Is there a way to setup a test docker swarm on a single machine?

Question

I am trying to setup a docker swarm on WSL2 for testing purposes. I want to know, if it is possible to have a swarm with multiple "dummy" nodes on a single machine.

Here are the two ways that I trid:

1. Run multiple WSL instances as suggested here.

PS C:\Users\jdu> wsl -l
 Windows-Subsystem für Linux-Distributionen:
 Ubuntu3
 Ubuntu
 Ubuntu2

Docker is installed and run in each WSL instance. So I manage to initialize a swarm on Ubuntu and let Ubuntu2 and Ubuntu3 to join as workers.

On Ubuntu

$ docker swarm init

Swarm initialized: current node (hude19jo7t9dqpe0akg55ipmy) is now a manager.

On Ubuntu2

$ docker swarm join --token SWMTKN-1-xxxxxxxxx-xxxxxxxxx 192.168.189.5:2377 --listen-addr 0.0.0.0:12377
This node joined a swarm as a manager.

Then if I check on Ubuntu

$ docker node ls
ID                            HOSTNAME        STATUS    AVAILABILITY   MANAGER STATUS   ENGINE VERSION
hude19jo7t9dqpe0akg55ipmy *   laptop-ebc155   Ready     Active         Leader           20.10.21
ozeq43yukgfbltjnfya0tlx08     laptop-ebc155   Ready     Active         Reachable        20.10.20

2. Inspired by the ideas here, I have tried with docker-in-docker containers, e.g. I deploy multiple docker instances on a single WSL.

# Init Swarm master
docker swarm init

# Get join token:
SWARM_TOKEN=$(docker swarm join-token -q worker)
echo $SWARM_TOKEN

# Get Swarm master IP (Docker for Mac xhyve VM IP)
SWARM_MASTER_IP=$(docker info | grep -w 'Node Address' | awk '{print $3}')
echo $SWARM_MASTER_IP

DOCKER_VERSION=dind

# setup deploy Docker-in-Docker containers and join them to a swarm
docker run -d --privileged --name worker-1 --hostname=worker-1 -p 12377:2377 docker:${DOCKER_VERSION}
docker exec worker-1 docker swarm join --token ${SWARM_TOKEN} ${SWARM_MASTER_IP}:2377

docker run -d --privileged --name worker-2 --hostname=worker-2 -p 22377:2377 docker:${DOCKER_VERSION}
docker exec worker-2 docker swarm join --token ${SWARM_TOKEN} ${SWARM_MASTER_IP}:2377

docker run -d --privileged --name worker-3 --hostname=worker-3 -p 32377:2377 docker:${DOCKER_VERSION}
docker exec worker-3 docker swarm join --token ${SWARM_TOKEN} ${SWARM_MASTER_IP}:2377

After that

$ docker node ls
ID                            HOSTNAME        STATUS    AVAILABILITY   MANAGER STATUS   ENGINE VERSION
s371tmygu9h640xfosn6kyca4 *   laptop-ebc155   Ready     Active         Leader           20.10.21
w1ina9ttvje4hn6r13p3gzbge     worker-1        Ready     Active                          20.10.20
m8mqky6jchjao01nz8t5e392a     worker-2        Ready     Active                          20.10.20
n29afhbb090tlyn9p0byga9au     worker-3        Ready     Active                          20.10.20

To test the above two swarm setup, I use a very simple compose file as suggested by the official docs. As you can expect, these two swarm setup didn't work that well :/

If the MongoDB and MongoExpress are deployed on different nodes, both of the swarm setups show a same error MongoNetworkError: failed to connect to server [mongo:27017] on first connect. My understanding to this error is, that MongoExpress can not reach MongoDB under mongo:27017, which seems like a problem of the docker internal DNS. Can someone help me out? Or just feel free to tell me, dont try this single-multi nodes ideas anymore :D I am very appreciate to any help!

