RIPE Atlas Docker Image

This is the RIPE Atlas software probe packaged as a Docker image.

Usage

Requirements

• 1 CPU core (of course)
• 20MiB memory
• 100MiB HDD
• A Linux installation with Docker installed
• Internet access

Tags

The following prebuilt tags are available at Docker Hub:

• latest, latest-probe, latest-anchor: latest stable version
• v{version}, v{version}-probe, v{version}-anchor: matches upstream version
• edge, edge-probe, edge-anchor: whatever from the master branch

Since version 5090, we do not provide -{arch} tags anymore.

Running

You can run the container manually with any OCI container runtime of your choice. There are some templates:

Using Docker Compose

An example docker-compose.yaml is provided.

cd contrib/docker-compose
docker compose pull
docker compose up -d

Using podman-systemd.unit

install --user=root --group=root --target /etc/containers/systemd/ -- contrib/podman-quadlet/*.container
systemctl reload
systemctl start ripe-atlas.service

Registering the Probe

Fetch the generated public key:

cat /etc/ripe-atlas/probe_key.pub

Register the probe with your public key. After the registration being manually processed, you'll see your new probe in your account.

Building

If you don't want to use the prebuilt image hosted on the Docker Hub, you can build your own image.

DOCKER_BUILDKIT=1 docker build --tag localhost/ripe-atlas:latest-probe --target ripe-atlas-probe .

Note that building this container image requires BuildKit.

Caveats

IPv6

Docker 27.0.1 enabled IPv6 (incl. ip6tables and NATv6) by default.

If you are on older versions: Docker does not enable IPv6 by default. If you want IPv6 support, some level of setup and a basic understanding of IPv6 is required. Swarm mode & some Kubernetes implementation supports IPv6 too with extra configuration.

Using native address assignment

If you happened to have a block of static IPv6 addresses routed to your host, you can directly assign one of the addresses to the container. Edit /etc/docker/daemon.json and add native IPv6 address blocks, then restart the Docker daemon. An example:

{
  "ipv6": true,
  "fixed-cidr-v6": "2001:db8:a1a3::/48"
}

Notes:

• These config work on Docker for Linux only
• If daemon.json exists, merge the config lines instead of directly overwriting it; if it doesn't exist, create it manually
• For more info, see the official doc

Using NAT (NPTv6)

If your ISP does not conform to BCOP 690 (very common), and/or your router cannot route smaller blocks of IPv6 to one server even if it has been assigned a block of valid IPv6 addresses (also very common), the method above might not work for you. As a workaround, you can setup NAT with either Docker's builtin experimental IPv6 NAT support, robbertkl/docker-ipv6nat or similar projects. Manual iptables/nftables NAT setup is also possible, but hanc marginis exiguitas non caperet.

Firstly, edit kernel parameters to enable IPv6 routing.

cat > /etc/sysctl.d/50-docker-ipv6.conf <<EOF
net.ipv6.conf.eth0.accept_ra=2
net.ipv6.conf.all.forwarding=1
net.ipv6.conf.default.forwarding=1
EOF
sysctl -p /etc/sysctl.d/50-docker-ipv6.conf

Notes:

• This potentially introduces more attack surface and might require you set up IPv6 firewall rules to make yourself safe
• This might break your network and your mileage may vary
• Swap eth0 with your primary network adapter name
• If you use static IPv6 assignment instead of SLAAC, change accept_ra to 0

Secondly, create a IPv6 NAT enabled network.

docker network create --ipv6 --subnet=fd00:a1a3::/48 ripe-atlas-network
docker run -d --restart=always -v /var/run/docker.sock:/var/run/docker.sock:ro -v /lib/modules:/lib/modules:ro --cap-drop=ALL --cap-add=NET_RAW --cap-add=NET_ADMIN --cap-add=SYS_MODULE --net=host --name=ipv6nat robbertkl/ipv6nat:latest

Finally, start the RIPE Atlas container with argument --net=ripe-atlas-network.

Auto Update

Use this recipe for auto updating the docker container.

docker run --detach --restart=always -v /var/run/docker.sock:/var/run/docker.sock --name watchtower containrrr/watchtower --cleanup --label-enable

Then start the RIPE Atlas container with argument --label=com.centurylinklabs.watchtower.enable=true.

Backup

Back up /etc/ripe-atlas is enough.

Resetting the Probe

If the probe is acting weird or not connecting to the server for a prelonged time without any error logs, you can try resetting the probe's internal state by deleting everything in /var/spool/ripe-atlas and /run/ripe-atlas then restarting the container.

Security

Upstream software does not correctly use Linux capabilities(7) and tries to mess up everything by using setuid executables. So:

Container Runtime	Container User	Network Namespace	Works	Caveats
root	root	separate	YES
root	non-root	separate	NO	daemons does not start
root	root	host	?
root	non-root	host	NO	daemons does not start
rootless	root	separate	YES	traceroute might not work
rootless	non-root	separate	NO	daemons does not start
rootless	root	host	NO	eooqd: socket: Operation not permitted
rootless	non-root	host	NO	daemons does not start

When the host distro is Debian 10 or similarly old ones, you might need to add --security-opt seccomp:unconfined to the docker run command to make things work (#19). You should upgrade your host distro ASAP.

Upgrading from 5080 to 5100 or Later

At version 5090, upstream introduced a lot changes that require manual intervention.

• You need to update the container startup arguments. See Running for an example. Note that new permissions are required to make the directory initialization process work.
• The SSH keys are stored at /etc/ripe-atlas now. Please mv /var/atlas-probe/etc /etc/ripe-atlas and make sure they are owned by 101:999 (before subuid/subgid mapping, if applicable).
• /var/atlas-probe is not used anymore and should be removed.
• /var/spool/ripe-atlas and /run/ripe-atlas are now used to store probe runtime info.
• If you are still using latest-{arch} tags, please update to use only latest.