Pipeline Runner: Communicate with Docker on host

[moritz_matchory_com]

We deploy from BitBucket Pipelines into a Docker Swarm Mode cluster. This requires setting up a remote SSH context to one of the cluster manager nodes, and executing the docker commands in that context. Essentially the following:

# Preserve the docker host, then unset it so it doesn't interfere 
# with the context
export PREVIOUS_DOCKER_HOST=$DOCKER_HOST
unset DOCKER_HOST

# Create and use the remote context
docker context create remote --docker "host=ssh://$DEPLOYMENT_USER@$DEPLOYMENT_HOST"
docker context use remote

# Log into the registry
echo $DOCKER_HUB_PASSWORD | docker login --username $DOCKER_HUB_USER --password-stdin

# Deploy the service
docker stack deploy \
  --with-registry-auth \
  --prune \
  --compose-file docker-compose.production.yaml \
$BITBUCKET_REPO_SLUG

# Restore the pipeline docker host, in case we need it later on
export DOCKER_HOST=PREVIOUS_DOCKER_HOST

This has several drawbacks:

• Without jumping through some annoying hoops, $DEPLOYMENT_HOST must be a single, static host that provides access to the cluster. This is inflexible.
• What's more, it needs to be a manager node. So every time we reshape the cluster, we'll need to update BitBucket configuration, too.
• The node needs to expose SSH, and provide access to a pipeline user. This means copy-pasting public keys to that node, manually, for every service, and adds another potential security breach.
• The whole context switching means lots of boilerplate for all pipelines.

A better way

Instead, we'd like to have a self-hosted runner, running in the cluster, so the deployment step can run within the cluster and deploy services to the local Docker instance directly.
This solves every drawback:

• As runners connect to BitBucket by themselves, the static host address is gone. 
• We can have several runners which can run on several manager nodes and load being spread evenly among them. If a runner moves to another node, it takes build steps with it.
• SSH does not need to be exposed externally, as communication happens via the runner web socket.
• Docker context would be replaced by simply setting the Docker Host to the local Docker on the cluster node.

The problem

What sounds so neat in theory doesn't work, because I can't get the step executed inside the runner to talk to the Docker daemon on my cluster node. I would need to mount the Docker socket from the host into the step container, or resolve the host IP to communicate with the TCP socket.

Is there a way to talk to the docker daemon on the runner host from within a build step?

Edit: For reference, here's the output of docker info inside the build step (emphasis mine):

+ DOCKER_HOST=$BITBUCKET_DOCKER_HOST_INTERNAL docker info
Client:
Context: default
Debug Mode: false
Server:
Containers: 0
Running: 0
Paused: 0
Stopped: 0
Images: 0
Server Version: 20.10.15
Storage Driver: overlay2
Backing Filesystem: extfs
Supports d_type: true
Native Overlay Diff: true
userxattr: false
Logging Driver: json-file
Cgroup Driver: cgroupfs
Cgroup Version: 1
Plugins:
Volume: local
Network: bridge host ipvlan macvlan null overlay
Authorization: pipelines
Log: awslogs fluentd gcplogs gelf journald json-file local logentries splunk syslog
Swarm: inactive
Runtimes: io.containerd.runc.v2 io.containerd.runtime.v1.linux runc
Default Runtime: runc
Init Binary: docker-init
containerd version: 212e8b6fa2f44b9c21b2798135fc6fb7c53efc16
runc version: v1.1.1-0-g52de29d7
init version: de40ad0
Security Options:
apparmor
seccomp
Profile: default
userns
Kernel Version: 5.4.0-105-generic
Operating System: Alpine Linux v3.15 (containerized)
OSType: linux
Architecture: x86_64
CPUs: 2
Total Memory: 3.746GiB
Name: 086d6e21a117
ID: WVBT:EBC5:HCCO:SKNB:P664:IUDG:AU2M:XDQE:BXR2:22E4:AXMH:HT3G
Docker Root Dir: /var/lib/docker/165536.165536
Debug Mode: false
Registry: https://index.docker.io/v1/
Labels:
Experimental: false
Insecure Registries:
127.0.0.0/8
Registry Mirrors:
http://localhost:5000/
Live Restore Enabled: false
Product License: Community Engine

WARNING: API is accessible on http://0.0.0.0:2375 without encryption.
Access to the remote API is equivalent to root access on the host. Refer
to the 'Docker daemon attack surface' section in the documentation for
more information: https://docs.docker.com/go/attack-surface/
WARNING: No swap limit support