Skip to content

Latest commit

 

History

History
326 lines (235 loc) · 12.5 KB

README.md

File metadata and controls

326 lines (235 loc) · 12.5 KB

Local Docker Host DNS - ldhdns

Build Status

A developer tool for providing DNS for Docker containers running on a local Linux development host.

Requirements

How it Works

ldhdns works by running a controller and a lighweight DNS server as docker containers.

The controller manages the DNS server and configures systemd's resolved to use it.

The DNS server, uses the Docker API to monitor when containers are started and stopped, and uses specific labels of the containers for the subdomain name to use.

The domain names are dynamically resolveable on the host and from within containers, so that the same names can be used in either scenario, and with the actual service ports so no container to host port mappings are required.

Usage

The Controller

Start the controller, attaching it to the host network, as follows:

Security Note: The container mounts the Docker socket so that it can consume the Docker API and it is run with the apparmor=unconfined security option and mounts the SystemBus socket so that it is able to configure systemd-resolved dynamically. Please inspect the code and build the image yourself if you are concerned about security.

LDHDNS_CONTAINER_NAME=ldhdns

docker run \
  --name $LDHDNS_CONTAINER_NAME \
  --detach \
  --network host \
  --security-opt "apparmor=unconfined" \
  --volume "/var/run/docker.sock:/tmp/docker.sock" \
  --volume "/var/run/dbus/system_bus_socket:/var/run/dbus/system_bus_socket" \
  --env LDHDNS_CONTAINER_NAME=$LDHDNS_CONTAINER_NAME \
  --restart unless-stopped \
  ghcr.io/virtualstaticvoid/ldhdns:latest

Visit the virtualstaticvoid/ldhdns image repository for available image tags.

Configuration

The network ID, domain name suffix and subdomain label are configured with environment variables:

  • LDHDNS_NETWORK_ID for docker network name to use. The default is ldhdns.
  • LDHDNS_DOMAIN_SUFFIX for domain name suffix to use. The default is ldh.dns.
  • LDHDNS_SUBDOMAIN_LABEL for label used by containers. The default is dns.ldh/subdomain.
  • LDHDNS_CONTAINER_NAME for the container name of the controller. The default is ldhdns.

NOTE: The LDHDNS_CONTAINER_NAME environment variable is required since the controller needs to be able to obtain the ID of the container which it is executing in. The OCI runtime specification doesn't currently provide a portable way to obtain the container ID from within and using hacks such as via /proc/self/cgroup and /proc/1/cpuset have proven to be unreliable.

Overridding the Domain Name

You can provide your own domain name via the LDHDNS_DOMAIN_SUFFIX environment variable as follows:

LDHDNS_CONTAINER_NAME=ldhdns

docker run \
  --name $LDHDNS_CONTAINER_NAME \
  --env LDHDNS_DOMAIN_SUFFIX=ldh.example.com \
  --detach \
  --network host \
  --security-opt "apparmor=unconfined" \
  --volume "/var/run/docker.sock:/tmp/docker.sock" \
  --volume "/var/run/dbus/system_bus_socket:/var/run/dbus/system_bus_socket" \
  --env LDHDNS_CONTAINER_NAME=$LDHDNS_CONTAINER_NAME \
  --restart unless-stopped \
  ghcr.io/virtualstaticvoid/ldhdns:latest

Tip: If you are using a "real" domain name, be sure to use a subdomain off the apex domain, such as ldh. to avoid any clashes with it's public DNS resolution.

E.g. Use ldh.example.com for the domain name so that a container "foo" will be resolvable to foo.ldh.example.com instead of foo.example.com.

Containers

To make containers resolvable, add the label "dns.ldh/subdomain=<subdomain>" with the desired subdomain to use.

This subdomain will be prepended to the domain name in the LDHDNS_DOMAIN_SUFFIX environment variable to form a fully qualified domain name.

To apply the label to a container using the command line:

docker run -it --rm --label "dns.ldh/subdomain=foo" nginx

Or with Docker Compose:

# docker-compose.yml
services:
  web:
    image: nginx
    labels:
      "dns.ldh/subdomain": "foo"

Note: Make sure to use the same label key you provided in the LDHDNS_SUBDOMAIN_LABEL environment variable.

Note: Labels cannot be added to existing containers so you will need to re-create them to apply the label if needed.

Testing

Start by running Nginx in a container:

docker run --detach --label "dns.ldh/subdomain=foo" nginx

The subdomain provided will now be resolvable to the container IP address.

E.g.

dig -t A foo.ldh.dns

; <<>> DiG 9.16.1-Ubuntu <<>> -t A foo.ldh.dns
;; global options: +cmd
;; Got answer:
;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 61163
...

