is super visible in early docker where you could not individually dial those settings. It’s still a useful way to explain containers in general in the sense that comparing two similar things helps you define both of them.
Also cgroups have evolved alongside containers at this point and work rather differently now compared to 18 years ago when cgroups were invented and this differentiation mattered more than now. We’re at the point where differentiation between VMs and Containers is getting really hard since both more and more often rely on the same kernel features that were developed in recent years on top of cgroups
But why can I “just install a program” on my windows machine or on my phone and it is that easy?
You might notice that your Windows installation is like 30 gigabytes and there is a huge folder somewhere in the system path called WinSXS. Microsoft bends over backwards to provide you with basically all the versions of all the shared libs ever, resulting in a system that can run programs compiled from decades ago just fine.
In Linux-land usually we just recompile all of the software from source. Sometimes it breaks because Glibc changed something. Or sometimes it breaks because (extremely rare) the kernel broke something. Linus considers breaking the userspace API one of the biggest no-nos in kernel development.
Even so, depending on what you’re doing you can have a really old binary run on your Linux computer if the conditions are right. Windows just makes that surface area of “conditions being right” much larger.
As for your phone, all the apps that get built and run for it must target some kind of specific API version (the amount of stuff you’re allowed to do is much more constrained). Android and iOS both basically provide compatibility for that stuff in a similar way that Windows does, but the story is much less chaotic than on Linux and Windows (and even macOS) where your phone app is not allowed to do that much, by comparison.
That’s just incorrect. Apart from 3 guys who have no better things to do no one in “Linux-land” does that.
In case of phones, there’s less of a myriad of operating systems and libraries.
A typical Android app is (eventually) Java with some bundled dependencies and ties in to known system endpoints (for stuff like notifications and rendering graphics).
For windows these installers are usually responsible for getting the dependencies. Which is why some installers are enormous (and most installers of that size are web installers, so it looks smaller).
Docker is more aimed at developers and server deployment, you don’t usually use docker for desktop applications. This is the area where you want to skip inconsistencies between environments, especially if these are hard to debug.
Caveat: I am not a programmer, just an enthusiast. Windows programs typically package all of the dependency libraries up with each individual program in the form of DLLs (dynamic link library). If two programs both require the same dependency they just both have a local copy in their directory.
Beat me to it.
So instead of having problems getting the fucking program to run, you have problems getting docker to properly build/run when you need it to.
At work, I have one program that fails to build an image because of a 3rd party package who forgot to update their pgp signature; one that builds and runs, but for some reason gives a 404 error when I try to access it on localhost; one that whoever the fuck made it literally never ran it, because the
Dockerfilewas missing some 7 packages in the apt install line.There are two ends here, as a user and as a developer. As a user Docker images just work, so you solve almost every problem you’re having which would be your users having them and giving up on using your software.
Then as a developer docker can get complicated, because you need to build a “system” from scratch to run your program. If you’re using an unstable 3d party package or missing packages it means that those problems would be happening in the deploy servers instead of your local machines, and each server would have its own set of problems due to which packages they didn’t have or had the wrong version, and in fixing that for your service you might be breaking other service already running there.
Yeah, it’s another layer, and so there definitely is an https://xkcd.com/927/ aspect to it… but (at least in theory) only having problems getting Docker (1 program) to run is better than having problems getting N problems to run, right?
(I’m pretty ambivalent about Docker myself, BTW.)
Building from source is always going to come with complications. That’s why most people don’t do it. A docker compose file that ‘just’ downloads the stable release from a repo and starts running is dramatically more simple than cross-referencing all your services to make sure there are no dependency conflicts.
There’s an added layer of complexity under the hood to simplify the common use case.