Description of problem: From the k8s master, $ kubectl get pods NAME READY STATUS RESTARTS AGE glusterfs 1/1 Running 0 3h systemd 1/1 Running 0 3h From the k8s node where the actual containers are running, # docker ps CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES 6dccfff97e63 rhel7 "/usr/sbin/init" 3 hours ago Up 3 hours k8s_systemd.7c2be022_systemd_default_72e636e0-5a53-11e6-9506-fa163e07a2df_59e2d256 b12abb6f58a0 gcr.io/google_containers/pause:2.0 "/pause" 3 hours ago Up 3 hours k8s_POD.6059dfa2_systemd_default_72e636e0-5a53-11e6-9506-fa163e07a2df_e8cc8ba6 70122ebd06ca fedora/nginx "/usr/sbin/nginx" 3 hours ago Up 3 hours k8s_glusterfs.6c8afde2_glusterfs_default_e37129dd-5a4f-11e6-9506-fa163e07a2df_812ff0c5 2dc1b9887f46 gcr.io/google_containers/pause:2.0 "/pause" 3 hours ago Up 3 hours k8s_POD.6059dfa2_glusterfs_default_e37129dd-5a4f-11e6-9506-fa163e07a2df_0d2786c6 # machinectl MACHINE CLASS SERVICE 2dc1b9887f4651272236c657669ebb19 container docker b12abb6f58a0981759393f98e62ed112 container docker Both of the containers were registered are pause containers. # machinectl status 2dc1b9887f4651272236c657669ebb19 2dc1b9887f4651272236c657669ebb19(32646331623938383766343635313237) Since: Thu 2016-08-04 14:29:54 UTC; 3h 35min ago Leader: 62649 (pause) Service: docker; class container Root: /var/mnt/overlay/devicemapper/mnt/d7a3b2298dbac708254ae80c66a3 Address: 172.17.0.2 fe80::42:acff:fe11:2 Unit: docker-2dc1b9887f4651272236c657669ebb1961c7e23401bfe6774fcfebd └─62649 /pause Version-Release number of selected component (if applicable): oci-register-machine-0-1.7.git31bbcd2.el7.x86_64 atomic host 7.2.6 How reproducible: always
We are dropping oci-register-machine from RHEL7 for now. There is a bug in the linux kernel that does not allow us to run docker in the host namespace. oci-register-machine does not work when you run in this state, because systemd will not see the mount points inside of the container since it will not be in docker's namespace. As far as this bug is concerned, oci-register-machine is only going to register the first container run all other containers are joining the initial container. Mrunal do you think the other containers even call into oci-register-machine? This will also cause oci-systemd-hook to not run properly, since it will not setup the containers correctly.
The hooks should be called for each container. I suspect that the fix for https://bugzilla.redhat.com/show_bug.cgi?id=1364237 isn't present in this setup.
Dan, do you think we should close this one since we are dropping oci-register-machine?
Actually I misspoke. We were disabling the RHEL7 oci-register-machine by default, but we still want to support it. I think the oci-register-machine should be working now.
I am going to mark this as fixed in the current release. If you can check to see if it is still broken.
Yes, this is working now as version 7.3.2.