Docker, boot2docker and DNS resolution of containers

… or how to setup a full development environment with docker (on Mac OS X)


This tutorial should help to setup a development environment on Mac OS X that relies heavily on docker and boot2docker. Based on my experience, docker is an excellent tool to create and run applications very effectively during dev, test and production. However, not having great development environment where docker can be used effectively slows you down and requires a lot of hand-tuning each an every time during the development process.

Part of this tutorial is based on information from other blogs, such as iSpyker and SkyDock. Thanks for all the great work guys!

Goal of this tutorial

The goal is to have the following capabilities available after you complete this guide.

  1. Run docker containers from your dev machine and have seamless access to all container using their IPs.
  2. Enable DNS capabilities for all containers and being able to resolve them from your dev machine.
  3. Ensure that this will also work if you have to use a corporate VPN client (like Cisco AnyConnect).
  4. Being able to quickly set this up and tear it down with a simple script. Even though I provide a script (at the end), I highly suggest to go through this manually to understand the details.


You need to have the following software installed.

  • Mac OS X (this guide is not required for Linux because it can run docker nativly)
  • boot2docker v1.1.1 (Note: I run the boot2docker ISO from here because I allows me to mount /Users in the containers which is great for testing code that I’m working on)
  • docker v1.1.1
  • git for checking about the git repo with the scripts

What is the current problem?

If you start a docker container on MacOS, it will actually be running inside a VirtualBox VM that boot2docker is setting up for you under the cover. It’s called boot2docker-vm and you can see the VM when you start the VirtualBox app. This VM actually runs the docker agent and your docker CLI on your Mac connects via the $DOCKER_HOST variable to your docker agent within the VM. The consequence is that each docker container cannot be pinged or resolved from the Mac which makes it hard to use during development. To illustrate the problem look as this simple example.

Start a docker container and run a server with nc

florian@mac:~$ docker run -i -t ubuntu bash
root@4cb806d8ddab:/# ifconfig eth0
eth0      Link encap:Ethernet  HWaddr be:70:6a:45:22:ee  
          inet addr:  Bcast:  Mask:
          inet6 addr: fe80::bc70:6aff:fe45:22ee/64 Scope:Link
          UP BROADCAST RUNNING  MTU:1500  Metric:1
          RX packets:8 errors:0 dropped:0 overruns:0 frame:0
          TX packets:8 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:1000 
          RX bytes:648 (648.0 B)  TX bytes:648 (648.0 B)
root@4cb806d8ddab:/# nc -l 7777

Try to connect the server

florian@mac:~$ telnet 7777
telnet: connect to address Operation timed out
telnet: Unable to connect to remote host

It will timeout because the IP cannot be resolved. So let’s fix that issues …

Before we jump to the solution you may already asked yourself the question “why is he not using port forwarding rules from VirtualBox?”. They simply arent’ enough if you using lots of containers and a highly dynamic system (e.g., based on microservices). For one or two containers, such as a mongo and simple redis service that you app is requiring, port forwarding is totally fine.

Setting up boot2docker

If you have boot2docker already running, your VirtualBox (the hypervisor that is used by boot2docker) VM instance boot2docker-vm has been created already.

Initialize the boot2docker-vm

$ boot2docker init --dhcp=false --hostip=

This will initialize a boot2docker-vm and also create a host-only network with the above mention IP and no DHCP for that network interface (we won’t need it).

If you already have a boot2docker-vm running, you have two choices: (a) destroy (boot2docker destroy) and re-create with the above command. This is recommended if you don’t have a lot of important containers running; (b) change the existing boot2docker-vm and manually add the host-only interface via VirtualBox (it’s pretty easy, you’ll figure it out).

Start boot2docker-vm

$ boot2docker up

This will bring up the boot2docker-vm. It will succeed but if you study the output below carefully, it complains about the missing IP address that could not be assigned to the host-only interfaces (this was done on purpose).

$ boot2docker up
2014/08/14 22:13:39 Waiting for VM to be started...
2014/08/14 22:14:18 Started.
2014/08/14 22:14:19   Trying to get IP one more time
2014/08/14 22:14:19 Auto detection of the VM's IP address failed.
2014/08/14 22:14:19 Please run `boot2docker -v up` to diagnose.

We will configure the IP in the next step.

Configure host-only network adapter

Run the following command to assign an IP and netmask to the newly configured host-only adapter eth1.

$ boot2docker ssh "sudo ifconfig eth1 netmask"

Let’s ensure you can ping eth1 from your Mac:

$ ping -c1 
PING ( 56 data bytes
64 bytes from icmp_seq=0 ttl=64 time=8.704 ms

--- ping statistics ---
1 packets transmitted, 1 packets received, 0.0% packet loss
round-trip min/avg/max/stddev = 8.704/8.704/8.704/0.000 ms

Setting up a route

So far you can ping the host-only adapter, however, you cannot ping the docker containers as we saw earlier. You need to setup a route for this.

$ sudo route -n add

This will ensure that all contains which get a 172.17/16 address can be resolved via the host-only adapter. Let’s make sure the route was added correctly:

$ netstat -nr |grep 172.17
172.17           UGSc            0        0 vboxnet

Look’s good so we can go back that test the initial problem we were going after.

