There are another important reason than most of the issues pointer out here that docker solves.
Security.
By using containerization Docker effectively creates another important barrier which is incredibly hard to escape, which is the OS (container)
If one server is running multiple Docker containers, a vulnerability in one system does not expose the others. This is a huge security improvement. Now the attacker needs to breach both the application and then break out of a container in order to directly access other parts of the host.
Also if the Docker images are big then the dev needs to select another image. You can easily have around 100MB containers now. With the "distroless" containers it is maybe down to like 30 MB if I recall correctly. Far from 1GB.
Reproducability is also huge efficiency booster. "Here run these this command and it will work perfecty on your machine" And it actually does.
It also reliably allows the opportunity to have self-healing servers, which means businesses can actually not have people available 24/7.
The use of containerization is maybe one of the greatest marvels in software dev in recent (10+) years.
Isn't Docker massively insecure when compared to the likes of Podman, since Docker has to run as a root daemon?
I don't have in-depth knowledge of the differences and how big that is. So take the following with a grain of salt.
My main point is that using containerization is a huge security improvement. Podman seems to be even more secure. Calling Docker massively insecure makes it seem like something we should avoid, which takes focus away from the enormous security benefit containerization gives. I believe Docker is fine, but I do use Podman myself, but that is only because Podman desktop is free, and Docker files seem to run fine with Podman.
Edit: After reading a bit I am more convinced that the Podman way of handling it is superior, and that the improvement is big enough to recommend it over Docker in most cases.
Oof. I'm anxious that folks are going to get the wrong idea here.
While OCI does provide security benefits, it is not a part of a healthly security architecture.
If you see containers advertised on a security architecture diagram, be alarmed.
If a malicious user gets terminal access inside a container, it is nice that there's a decent chance that they won't get further.
But OCI was not designed to prevent malicious actors from escaping containers.
It is not safe to assume that a malicious actor inside a container will be unable to break out.
Don't get me wrong, your point stands: Security loves it when we use containers.
I just wish folks would stop treating containers as "load bearing" in their security plans.
This
I've tried making this argument before and people never seem to agree. I think Google claims their Kubernetes is actually more secure than traditional VMs, but how true that really is I have no idea. Unfortunately though there are already things we depend upon for security that are probably less secure than most container platforms, like ordinary unix permissions or technologies like AppArmour and SELinux.
Sounds like an ugly retrofit of bsd jail
You don't have to ship a second OS just to containerize your app.
Containerized software is huge in the sciences for reproducible research. Or at least it will/should be (speaking as someone adjacent to bioinformatics and genomics)
Always someone who needs to explain and ruin the joke..
Not everyone is experienced in the space. I appreciate the reader notes.
Based on many of the other comments, I don't think most people understood the joke.
Call me crusty, old-fart, unwilling to embrace change... but docker has always felt like a cop-out to me as a dev. Figure out what breaks and fix it so your app is more robust, stop being lazy.
I pretty much refuse to install any app which only ships as a docker install.
No need to reply to this, you don't have to agree and I know the battle has been already lost. I don't care. Hmmph.
Docker is more than a cop out for than one use case. It's a way for quickly deploy an app irrespective of an environment, so you can scale and rebuild quickly. It fixes a problem that used to be solved by VMs, so in that way it's more efficient.
If this is your take your exposure has been pretty limited. While I agree some devs take it to the extreme, Docker is not a cop out. It (and similar containerization platforms) are invaluable tools.
Using devcontainers (Docker containers in the IDE, basically) I'm able to get my team developing in a consistent environment in mere minutes, without needing to bother IT.
Using Docker orchestration I'm able to do a lot in prod, such as automatic scaling, continuous deployment with automated testing, and in worst case near instantaneous reverts to a previously good state.
And that's just how I use it as a dev.
As self hosting enthusiast I can deploy new OSS projects without stepping through a lengthy install guide listing various obscure requirements, and if I did want to skip the container (which I've only done a few things) I can simply read the Dockerfile to figure out what I need to do instead of hoping the install guide covers all the bases.
And if I need to migrate to a new host? A few DNS updates and SCP/rsync later and I'm done.
I've been really trying to push for more usage of dev containers at my org. I deal with so much hassle helping people install dependencies and deal with bizarre environment issues. And then doing it all over again every time there is turnover or someone gets a new laptop. We're an Ops team though so it's a real struggle to add the additional complexity of running and troubleshooting containers on top of mostly new dev concepts anyway.
You know, all this talk about these benefits... when PHP has had this for ages, no BS needed.
I'll see myself out.
I hate that it puts package management in Devs hands. The same Devs that usually want root access to run their application and donβt know a vulnerability scan for the life of them. So now rather than having the one up to date version of a package on my system I may have 3 different old ones with differing vulnerabilities and devs that donβt want to change it because βI need this version because it works!β
I'd argue that's just a ripple effect of being a bad dev, not necessarily the tools fault, but I do get where you are coming from. But also vulnerabilities in some package in a container would be isolated to that container without a further exploit chain
Docker or containers in general provide isolation too, not just declarative image generation, it's all neatly packaged into one tool that isn't that heavy on the system either, it's not a cop out at all.
