Docker : /var/run/docker.sock: no such file or dir

Infra

I am trying to set up a mesos<->marathon cluster distributing containers. On my main server I have:

zookeeper
docker registry v2 (port 5000) wo credentials
a container with supervisord + mesos + marathon

Furthermore I have a slave (on the same server).

$docker ps 
192.168.0.38:5000/mesos-slave:prod                                                                                  mesos-slave-1
192.168.0.38:5000/mesos-master:generic                                                                              mesos-master
jplock/zookeeper                           0.0.0.0:2181->2181/tcp, 0.0.0.0:2888->2888/tcp, 0.0.0.0:3888->3888/tcp   nostalgic_visvesvaraya
registry:2                                 0.0.0.0:5000->5000/tcp                                                   registry

Issue

My issue is that I can't have marathon sending a docker image to the slave.

I think this comes from the configuration of the slave.

$ docker exec -ti mesos-slave-1 /bin/bash
root@xw8600-Workstation:/# docker info
Get http:///var/run/docker.sock/v1.19/info: dial unix /var/run/docker.sock: no such file or directory. Are you trying to connect to a TLS-enabled daemon without TLS?

The issue is on the {dial unix /var/run/docker.sock: no such file or directory}.

Slave's Dockerfile

The Dockerfile for the slave looks more or less like that :

FROM 192.168.0.38:5000/supervisor:prod

RUN apt-get -y --fix-missing update
RUN apt-get install -y --force-yes mesos

#Mesos
RUN echo 'docker,mesos' > /etc/mesos-slave/containerizers
RUN echo '5mins' > /etc/mesos-slave/executor_registration_timeout

COPY mesos-slave.conf /etc/supervisor/conf.d/mesos-slave.conf

CMD  supervisord -c /etc/supervisor.conf

The mesos-slave.conf looks like that :

[program:mesos-slave]
command=mesos-slave --master=%(ENV_MASTER)s --ip=%(ENV_IP)s --hostname=%(ENV_HOSTNAME)s --resources=%(ENV_RESOURCES)s --attributes=%(ENV_ATTRIBUTES)s --port=%(ENV_SLAVE_PORT)s --log_dir=/etc/mesos/logs --containerizers=docker,mesos --no-switch_user

I get always the same error on the error log. Example to pull rstudio :

failed to start: Failed to 'docker -H unix:///var/run/docker.sock pull 192.168.0.38:5000:rocker/rstudio': exit status = exited with status 1 stderr = Post http:///var/run/docker.sock/v1.19/images/create?fromImage=192.168.0.38%3A5000%3Arocker%2Frstudio%3Alatest: dial unix /var/run/docker.sock: no such file or directory. Are you trying to connect to a TLS-enabled daemon without TLS?

What do I do wrong ?

Edit

It seems I can not start docker

root@xw8600-Workstation:/var/log/upstart# sudo /etc/init.d/docker status
 * Docker is not running
root@xw8600-Workstation:/var/log/upstart# sudo /etc/init.d/docker start
 * Starting Docker: docker
   ...done.
root@xw8600-Workstation:/var/log/upstart# sudo /etc/init.d/docker status
 * Docker is not running

Edit 2

Following my question here : I have a mesos slave (v1.7) on which I try to get a docker in docker running. When I am inside the container, I get the following error log :

$ cat /var/log/docker.log 
time="2015-10-17T12:27:40.963674511Z" level=fatal msg="Error starting daemon: error initializing graphdriver: operation not permitted"

I can't find much info on internet, here are some similar questions:

https://github.com/jpetazzo/dind/issues/89
https://github.com/docker/docker/issues/14543
Docker hello-world on Ubuntu - permission denied

回答1:

I run a completely dockerized Mesos/Marathon infrastructure on CoreOS. If you want to have a look at the systemd service definitions, feel free...

I think you're missing to pass in the Docker unix socket and the cgroups to your slave container, see

https://github.com/tobilg/coreos-setup/blob/master/etc/systemd/system/mesos-slave.service#L17

Docker run command:

docker run -d \
  -v /sys/fs/cgroup:/sys/fs/cgroup \
  -v /usr/bin/docker:/usr/bin/docker:ro \ 
  -v /var/run/docker.sock:/var/run/docker.sock \
  [your other options] \
  192.168.0.38:5000/mesos-slave:prod

Personally, I'd also run another separate container for Marathon, instead of having it in the same image as the Mesos Master, but that may be a matter of taste.

回答2:

You cannot restart docker within a docker container. You can run docker within a docker containers But restarting? (Correct me if I'm wrong)

Essentially, you are stopping the docker daemon in your container 'xw8600-Workstation' while the container itself is using the very same docker daemon.

A better architecture is to just have your mesos-slave in a plain VM or bare-metal then run your docker daemon there and let mesos/marathon manage your resources through Docker. Hope it makes sense.