Back in

[Dirk Hohndel]

On Tue, Sep 16, 2014 at 08:53:09PM +0200, Anton Lundin wrote:
> 
> Docker is as high tech as a sledgehammer compared to Chef or Puppet.
> They kinda aim for different targets.
> 
> Docker is probably a way simpler way to go for you.

Yes. Chef and Puppet aim at automating large data centers. I am running a
few services on a single server (with maybe a vague idea to have a
failover in a different geo at some point).

Docker helps me to do some separation of services and to make the services
easier to reproduce if things fall down.

> > The biggest challenge with Docker is that it's not really designed for the
> > type of services I'm running... you cannot really do the "one app one
> > container" thing Docker wants you to do. Trac requires a web server, a git
> > server, a mail server, and it's entirely non-trivial and
> > counter-productive to spread those out across multiple containers - at
> > least as far as I can tell...
> 
> You can do some trickery with volume containers and intra container
> routing, but its probably simpler for you to just lump a slew of
> services into one container.
> 
> I would recommend you at least use volumes for data, and containers for
> code.

I have a container for MySQL.
A ton of other data actually comes as volumes from the 'real' host. This
makes it so much easier to back things up and to be able to quickly tear a
container down, fix something and bring it back up.

> > So I'm using Docker/baseimage to run multiple services in one container
> > and basically use Docker as a set of tools to be able to encapsulate
> > larger logical blocks. E.g. the MySQL server is its own container. As is
> > the WordPress site (that one had been hacked before). I'm still in the
> > experimentation phase regarding the separation of the other services -
> > especially the trac/git server will likely be one single container...
> > 
> > THis means that multiple containers will be running apache and there needs
> > to be a reverse proxy in front of that (also apache) which means that I
> > have a lot of independent apache processes running. I'll have to monitor
> > how much that increases system resource load. I did switch to a 16 core
> > Xeon server with 24GB of memory, so this should be big enough for a few
> > years (famous last words).
> > 
> 
> 16 cores and 24GB ram is a huge machine for such a task. You could have
> used VM's for everything and still had ram left with such a machine.

I know. This machine was much cheaper than a more reasonably sized one :-)

> ( And you don't need to run a apache for each of them. You could as
> easily just run tracd in your container and let apache proxy to that )

That I'm still figuring out. Trac is a pain in the rear...

> > My biggest problem is time. I just don't have enough. This day job keeps
> > distracting me from working on Subsurface infrastructure :-)
> 
> Pity that someone have nabbed https://github.com/subsurface , that could
> be a simple place to just use as our infrastructure.

I HATE the github infrastructure. What a pile of... no, let's not go
there. I have control issues. As long as I maintain Subsurface it will run
on my own servers. It's that simple.

/D