Hacker News Reader

Comment by justinsb | Hacker News Reader

I would agree that Docker has some flawed fundamentals, that it has many implementation flaws, and that is over-hyped. Out-of-the-box it is even almost useless for real-world production usage.

But - despite agreeing with your words - I wouldn't put them together to reach the same conclusion. The container model with managed images seems to be a big step forwards vs the VM model: it encourages having immutable 'machines' that are disposable, which works very well for cloud architectures.

I would check out the kubernetes project and where it is heading: containers orchestrated across a cluster of machines, with seamless/automatic networking between the containers.

View on HN

Thanks for the feedback, I'll check out kubernetes. Although I agree that containerisation is having an impact in terms of encouraging immutable infrastructure, I would argue that this is easily achieved using virtualisation.

Other people have raised a similar argument of Docker ecosystem, which is a fair point and I'll be amending the article to reflect this.

Immutable architecture seems to be extremely hard to achieve with virtualization, while there are numerous success stories doing this with Docker.

I would have to strongly disagree with this, not just because AWS has been capable of doing this for quite some time, but also because of my own success stories. I'm due to post a follow up article, explaining the pros/cons of the alternative solutions, within a month or so.

Well you could have managed images in a VM, many do that, incl. ppl just using ec2. you could also do that using lxc, or what not, not necessarily with docker.

So basically, ocker adds nothing to these other solutions except complication, bugs, etc. (which is also my experience).

Moreover, and to push it a little further, if anything, it's a better idea to call some container library so that you use namespace for your very use case, inside a VM. That VM being an image that you manage and deploy of course (ie that you dont tweak when its running unless its for debugging)

For me the biggest win with Docker is how composable it makes containers. Building new images only takes moments because of how fast containers can be downloaded, started and modified.

It's certainly possible to do all of the same things with virtual machines, but the difference is the process is really cumbersome. No one has made an equivalent service of Docker Hub for virtual machine images and snapshot deltas.

Exactly this. Also in my experience Docker wraps LXC much more nicely then raw LXC, which on my current work machine with Ubuntu 14.04 doesn't work at all (whereas I've got a great chain of Docker images set up for web dev right now which gives me a complete overview of what I'm building).

Although I haven't yet reviewed LXD, I agree that plain LXC is an absolute pita to use, at least it was in 2012 (the last time I used it in prod, aside from Docker)

Docker doesn't wrap LXC any more; they re-implemented that stuff in libcontainer. Well, the code to call LXC is in there somewhere as a backwards-compatibility option, but you won't be using it unless you really try.

Good points. I failed to mention marketplace AMIs which, despite my disliking of AWS, is an equivalent ecosystem for sharing images.

Orchestrating things across a cluster of machines with seamless networking isn't new from kube and the docker ecosystem. Check out mesos some time, which has enabled that kind of thing for quite a while.

Well, if you want to play that game, you can look further back to HTCondor which has been doing this since 1989. I'm sure earlier examples exist as well, given that the concept itself isn't all that esoteric.

Interesting, I hadn't heard of HTCondor before but have recently become more interested in bioinformatics, which seems to use this quite a bit (at least in Cambridge). I'll have a read, thanks for the heads up :)

Reply on news.ycombinator.com