Docker DNS-gen

dns-gen sets up a container running Dnsmasq and docker-gen. docker-gen generates a configuration for Dnsmasq and reloads it when containers are started and stopped.

By default it will provide thoses hosts: containername.docker and servicename.projectfolder.docker pointing to the corresponding container.

Simple usage

First, you have to know the IP of your docker0 interface. It may be 172.17.42.1 but it could be something else. To known your IP, run the following command:

$ /sbin/ifconfig docker0 | grep "inet" | head -n1 | awk '{ print $2}' | cut -d: -f2

Now, you can start the dns-gen container:

$ docker run --detach \
  --name dns-gen \
  --publish 172.17.42.1:53:53/udp \
  --volume /var/run/docker.sock:/var/run/docker.sock \
  jderusse/dns-gen

Last thing: Register you new DnsServer in you resolv.conf

$ echo "nameserver 172.17.42.1" | sudo tee --append /etc/resolvconf/resolv.conf.d/head
$ sudo resolvconf -u

This is it. You can now start your containers and retrieve their IP:

$ docker run --name my_app --detach nginx
$ dig my_app.docker
$ dig sub.my_app.docker

You can customize the DNS name by providing an environment variable, like this: DOMAIN_NAME=subdomain.youdomain.com

$ docker run --env DOMAIN_NAME=foo.docker --detach nginx
$ dig foo.docker
$ dig sub.foo.docker
$ docker run --env DOMAIN_NAME=bar.docker,baz.docker --detach nginx
$ dig bar.docker
$ dig baz.docker

Start the container automatically after reboot

You can tell docker (version >= 1.2) to automatically start the DNS container after booting, by passing the option --restart always to your run command.

$ docker run -d --name dns-gen \
  --restart always \
  --publish 172.17.42.1:53:53/udp \
  --volume /var/run/docker.sock:/var/run/docker.sock \
  jderusse/dns-gen

beware! When your host will restart, it may change the IP address of the docker0 interface. This small change will prevent docker to start your dns-gen container. Indeed, remember our container is configured to forward port 53 to the previous docker0 interface which may not exist after reboot. Your container just will not start, you will have to re-create it. To solve this drawback, force docker to always use the same IP range by editing the default configuration of the docker daemon (sometimes located in /etc/default/docker but may change regarding your distribution). You have to restart the docker service to take the changes into account. Sometimes the interface is not updated, you will have to restart your host.

# For systemd users (Fedora and recent Ubuntu versions) :
$ vim /lib/systemd/system/docker.service
# append the --bip="172.17.42.1/24" option to the ExecStart line
# then
$ sudo systemctl daemon-reload

# For other users
$ vim /etc/default/docker

DOCKER_OPTS="--bip=172.17.42.1/24"

# In any cases
$ sudo service docker restart

One more thing When you start your host, the docker service is not fully loaded. Until this daemon is loaded, the dns container will not be automatically started and you will notice bad performance when your host will try to resolve DNS. The service is not fully loaded, because it uses a feature of systemd called socket activation: The first access to the docker socket will trigger the start of the true service. To skip this feature, you simply have to activate the docker service.

$ sudo update-rc.d docker enable

Et voila, now, docker will really start with your host, it will always use the same range of IP addresses and will always start/restart the container dns-gen.

Advanced usage

The previous method is simple to use, but suffer from drawback:

Containers won't resolve DNS from other containers
External DNS resolution may be slower
Some VPN changes the /etc/resolv.conf file which remove your configuration

Instead of using the docker0 interface and modifying /etc/resolv.conf, an other solution is to install localy a dnsmasq server (some distribs like ubuntu or debian are now using it by default) and forward requests to the dns-gen container and configure containers to use it.

step 1 Configure the local dnsmasq to forward request to 127.0.0.1:54 And listen to interfaces lo and docker0.

$ sudo vim /etc/NetworkManager/dnsmasq.d/01_docker`
bind-interfaces
interface=lo
interface=docker0

server=/docker/127.0.0.1#54

$ sudo systemctl status NetworkManager

step 2 Run dns-gen and bind port 53 to the 54's host

$ docker run --daemon --name dns-gen \
  --restart always \
  --publish 54:53/udp \
  --volume /var/run/docker.sock:/var/run/docker.sock \
  jderusse/dns-gen -R

the option -R just tell dns-gen to not fallback to the default resolver which avoid an infinity loop of resolution

step 3 Configure docker to use the docker0 as DNS server

# For systemd users (Fedora and recent Ubuntu versions) :

$ vim /etc/systemd/system/docker.service

# append
[Service]
ExecStart=
ExecStart=/usr/bin/docker daemon -H fd:// --bip=172.17.42.1/24 --dns=172.17.42.1

# then
$ sudo systemctl daemon-reload

# For other users
$ vim /etc/default/docker
DOCKER_OPTS="--dns=172.17.42.1 --bip=172.17.42.1/24"

# In any cases
$ sudo service docker restart

Thank to this configuration the resolution workflow is now:

the host want to resolve google.com: host -> dnsmasq -> external dns
the host want to resolve foo.docker: host -> dnsmasq -> 127.0.0.1:54 -> dns-gen
a container want to resolve google.com: container -> 172.17.42.1 -> dnsmasq -> external dns
a container want to resolve foo.docker: container -> 172.17.42.1 -> dnsmasq -> 127.0.0.1:54 -> dns-gen

Automate configuration

@jpic writes an ansible playbook to configure the host. Thanks to him.

Troubleshooting

On restart, if you loose the dns resolution, check the NetworkManager service status.

$ service NetworkManager status

If the service is down, check your syslog, grep on dnsmasq:

$ grep dnsmasq /var/log/syslog

If this error is logged:

dnsmasq[12345]: unknow docker0 interface

In this case, the NetworkManager service try to start before the docker service.
dnsmasq can't listen to an interface who is not already defined.

You can fix it by increasing the docker start priority (higher than NetworkManager).