I have recently been working on fixing some personal pain-points for Docker. It came about because of a course I am taking at UConn this fall on Troubleshooting Distributed Systems - Excellent course by the way.

The current issue I’m looking at is the one about logging. Yep, I somehow managed to volunteered to do this in a month with a full-time job and 4 hours a week of driving back and forth from UConn. Anyway, I needed to run a dev version of Docker in my local machine alongside the stable version. I’m on a mac running boot2docker. There are other ways to do this but I find boot2docker sufficient for almost all the work I’d want to do in this case.

Compiling Docker

Basically, follow the instruction in http://docs.docker.com/contributing/devenvironment/. The short form of it is:

git clone git@github.com:docker/docker.git
cd docker
make BINDDIR=. cross

Docker compiles itself in a docker container (of course) and it was a seamless operation aside from the fact that I could not access the binaries after the compilation was done. The problem is that your BINDDIR is not set to the current directory. I have come to run the make command with that environment variable, for tasks other than tests.

You will find the compiled versions in the following directory structure (inside the bundles directory)

bundles
├── 1.3.1
│   └── binary
│       ├── docker -> docker-1.3.1
│       ├── docker-1.3.1
│       ├── docker-1.3.1.md5
│       └── docker-1.3.1.sha256
└── 1.3.1-dev
    ├── binary
    │   ├── docker -> docker-1.3.1-dev
    │   ├── docker-1.3.1-dev
    │   ├── docker-1.3.1-dev.md5
    │   └── docker-1.3.1-dev.sha256
    └── cross
        ├── darwin
        │   ├── 386
        │   │   ├── docker -> docker-1.3.1-dev
        │   │   ├── docker-1.3.1-dev
        │   │   ├── docker-1.3.1-dev.md5
        │   │   └── docker-1.3.1-dev.sha256
        │   └── amd64
        │       ├── docker -> docker-1.3.1-dev
        │       ├── docker-1.3.1-dev
        │       ├── docker-1.3.1-dev.md5
        │       └── docker-1.3.1-dev.sha256
        ├── freebsd
        │   ├── 386
        │   │   ├── docker -> docker-1.3.1-dev
        │   │   ├── docker-1.3.1-dev
        │   │   ├── docker-1.3.1-dev.md5
        │   │   └── docker-1.3.1-dev.sha256
        │   ├── amd64
        │   │   ├── docker -> docker-1.3.1-dev
        │   │   ├── docker-1.3.1-dev
        │   │   ├── docker-1.3.1-dev.md5
        │   │   └── docker-1.3.1-dev.sha256
        │   └── arm
        │       ├── docker -> docker-1.3.1-dev
        │       ├── docker-1.3.1-dev
        │       ├── docker-1.3.1-dev.md5
        │       └── docker-1.3.1-dev.sha256
        └── linux
            ├── 386
            │   ├── docker -> docker-1.3.1-dev
            │   ├── docker-1.3.1-dev
            │   ├── docker-1.3.1-dev.md5
            │   └── docker-1.3.1-dev.sha256
            ├── amd64
            │   ├── docker -> ../../../binary/docker
            │   ├── docker-1.3.1-dev -> ../../../binary/docker-1.3.1-dev
            │   ├── docker-1.3.1-dev.md5 -> ../../../binary/docker-1.3.1-dev.md5
            │   └── docker-1.3.1-dev.sha256 -> ../../../binary/docker-1.3.1-dev.sha256
            └── arm
                ├── docker -> docker-1.3.1-dev
                ├── docker-1.3.1-dev
                ├── docker-1.3.1-dev.md5
                └── docker-1.3.1-dev.sha256

Since I’m using a Macbook Pro - when I try to run ./bundles/1.3.1-dev/cross/darwin/amd64/docker ps I get

2014/11/11 04:17:49 Error response from daemon: client and server don't have same version (client : 1.16, server: 1.15)

This means the versions of client and servers do not match. Now I don’t really want to replace the stable docker instance running in boot2docker.

The solution is slightly hacky but works well enough for what I need.

Move Docker binary to the VM

This would be similar for other systems too - I’m going to assume boot2docker for simplicity again

Find boot2docker ssh port

It’s mostly 2022 but doesn’t hurt to look

$ boot2docker info

{
  "Name": "boot2docker-vm",
  "UUID": "c0bad522-2c24-4a96-bf54-d2c0225be3a2",
  "Iso": "/Users/nextdooerhacker/.boot2docker/boot2docker.iso",
  "State": "running",
  "CPUs": 8,
  "Memory": 2048,
  "VRAM": 8,
  "CfgFile": "/Users/nextdoorhacker/VirtualBox VMs/boot2docker-vm/boot2docker-vm.vbox",
  "BaseFolder": "/Users/nextdoorhacker/VirtualBox VMs/boot2docker-vm",
  "OSType": "",
  "Flag": 0,
  "BootOrder": null,
  "DockerPort": 0,
  "SSHPort": 2022,
  "SerialFile": "/Users/nextdoorhacker/.boot2docker/boot2docker-vm.sock"
}%

If you want to script it, I recommend the excellent jq tool to get the number. There are ways to do this with awk/sed/grep tools too.

bot2docker info | jq '.SSHPort'
2022

Lets save that to a variable

PORT=$(boot2docker info | jq '.SSHPort')

Move the binaries to boot2docker

From your docker directory, the following command:

scp -r -P $PORT bundles/1.3.1-dev/cross/linux/amd64/ docker@localhost:

The password according to https://github.com/boot2docker/boot2docker is tcuser.

SSH into boot2docker

This should drop you into the boot2docker shell

boot2docker ssh

Run the dev docker instance

I picked the port 2378 pretty much arbitrarily here. I wanted to run with just changing the DOCKER_HOST variable. Unless I’m hacking on the TLS part, I shouldn’t have to change that. This is the command I run:

cd amd64
sudo ./docker -d -D -g . -H unix:// -H tcp://0.0.0.0:2378 --tlsverify --tlscacert=/var/lib/boot2docker/tls/ca.pem --tlscert=/var/lib/boot2docker/tls/server.pem --tlskey=/var/lib/boot2docker/tls/serverkey.pem -p docker.pid

Only things that have changed are

• the -D argument for debug mode - I find it useful to have it running. Docker is not that verbose to bother me
• the pidfile to be in the current directory

Using the new client

Now prefix with the new port number – everything else will remain the same.

DOCKER_HOST=tcp://192.168.59.103:2378 ./bundles/1.3.1-dev/cross/darwin/amd64/docker run -t -i ubuntu  /bin/bash
root@c886080a6024:/#

Congrats you have a dev version of docker running. Commence hackery!