Test if you can reach a container

Let’s start a container back up and run nc:

$ docker run -i -t ubuntu bash
root@52eee3e2c7f6:/# ifconfig eth0 | grep addr
eth0      Link encap:Ethernet  HWaddr 16:85:bf:22:57:fa  
          inet addr:  Bcast:  Mask:
          inet6 addr: fe80::1485:bfff:fe22:57fa/64 Scope:Link
root@52eee3e2c7f6:/# nc -l 7777

Let’s make sure we can connect to “service” offered by the container:

$ telnet 7777
Connected to
Escape character is '^]'.

Voila, we are connected. Type something in the telnet session and you’ll see it on the other side. Now we can move on to getting DNS to work.

Setup DNS for Docker

Now that we are able to ping docker containers seamlessly from our Mac, it would be great to have DNS names automatically registered whenever a docker container comes up. There are two related projects out there called SkyDock and SkyDNS that fill this gap. Setting them up is fairly easy. I won’t go into the details of the project and the internals because Michael Crosby, the founder of the project does a much better job.

In a nutshell, you need to start two services, SkyDock and SkyDNS, which are both provided as pre-built containers. SkyDNS is a classical DNS server and SkyDock is listening at the docker host for container being started and stopped, keeps track of them and creates the DNS entries in SkyDock.

To get started, we need to “tune” the docker agent commands in the boot2docker-vm a bit. Let’s kill the existing docker daemon running inside the boot2docker-vm.

Changing the arguments of the docker daemon

$ boot2docker ssh 
docker@boot2docker:~$ sudo pkill bin/docker

Now restart the docker daemon with the following options:

docker@boot2docker:~$ sudo /usr/local/bin/docker -d -g /var/lib/docker -H unix:// -H tcp:// --bip= --dns= &

The --bip flag points to the IP address of the docker0 bridge. You can find out by running ifconfig docker0 |grep addr inside boot2docker. The DNS server will bind to that IP address. The --dns flag will tell the docker agent that each container that is launched should have as the DNS server in /etc/resolv.conf.

Launching SkyDNS and SkyDock

Now we are ready to launch the two containers. First we launch SkyDNS, then SkyDock. The first time this may take a while because docker will have to pull both containers from the public registry.

$ docker run -d -p --name skydns crosbymichael/skydns -nameserver -domain docker
$ docker run -d -v /var/run/docker.sock:/docker.sock --name skydock crosbymichael/skydock -ttl 30 -environment dev -s /docker.sock -domain docker -name skydns

The SkyDock container has two interesting arguments: --domain is the name of the “fake” domain that all hosts will get. The --environment is a prefix to the domain name. We will see an example later. Ensure now that both containers where started and check the output:

$ docker ps
CONTAINER ID        IMAGE                          COMMAND                CREATED             STATUS              PORTS                              NAMES
f0eeaff0a54e        crosbymichael/skydock:latest   /go/bin/skydock -ttl   6 minutes ago       Up 6 minutes                                           skydock             
cff583466d45        crosbymichael/skydns:latest    skydns -http   10 minutes ago      Up 10 minutes       8080/tcp,>53/udp   skydns              

Testing the DNS server

We should now be able to get results from the DNS server. Let’s start a docker container as follows:

$ docker run -i -t --name u1 ubuntu bash

With SkyDock, the general rule for a DNS name is as follows:


So in this case the DNS name is Let’s query the result with dig:

$ dig @ +short

If we start another container docker run -i -t --name u2 ubuntu bash and the use dig to query, we can get a list of all ubuntu containers as follows:

$ dig @ +short

Update /etc/resolv.conf

If you want seamless DNS resolution while developing, you need to enter the DNS server as the first in /etc/resolv.conf. Something like this:

$ cat /etc/resolv.conf 
# Mac OS X Notice
# This file is not used by the host name and address resolution
# or the DNS query routing mechanisms used by most processes on
# this Mac OS X system.
# This file is automatically generated.
domain home

That’s it, you can now enjoy full DNS-enabled docker containers. If you get tired of manually adding/removing the DNS server the /etc/resolv.conf after you stop the SkyDock container, you can create “Network Locations” in OS X. Go to System Menu (top left Apple symbol) -> Location and configure two profiles, one with SkyDNS server one without. Then you can easily switch.

Connecting to VPN will break things

If you have to use Cisco VPN to connect to your corporate network, you will soon realize that once you do that, pinging the docker containers will no longer work nor will DNS. I don’t know why but it seems that VPN is doing some magic with firewalls.

Try connecting to VPN and test yourself. If the previous dig query still works you are fine and you can skip this step. If you are getting a message like ;; connection timed out; no servers could be reached you are in trouble.

The fix that works for me is to find a firewall rule that blocks all traffic (from any to any):

$ sudo ipfw -a list | grep "deny ip from any to any"
00411       94       10039 deny ip from any to any

Now try to delete this rule:

$ sudo ipfw delete 00411

And now test if your dig query works again. Mine does :). Keep in mind that you have to redo this step every time you reconnect your VPN. The git repository below also has a script that automates that process as well.

A simple script that ties all this together

I provided two scripts in my git repo. There are some instructions in the readme. In a nutshell, the implements the whole procedure described in this blog. When using those scripts, be aware that they are tested only in my environment and work for me. If you have problem please file and issue in github as there are many corner cases that I may not handle.

The second script is the script that I run every time I reconnect to my corporate VPN. It cleans out some conflicting rules.