Answer 1

I just tried the same two exercises :)

Approach 1 - swarm nodes in WSL instances

I think it is currently impossible because of WSL2 design see https://github.com/microsoft/WSL/issues/4304. WSL2 instances are in fact sharing network setup - ip, interfaces, network namespaces, and so on. Every change made in one of them is immediately visible in all others and this conflicts with virtual interfaces and namespaces created by docker swarm nodes when they start up.

I tried configuring multiple ip addresses on eth0 interface, so that each node can have it's own (like here), and then used --advertise-addr --listen-addr options in docker swarm init and docker swarm join commands. Still I'm getting this error in dockerd logs:

moving interface ov-001000-yis5e to host ns failed, invalid argument, after config error error setting interface \"ov-001000-yis5e\" IP to 10.0.0.1/24: cannot program address 10.0.0.1/24 in sandbox interface because it conflicts with existing route {Ifindex: 4 Dst: 10.0.0.0/24 Src: 10.0.0.1 Gw: <nil> Flags: [] Table: 254}"

I believe here docker swarm hits a problem, because it already sees master's interfaces when it tries to to set up routing mesh networking for the worker. All because master and node share network config.

Approach 2 - swarm nodes as docker containers (docker-in-docker)

But I've got no 2. working with just a small change in swarm init command:

# advertise swarm on default bridge network
docker swarm init --advertise-addr 172.17.0.1

For me, the standard docker swarm init selected by default the eth0 address, which was only working for communication from dind -> wsl, but not the other way round.

Another but probably unrelated problem was that I could not access services/stacks executed this way from Windows host. This seems to be a wls bug and luckily there is a workaround.

One last hint about this mongo stack is ... patience. The stack consists of 2 services: mongo - the database and mongo-express - the client. Mongo image is a lot bigger ~600MB while mongo-express just ~135MB. The mongo-express image will be downloaded faster and it will be recreated by swarm multiple times before mongo is even started. Note also that docker images are independently downloaded for each worker in this setup, so also rebalancing may take some time.

I found these commands useful to see what is really happening:

# overview of services
docker service ls

# containers in each swarm service
docker service ps $(docker service ls --format {{.Name}})

# images in each dind worker
for i in $(seq "${NUM_WORKERS}"); do
   docker exec worker-${i} docker images
done

#containers in each dind worker
for i in $(seq "${NUM_WORKERS}"); do
   docker exec worker-${i} docker ps -a
done

Full listing of commands necessary to get working docker swarm using dind:

docker swarm init --advertise-addr docker0

SWARM_TOKEN=$( docker swarm join-token -q worker)
echo $SWARM_TOKEN

SWARM_MASTER_IP=$( docker info 2>&1 | grep -w 'Node Address' | awk '{print $3}')
echo $SWARM_MASTER_IP

DOCKER_VERSION=20.10.12-dind
NUM_WORKERS=3

# Run NUM_WORKERS workers with SWARM_TOKEN
for i in $(seq "${NUM_WORKERS}"); do
   docker run -d --privileged --name worker-${i} --hostname=worker-${i} docker:${DOCKER_VERSION}
   sleep 5
   docker exec worker-${i} docker swarm join --token ${SWARM_TOKEN} ${SWARM_MASTER_IP}:2377
done

# Setup the visualizer
docker service create \
  --detach=true \
  --name=viz \
  --publish=8000:8080/tcp \
  --constraint=node.role==manager \
  --mount=type=bind,src=/var/run/docker.sock,dst=/var/run/docker.sock \
  dockersamples/visualizer

####### play with mongo 

mkdir mongodemo && cd mongodemo
wget https://raw.githubusercontent.com/docker-library/docs/f6c9b596064e2eed9c3b6ac75bea606cb6d94099/mongo/stack.yml
docker stack deploy -c stack.yml mongo

# from windows:
# mongo will be available under <eth0>:8081
# visualizer under <eth0>:8000
ip -4 addr | grep eth0