Rocket vs. Docker and the Myth of the “Simple, Lightweight Enterprise Platform”
blog.xebialabs.com
blog.xebialabs.com
As soon as you hit an environment where there are > 20 people using the system you need a plethora of tools and systems involved which remove the terms, "simple" and "lightweight," from the specification.
Access control, authentication, auditing, operations... you can't eschew complexity by ignoring it.
I've learned this time and again. If you start out with "simple" and "lightweight" you're going to hit that "enterprise wall" and really pay the price of not considering those requirements up-front. The price of either re-writes or layers of out-of-band processes and tools for orchestrating light-weight systems that were not designed to be composed in such a way. It's quite painful and terrifyingly complicated.
Update: In the case of Docker/Rocket... I think these are just early iterations. I wouldn't be surprised if even Rocket becomes bloated as they realize they cannot design for simplicity when complexity is a requirement.
That was pretty much what I was trying to get at with the question of how a "Rocket ecosystem" would address these requirements. I think vidarh makes a valid point that these could be met by tools outside the container system that are better suited...but I am curious as to what this combination of tools would be.
I think we're finally seeing the first iterations of the operating system disappearing into the network... an idea that has been touted for some decades.
The Rocket ecosystem will be necessarily complex.
Of course Docker in an enterprise will never be as simple as running a Rails app on your Macbook, but it provides an encapsulation primitive that looks as powerful for deployment as object orientation and functional programming were for development.
SICP lecture #1: we manage complexity through means of abstraction and means of combination. Docker is a means of abstraction, and it leaks less than the others we have so far (Chef cookbooks, etc.) We haven't quite figured out what the means of combination is. It won't be trivial - writing programs certainly isn't - but it will be better than handrolled assembly.
Definitely. And I think if we can move more to discussions along the lines of "how do we best abstract/encapsulate/handle the inevitable complexity," rather than "wow, this technology looks like it will solve all our problems," that would be great.
I know the latter sounds absurdly simplistic, but I still hear an awful lot of it around.
Who is propagating this myth of the "Simple, Lightweight Enterprise Platform" to begin with? Some searches did not net me any references to that other than this blog post.
I was referring mainly to this statement: "We should stop talking about Docker containers, and start talking about the Docker Platform. It is not becoming the simple composable building block we had envisioned."
I agree with you that that doesn't necessarily mean that the Rocket team disputes the need for enterprise features, but (as you say) may think other technologies are better suited. The announcement itself doesn't go into much detail as to which technologies those would be, though.
> Who is propagating this myth of the "Simple, Lightweight Enterprise Platform" to begin with?
It certainly hasn't been associated with this specific phrase, but the idea that there is a New Cool Tech around the corner that will be simpler, easier and less heavyweight than the Last Cool Tech seems to be a recurring theme in our industry, in my experience.
That's not to say that new technologies don't introduce dramatic improvements in certain areas. Personally, I'm a big fan of microservices and, specifically, containers as a possible implementation mechanism.
Equally, I don't think it's surprising that technologies such as Docker that start out as 'simple, composable building blocks' become part of more complex, "heavyweight" solutions as they evolve.
That's part of the point. They want to be free to compose components depending on their customers needs.
CoreOS already includes Fleet (orchestration of individual units), and their are putting effort into Flannel (overlay network), and Kubernetes (orchestration of "pods" of containers).
But their market is enterprises that in many cases will want to fit CoreOS into an existing enterprise platform.
Nothing in how CoreOS is acting has shown any indication that they think enterprise is "simple". On the contrary, Rocket, while smaller in scope than Docker, if anything is more complex (with the signing, composability of filesets, and implied support for running systemd in the containers).
I'm not convinced Rocket is the right thing - they'll have to show me tooling that makes it as simple as Docker first - but some of the decisions appeals to me (a clear spec for the format, that alternative runtimes can conform to; e.g. nothing appears to prevent someone from making a Rocket runtime ("ACE") that uses KVM or Virtualbox or Xen, and hence it could be potentially supported by things like OpenStack, or even shoehorned into a VMware deployment; making things like whether or not an overlay filesystem is used an implementation detail; and while I won't need it often, and it may be an annoyance in some cases, support for signed images).
So I just see nothing to imply that CoreOS is a good exponent for some idea of "simple, lightweight enterprise platform". They are providing a set of tools that individually are simple and lightweight, that you can use to start building the lower levels of an enterprise platform with, though, and understandably don't want one of those tools to grow into a platform itself (whether or not Docker will grow into that in a way that doesn't allow the core container execution mechanism to be easily lifted out, is a separate discussion).
I'd say that's completely valid. I was not trying to imply that the combination of CoreOS and tools such as Fleet, Flannel, Kubernetes etc. claimed to be a "simple, lightweight platform". Indeed, as you point out, the core component, i.e. Rocket itself, may well be more complex than "Docker classic."
I think your comments demonstrate accurately that the overall "Rocket ecosystem" will also address the inevitable enterprise requirements. Since most of the other tools you mention did not appear in the launch announcement, I think it's fair to say that describing Rocket as a response to the emerging 'non-simple Docker Platform' is, at the very least, inviting a comparison that isn't exactly apples to apples.
The overall point I was trying to make goes beyond Rocket, Docker, containers or any specific current technology or tool, however. The way the Rocket announcement was worded simply prompted on ongoing train of thought. Indeed, it's the many conversations in the past months around Docker that I've been involved in that have exemplified the belief/hope/wish that there finally is a new tool that will make boring enterprise problems magically "go away."
Docker, Rocket, and whatever tomorrow's favorite container framework turns out to be, are unlikely to be the last examples of that pipe dream, I think.
The Docker simplicity is why they've had so much success. That's how it started: as a tool to make using LXC simpler.
Do you feel Machine, Swarm and Compose add to or detract from the value proposition of Docker, in that case?
Just to ensure I'm understanding your point about Rocket benefiting from 40 years of Unix evolution correctly: do you think that Rocket will be able to remain simple as a result of this? And that Docker perhaps will not, even though up until this point it's been "trying to do things the simple way?"
Skimming through it, yes whatever you define as "enterprise" can fit your definition and whatever claim you are doing in the article or blog, can be fulfilled.
Sorry about that, antocv. I've spoken to our web team about the colour scheme a few times already.
Thanks to agumonkey for the link!