Generate CoreOS cloud configs that allow installing Kubernetes via Fleet on Vagrant or AWS. Kubernetes is configured with secure connections and token authentication.
On AWS, these scripts target a CoreOS cluster consisting of a three node etcd-cluster, where one etcd node is the Kubernetes master. Any number of Kubernetes nodes can specified, but these do not run etcd themselves.
Configuration overview:
global.conf: global configuration that applies to all targets (AWS, Vagrant).defaults.yamldefault cloud-config configuration that applies to all targets.vagrant.yamlconfiguration specific to Vagrant (overrides values indefaults.yaml).aws.yamlconfiguration specific to AWS (overrides values indefaults.yaml).
You can run these cloud-configs with Vagrant. Simply do:
vagrant up
and this will launch one master and two nodes where only the master
runs etcd. Set a different number of Kubernetes nodes via the
NUM_NODES environment variable, e.g.,:
NUM_NODES=3 vagrant up
On AWS, this cloud config targets a cluster of two etcd-running
instances, one Kubernetes master (also running etcd), and any number
of Kubernetes nodes (matching a CoreOS production cluster). If needed,
edit the settings in the above configuration files, if needed
(specifically, you might want to add a user and ssh keys to
defaults.yaml).
To generate the cloud-config for a target, run:
./make_cloud_config.py -t aws
If the -t option is omitted the target defaults to Vagrant. The
actual cloud-configs ends up in /tmp/etcd.yaml, /tmp/master.yaml
and /tmp/node.yaml.
Now launch two etcd instances, and one master. Note the private IP
addresses of those instances and manually update the etcd_servers
for Fleet and etcd_endpoints for Flannel in the node.yaml cloud
config. Then launch any number of nodes using the updated cloud
config. Note that the master and the nodes need to be associated with
a AWS security role with full EC2 access policy (or use the policy
files from the official Kubernetes project).
When a CoreOS cluster is running, either on Vagrant or AWS, install Kubernetes on the CoreOS cluster using Fleet:
./kube-up.sh [ MASTER_IP ]
On Vagrant, the MASTER_IP can be omitted. On AWS, you obviously need
access to Kubernetes on all relevant ports (e.g, 2379, 8080),
typically from another machine in the same AWS VPC.
After a short while, you should be able to query the Kubernets master (e.g, on Vagrant):
./kubectl --server=http://172.17.8.101:8080 get pods