Control socket connect(...): Connection refused

Archives Ansible Project
1

I’m running ansible on Ubuntu 14.04.2 LTS running OpenSSH_6.6.1p1 inside of a Docker container with a CoreOS host, and running this command as root:

ansible key_tleyden -i inventory -m ping -u centos -vvvv

and seeing this error:

ec2-52-152-154-133.compute-1.amazonaws.com | FAILED => SSH Error: debug2: set_control_persist_exit_time: schedule exit in 60 seconds
while connecting to 10.152.151.154:22

After quite a lot of digging (I wish ansible gave a clearer error message here), it seemed to be related to SSH ControlPersist, so I tried running this command (also as root):

ssh -o ControlMaster=auto -o ControlPersist=60s -o ControlPath=$CP centos@10.119.181.141

and with that I get the error:

Control socket connect(/root/.ansible/cp/ansible-ssh-10.119.181.141-22-centos): Connection refused

I can see the domain socket here:

srw------- 1 root root 0 Jun 10 18:11 ansible-ssh-ec2-54-157-151-133.compute-1.amazonaws.com-22-centos

but am unclear on why ssh can’t seem to use it.

If I change my ansible.cfg to disable the ControlPersist settings via:

[ssh_connection]

ssh_args =

Then the ansible errors completely go away.

2

So far this is looking like a really obscure CoreOS issue. I tried against Ubuntu running inside a Docker container on an Ubuntu host, and couldn’t reproduce this. Sorry for the noise. (but any insight would still be welcome)

3

Please check:

https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1262287

In a nutshell, the latest ssh versions and overlayfs don’t play well together, after much trouble I just configured the ControlPath ssh option to point to a tmpfs path inside the docker container and voilá.

I’m using the castawaylabs/semaphore docker image on CoreOS stable (835.9.0)

Cheers

4

@Nel, thanks for figuring this one out, this seems like a pretty nasty bug.

5

I can confirm changing the the control path to a tmpfs location fixed the problem for me. In my case I used /dev.