Users may now deploy containerized versions of OpenShift Origin, OpenShift Enterprise on Atomic Host or RHEL, Centos, and Fedora. This includes OpenvSwitch based SDN.
When installing on Atomic Host you will automatically have containerized installation methods selected for you based on detection of /run/ostree-booted
Currently the default installation method for traditional operating systems is via RPMs. If you wish to deploy using containerized installation you may set the ansible variable 'containerized=true' on a per host basis. This means that you may easily deploy environments mixing containerized and RPM based installs. At this point we don't suggest deploying heterogeneous environments.
When using containerized installations openshift-ansible will deploy a wrapper script on each master located in /usr/local/bin/openshift and a set of symbolic links /usr/local/bin/oc, /usr/local/bin/oadm, and /usr/local/bin/kubectl to ease administrative tasks. The wrapper script spawns a new container on each invocation so you may notice it's slightly slower than native clients.
The wrapper scripts mount a limited subset of paths, ~/.kube, /etc/origin/,
and /tmp. Be mindful of this when passing in files to be processed by oc or
oadm. You may find it easier to redirect input like this :
oc create -f - < my_file.json
Based on your openshift_deployment_type the installer will make use of the following
images. Because you may make use of a private repository we've moved the
configuration of docker additional, insecure, and blocked registries to the
beginning of the installation process ensuring that these settings are applied
before attempting to pull any of the following images.
Origin
docker.io/openshift/origin
docker.io/openshift/node (node + openshift-sdn + openvswitch rpm for client tools)
docker.io/openshift/openvswitch (centos7 + openvswitch rpm, runs ovsdb ovsctl processes)
registry.redhat.io/rhel7/etcd
OpenShift Container Platform
registry.redhat.io/openshift3/ose
registry.redhat.io/openshift3/node
registry.redhat.io/openshift3/openvswitch
registry.redhat.io/rhel7/etcd
- note openshift3/* images come from registry.redhat.io and rely on the --additional-repository flag being set appropriately.
The installer will create relevant systemd units which can be used to start, stop, and poll services via normal systemctl commands. These unit names match those of an RPM installation with the exception of the etcd service which will be named 'etcd_container'. This change is necessary as currently Atomic Host ships with etcd package installed as part of Atomic Host and we will instead use a containerized version. The installer will disable the built in etcd service. etcd is slated to be removed from os-tree in the future.
All configuration files are placed in the same locations as RPM based installations and will survive os-tree upgrades.
The examples are installed into /etc/origin/examples rather than /usr/share/openshift/examples because that is read-only on Atomic Host.
Atomic Host installs by default have a small root filesystem. However
the etcd, master, and node containers will persist data in
/var/lib. Please ensure that you have enough space on the root
filesystem. This is usually handled by
docker-storage-setup;
set e.g. ROOT_SIZE=20G in /etc/sysconfig/docker-storage-setup in
early host boot, such as inside a cloud-init boot hook.
OpenShift SDN initialization requires that the docker bridge be reconfigured and docker is restarted. This complicates the situation when the node is running within a container. When using the OVS SDN you'll see the node start, reconfigure docker, restart docker which will restart all containers, and finally start successfully.
The node service may fail to start and be restarted a few times because the master services are also restarted along with docker. We currently work around this by relying on Restart=always in the docker based systemd units.