If I could choose, not for laziness, but for reproducibility and compatibility, I would only package software in 3 formats:
- Nix package
- Container image
- Flatpak
The rest of the native packaging formats are all good in their own way, but not as good. Some may have specific use cased that make them best like Appimage, soooo result...
Yeah, no universal packaging format yetI love docker, it of course comes with some inefficiencies, but let's be real, getting an app to run on every possible environment with any possible other app or configuration or.... that could interfere with yours in some way is hell.
In an ideal world, something like docker is indeed not needed, but the past decades have proven beyond a doubt that alas, we don't live in this utopia. So something like docker that just sets up a private environment for the app so that nothing else can interfere with it.... why not? Anything i've got running on docker is just so stable. I never have to worry that any change i do might affect those apps. Updating them is automated, ...
Not wasting my and the developers time in exchange for a bit of computer resources, sounds like a good deal. If we find a better way for apps to be able to run on any environment, that would of course be even better, but we haven't, so docker it is :).
Fair enough... I admit I'm a bit of an old curmudgeon, set in my ways. :s
I'm not a dev exactly. But I got my Linux skills using Slackware and I still have no problem compiling something if there is no package for it. In some cases I will use a appimage(Cura) but for the most part I just install natively. I use ubuntu but always start eliminating snaps on any install and it really doesn't take that long.
It's about predictable troubleshooting for a bug, not about whether you can install it. No doubt you can, but now the dev has to figure out what particular feature in your OS is causing the issue.
I had this recently, installed Distrobox, which is just a set of scripts, on Aurora. Could not --clone a container, no how. Blow the OS out and install Fedora 41 which is what Aurora is derived from except it's rpm-ostree, and not a problem cloning a Distrobox. Closed the bug as there was no point trying to figure out what went on there for some weird edge case of using a specific distro.
I agree that it's a "cop-out", but the issue it mitigates is not an individual one but a systemic one. We've made it very, very difficult for apps not to rely on environmental conditions that are effectively impossible to control without VMs or containerization. That's bad, but it's not fixable by asking all app developers to make their apps work in every platform and environment, because that's a Herculean task even for a single program. (Just look at all the compatibility work in a codebase that really does work everywhere, such as vim.)
You ever notice how most docker images are usually based from Ubuntu, the arguably worse distro to use for dependency management.
The other core issue is people using docker as a configuration solution with stuff like compose.
If I want containers, I usually just use LXC.
Only docker project I liked was docker-osx which made spinning up OSX VMs easy, but again it was basically 80% configuration for libvirt.
It eliminates the dependency of specific distributions problem and, maybe more importantly, it solves the dependency of specific distribution versions problem (i.e. working fine now but might not work at all later in the very same distribution because some libraries are missing or default configuration is different).
For example, one of the games I have in my GOG library is over 10 years old and has a native Linux binary, which won't work in a modern Debian-based distro by default because some of the libraries it requires aren't installed (meanwhile, the Windows binary will work just fine with Wine). It would be kinda deluded to expect the devs would keep on updating the Linux native distro (or even the Windows one) for over a decade, whilst if it had been released as a Docker app, that would not be a problem.
So yeah, stuff like Docker does have a reasonable justification when it comes to isolating from some external dependencies which the application devs have no control over, especially when it comes to future-proofing your app: the Docker API itself needs to remain backwards compatible, but there is no requirement that the Linux distros are backwards compatible (something which would be much harder to guarantee).
Mind you, Docker and similar is a bit of a hack to solve a systemic (cultural even) problem in software development which is that devs don't really do proper dependency management and just throw in everything and the kitchen sink in terms of external libraries (which then depend on external libraries which in turn depend on more external libraries) into the simplest of apps, but that's a broader software development culture problem and most of present day developers only ever learned the "find some library that does what you need and add it to the list of dependencies of your build tool" way of programming.
I would love it if we solved what's essentially the core Technical Architecture problem of in present day software development practices, but I have no idea how we can do so, hence the "hack" of things like Docker of pretty much including the whole runtime environment (funnilly enough, a variant of the old way of having your apps build statically with every dependency) to work around it.
I don't refuse to install dockerized software - but my system does. While for some people this might be unthinkable, not everyone runs Linux or some proprietary shit. There are many reasons to be unhappy with the trend.
Now if only Docker could solve the "hey I'm caching a layer that I think didn't change" (Narrator: it did) problem, that even setting the "don't fucking cache" flag doesn't always work. So many debug issues come up when devs don't realize this and they're like, "but I changed the file, and the change doesn't work!"
docker system prune -aand beat that SSD into submission until it dies, alas.My favourite is when docker refuses to build unless u enable the no cache options. It claims no disk space (it has plenty, might be some issue with qubes and dynamic disk sizing). I set up a network cache to cache packages at the network level its saved me years in build time.
This is a true story
I had some numpty telling me that installing an application on whatever dog's breakfast distro someone happened to put on an LXC was functionally the same as a shipped docker container for troubleshooting.
Lol π
