Age | Commit message (Collapse) | Author | Files | Lines |
|
Adds the external_ip variable that can be used to set ip different
from cluster_ip, like public/floating/elastic ip, that can be used to externally
communicate with kubernetes nodes. Mainly useful
for the control plane and communication with
api server.
Change-Id: Ie840bff4b8479081b7397a42935c9db53db1bf4f
Issue-ID: OOM-2112
Signed-off-by: Michal Zegan <m.zegan@samsung.com>
|
|
This group will contain kubernetes etcd cluster.
It is separated from kubernetes control plane and, especially, nodes.
There are of course no restrictions as to which
machines can be etcd.
Default is infrastructure-server.
Change-Id: I8d3ab9b9e4680f57ea8f595d7be3ed6e2d32764c
Issue-ID: OOM-1778
Signed-off-by: Michal Zegan <m.zegan@samsung.com>
|
|
Moved all kubernetes nodes from the kubernetes to kubernetes-node group,
and made the kubernetes group a parent of both kubernetes-node and
kubernetes-control-plane.
The reason is that we still need to have separate groups for
control planes and for nodes, but some
operations are performed equally on any kind of kubernetes
cluster member, and currently one would need to separately include all
related groups one by one. Example of
such common operation is installation of docker, that was not
performed for kubernetes-control-plane group, so that
docker was not installed on control planes that were neither infra nor nodes.
The side effect of this change is that if infrastructure
server is also a cluster member, it lands in a kubernetes group by default,
and node specific setup would be performed on it.
For that reason, playbook is modified to perform this setup on all
kubernetes cluster members except the infra server.
Change-Id: Ic827002d28e535334dbab7e5ad4aed8aa95c97a9
Issue-ID: OOM-1778
Signed-off-by: Michal Zegan <m.zegan@samsung.com>
|
|
Added a new playbook rke.yml and role rke which uses rancher RKE:
https://github.com/rancher/rke
It's an implementation of the kubernetes from rancher.com folks
and it is an alternative to the official kubernetes tool: kubeadm.
NOTE: Rancher has a notion of a 'control plane' which cannot run with
etcd on all nodes in a multi-node setup. Control-plane node is the
first kubernetes node from the inventory (as of now).
Change-Id: I0bf669442a5183efa20d44fb1cac823e7ce54348
Issue-ID: OOM-1778
Signed-off-by: Petr Ospalý <p.ospaly@partner.samsung.com>
Signed-off-by: Michal Zegan <m.zegan@samsung.com>
|
|
This fixes adding of dns address to
/etc/docker/daemon.json to always use infra's ip address
instead of possibly using it's hostname.
Also similar change is done in handling of
/etc/resolv.conf in infrastructure.yml.
Ip address is specified by the "cluster_ip" variable placed on a host.
Inventory specifies ip for everything except resource server (just for consistency).
Change-Id: I2baa3f4c7d69b8dff109b5f8ae984134caf2af7d
Issue-ID: OOM-1573
Signed-off-by: Michal Zegan <m.zegan@samsung.com>
|
|
Issue-ID: OOM-1551
Change-Id: I6ed8cae7fae5b482b5934d0e974434083714cf75
Signed-off-by: Petr Ospalý <p.ospaly@partner.samsung.com>
|