Vagrant single machine not starting

[jma...@redhat.com]

Hi All,

I'm going through the Vagrant instructions for running a single-machine and I'm hitting issues. It appears that several services are not starting correctly [1] and I waited awhile for all the downloads to happen like the instructions say [2] and then I looked at docker service status [3] and logs and the service was not able to start. Has anyone else see anything like this?

Thanks,
Jason

[1]
$ vagrant ssh                                                                                                                                                   master 1876aac ✗
CoreOS alpha (1151.0.0)
Last login: Fri Aug 26 13:48:26 2016 from 10.0.2.2
Update Strategy: No Reboots
Failed Units: 3
  docker.service
  update-engine.service
  docker.socket

[2] https://coreos.com/kubernetes/docs/latest/kubernetes-on-vagrant-single.html

[3]
core@localhost ~ $ systemctl status docker.service
● docker.service - Docker Application Container Engine
   Loaded: loaded (/usr/lib/systemd/system/docker.service; disabled; vendor preset: disabled)
  Drop-In: /etc/systemd/system/docker.service.d
           └─40-flannel.conf
   Active: failed (Result: start-limit-hit) since Fri 2016-08-26 13:37:06 UTC; 12min ago
     Docs: http://docs.docker.com
  Process: 5031 ExecStart=/usr/lib/coreos/dockerd daemon --host=fd:// $DOCKER_OPTS $DOCKER_CGROUPS $DOCKER_OPT_MTU (code=exited, status=1/FAILURE)
 Main PID: 5031 (code=exited, status=1/FAILURE)

Aug 26 13:37:06 localhost systemd[1]: Starting Docker Application Container Engine...
Aug 26 13:37:06 localhost dockerd[5031]: dockerd: "dockerd" requires 0 arguments.
Aug 26 13:37:06 localhost dockerd[5031]: Usage:        dockerd [OPTIONS]
Aug 26 13:37:06 localhost systemd[1]: docker.service: Main process exited, code=exited, status=1/FAILURE
Aug 26 13:37:06 localhost systemd[1]: Failed to start Docker Application Container Engine.
Aug 26 13:37:06 localhost systemd[1]: docker.service: Unit entered failed state.
Aug 26 13:37:06 localhost systemd[1]: docker.service: Failed with result 'exit-code'.
Aug 26 13:37:06 localhost systemd[1]: docker.service: Start request repeated too quickly.
Aug 26 13:37:06 localhost systemd[1]: Failed to start Docker Application Container Engine.
Aug 26 13:37:06 localhost systemd[1]: docker.service: Failed with result 'start-limit-hit'.

[pb]

Hello,

I discovered this bug yesterday. The simple fix, if your planning to stay on CoreOS alpha with Docker v1.12, is to remove the word 'daemon' from '/etc/systemd/system/docker.service.d/40-flannel.conf'.

Basically, if you setup kubernetes from scripts in the github.com/coreos/coreos-kubernetes repo on CoreOS,\ between the ~13 day window of PR #551 and #642(the fix) then rolling to CoreOS v1151.0.0+ will break your docker.service until you modify the drop-in.

The issue is that PR #551 added a drop-in to docker.service that overrides the `ExecStart` line to drop some parameters (for a different network configuration style). The problem is that the command line parameters to start dockerd changed between Docker v1.11 and v1.12 and v1.12 baulks at the use of the daemon argument. If the drop-in isn't present, CoreOS would just update the ExecStart line across versions correctly via the included unit file in /usr/lib.... Its really my fault for introducing a dependance on the ExecStart line remaining correct when I just needed to override two environment variables.

The more correct fix, that works for both Docker v1.11 and v1.12, is here: https://github.com/coreos/coreos-kubernetes/pull/642

Hope this helps, let me know if you have further questions.

Regards,

Patrick

[Jason Marley]

Unlucky on my part :). Cool thanks Patrick. That worked, but I had to reload the daemon afterwards I updated the 40-flannel.conf.

Out of curiosity, why isn’t kubernetes hooked into Rocket in lieu of Docker?

Jason Marley

jma...@redhat.com

[pb]

Hey Jason, glad it worked. Its actually been possible to run kubernetes on rkt since the same PR#551 in the github.com/coreos/coreos-kubernetes repo. Just set the container runtime flag to 'rkt'. 

We're planning on making a coreos-kubernetes release soon that includes support for configuring rkt as the kubernetes runtime. Its only just gotten to a stable point in both upstream kubernetes and in our kubernetes tooling. Check out http://kubernetes.io/docs/getting-started-guides/rkt/notes/ for known issues and general information.

Patrick