Kubernetes Formula

A salt formula to deploy a ready to use kubernetes cluster(Tested in Ubuntu 14.04). This formula mainly referenced from official saltbase code. The following component is deployed:

Master node:

• etcd: deployed as static pod manifest
• kubelet: using hyperkube bin and initd to ensure service running
• kube-api-server: deployed as static pod manifest
• kube-control-manager: deployed as static pod manifest
• kube-scheduler: deployed as static pod manifest

Pool node:

• kubelet: using hyperkube bin and initd to ensure service running
• kube-proxy: deployed as static pod manifest

Pre-request

1. Docker installation is not in the scope of this formula. It is required docker installed and running. Please make sure iptables=false and ip-masq=false docker opts are set.
2. A mine function network.internal_ip, distributed to all minions, that returns the internal IP address to use for apiserver communication. Example pillar data:

  mine_functions:
    network.internal_ip:
      - mine_function: grains.get
      - 'ip_interfaces:eth1:0'

How to use?

1. Create pillar data file based on pillar.example file and modify as needed. More configurable pillar data can be referenced from default.yml.

2. (Optional) To use flannel+calico as network plugin, please modify/add pillar data as needed(bind_iface: eth1, network_provider: cni, allocate_node_cidrs: true).

3. (Optional, only required for multi-master nodes setup) Generate TLS certificates and share among master nodes:

   a. Scope master nodes by setting pillar data kubernetes:master to True.

   b. Apply kuberentes.generate-cert state to one master node.

   c. Copy all newly generated certs under /srv/kubernets/(ca.crt, server.key, server.cert, kubecfg.key, kubecfg.crt) into pillar kubernetes:certs with filename as key and file content as value.

4. Apply kubernetes state among cluster master and pool nodes.

Note for setting up HA etcd/apiserver

• HA etcd require at least 3 node to be fault-tolerant, with the current setup meaning at least 3 master nodes
• HA apiserver require to put all apiserver endpoint(https://MASTER_NODE_IP:6443) behind one or multiple load balancer. Then add the load balancer ip and port to pillar data kubernetes:api_server:ip and kubernetes:api_server:port before step#3.

Note for add/remove node into existing cluster

Master node

• Add steps:

  1. make sure pillar data kubernetes:master is true for the host need to add
  2. make sure grain data initial_etcd_cluster_state is "existing" for the host need to add
  3. run salt 'ALL_NODE_*' saltutil.refresh_pillar to update pillar data
  4. run salt 'TARGET_MASTER_NODE' state.sls kubernetes to apply kubernetes state
  5. confirm the ip address of the host need to add with command like salt TARGET_MASTER_NODE grains.get ip_interfaces:eth0
  6. add the new etcd node into existing etcd cluster by running the following commands within an etcd container. For example the ip of the new master node is 192.168.51.7, then:

  * (for add kubernete main etcd cluster) `etcdctl member add etcd-192-168-51-7 http://192.168.51.7:2380`
  * (for add kubernete event etcd cluster) `etcdctl --endpoints=http://127.0.0.1:4002 member add etcd-192-168-51-7 http://192.168.51.7:2381`
  

  7. make sure load-balancer(e.g. haproxy) is updated if configured

• Remove steps:

  1. evict all scheduled pods by using command hyperkube kubectl taint node TARGET_NODE evict=true:NoExecute
  2. remove the target etcd node from existing etcd cluster by running the following commands within an etcd container. For example the ip of the new master node is 192.168.51.7, then:

  * locate member_id by matching ip with result of command `etcdctl member list`; then remove from kubernete main etcd cluster by command `etcdctl member delete TARGET_MEMBER_ID`
  * locate member_id by matching ip with result of command `etcdctl --endpoints=http://127.0.0.1:4002 member list`; then remove from kubernete event etcd cluster by using `etcdctl --endpoints=http://127.0.0.1:4002 member delete TARGET_MEMBER_ID` 
  

  3. remove node from kubernete cluster by using hyperkube kubectl delete node TARGET_NODE on a master node
  4. using salt-cloud/salt-key to delete the node from saltstack cluster
  5. run salt 'ALL_NODE_*' saltutil.refresh_pillar to update pillar data
  6. run salt 'TARGET_MASTER_NODE' state.sls kubernetes to apply kubernetes state to update
  7. make sure load-balancer(e.g. haproxy) is updated if configured

Pool node

• Add steps:

  1. make sure pillar data kubernetes:pool is true for the host need to add
  2. run salt 'TARGET_POOL_NODE' state.sls kubernetes to apply kubernetes state

• Remove steps:

  1. evict all scheduled pods by using command hyperkube kubectl taint node TARGET_NODE evict=true:NoExecute
  2. remove node from kubernete cluster by using hyperkube kubectl delete node TARGET_NODE" on a master node
  3. using salt-cloud/salt-key to delete the node from saltstack cluster

Note for upgrade/downgrade node

1. evict all scheduled pods by using command hyperkube kubectl taint node TARGET_NODE evict=true:NoExecute
2. modify version pillar data needed(currently major version update is handled by formula default file; but each component version is configurable through respect pillar data)
3. remove existing pkg by using command salt TARGET_NODE state.sls kubernetes.kubelet.removed
4. run salt 'TARGET_NODE' state.sls kubernetes to apply kubernetes state again
5. mark node as working status with command hyperkube kubectl taint node TARGET_NODE evict:NoExecute-

Note for cluster auto-scaler

• cluster auto-scaler feature rely on cloud provider, for AWS please reference here
• To enable auto-scaler make sure pillar data kubernetes:cluster_autoscaler:enabled is true(Default is false) and set kubernetes:cluster_autoscaler:params as needed.

Notes

• For easier debugging purpose: a kuberenetes.kubelet.reset state can be apply to all nodes to make sure kubelet is stop and all generated container is removed. e.g. To hard restart the cluster: kubernetes state can be apply to include this kubelet.reset state like this:

  salt minion* state.sls kubernetes pillar='{"kubernetes":{"reset_kubelet": true}}'

TODO

1. add logrotate to generate logs
2. organize defaults.yml