;; ANSWER SECTION:
foo.ldh.dns.    15  IN  A 172.17.0.2

...

And you can go ahead and consume the service.

From a terminal running on the host:

curl -v http://foo.ldh.dns

<!DOCTYPE html>
<html>
...
<h1>Welcome to nginx!</h1>
...
</html>

Or from within another container:

docker run -it --rm curlimages/curl -v http://foo.ldh.dns

<!DOCTYPE html>
<html>
...
<h1>Welcome to nginx!</h1>
...
</html>

Hacking

The development experience is based on docker compose. The docker-compose.yml file contains two instances of the ldhdns service; one configured using the defaults and the other with an alternative configuration. It also contains sample services with nginx and postgres so that tests can be conducted against these types of services.

The default configuration uses the ldh.dns DNS suffix with the dns.ldh/subdomain label, and the alternative configuration uses the alt.dns DNS suffix with the alt.ldh/subdomain label.

Building

Use docker compose build to build the ldhdns docker image locally.

Running

Use docker compose up to run the services locally.

Testing

Once the services are running, use docker compose run test to run the tests from within the test service, execute test.sh directly from your host, which uses the curl and psql tools, or navigate to the following URL's with your web browser:

Background

Consider a scenario in development where you are building a Single Page Web Application (SPA) and REST API, with a PostgreSQL database, with each service running in Docker containers on your local machine.

A web browser connects to the Web Application and the REST API, and the API connects to the PostgreSQL database.

In development, to access these services there are number of options:

  1. Map the container ports to host ports and access the services using localhost together with the host port number,
  2. Obtain the IP address of each respective container and access the services using the IP address together with the container port number,
  3. Using domain names instead of IP addresses, adding them to your /etc/hosts, mapping each container IP address to a name.

Each of these methods have difficulties, short-comings and implications, such as:

  • No consistent convention for mapping container ports to host ports.
  • Potential host port clashes when running multiple instances of a container.
  • Manual steps needed to get the IP addresses of containers.
  • Editing /etc/hosts requires root permissions.
  • Manual updates to /etc/hosts required each time an IP address changes.
  • Portability issues for other developers on their machines when collaborating on projects.
  • Configuration differences on the host vs within the container.

Furthermore, when host to container port mappings are typically used, the mapping could be 8080 to 80 for the Web Application and 8090 to 80 for the REST API. The SPA Web Application would therefore have to be configured to use http://localhost:8090 to access the API. However the API connects directly to PostgreSQL so it would have to configured to use the PostgreSQL container name.

You may also want to run some ad-hoc SQL queries whilst debugging, so connecting a tool such as psql would require a further port mapping of 8432 to 5432.

As you can see this setup gets messy and complicated quickly and isn't a great developer experience!

Now imagine adding SSL ports (443) so that you can debug under more production like conditions with TLS certificates; the situation gets nasty fast. Don't even think about having more than one instance of a container, such as when using the docker-compose up --scale api=N to add more container instances of a service!

Solution

ldhdns provides a simple solution. It monitors running containers, looking for labels which contain the domain name to use, and configures and runs a lightweight DNS server.

The domain names are dynamically resolveable on the host and from within containers, so that you can use the same fully qualified domain names in each scenario and use the actual service ports just like in production.

In the above mentioned example, you could use web, api and pgsql as the subdomains for the respective containers, making the Web Application and REST API accessible via http://web.ldh.dns and http://api.ldh.dns respectively, and the PostgreSQL service accessible via pgsql.ldh.dns.

Architecture

ldhdns consists of two services which are packaged in the same Docker container.

The following diagram illustrates the components which make up the solution, and how they interact with the host machine, the docker API, systemd-resolved and other applications such as a browser or psql.

The controller creates and configures a Docker bridge network and configures systemd-resolved to resolve DNS queries for the configured domain name. It spawns a second container to monitors the Docker API for when containers are started or stopped, creating and removing DNS records accordingly, and runs dnsmasq to resolve DNS queries for A (ipv4) and AAAA (ipv6) type records for the configured domain.

Inspiration

  • I got tired of running docker ps to figure out the container name, followed by docker inspect to get the IP address and then manually editing /etc/hosts.
  • I couldn't come up with a consistent convention for mapping host to container ports. What comes after 8099?
  • Finding IPv4 and IPv6 CIDR blocks which aren't in use so that static IP's can be used.
  • Not being able to create SSL certificates for *.xip.io or *.nip.io domains.

Credits

License

MIT License. Copyright (c) 2020 Chris Stefano. See LICENSE for details.