4.3 KiB
4.3 KiB
Kubespray's roadmap
Kubeadm
- Propose kubeadm as an option in order to setup the kubernetes cluster. That would probably improve deployment speed and certs management #553
Self deployment (pull-mode) #320
- the playbook would install and configure docker/rkt and the etcd cluster
- the following data would be inserted into etcd: certs,tokens,users,inventory,group_vars.
- a "kubespray" container would be deployed (kubespray-cli, ansible-playbook, kpm)
- to be discussed, a way to provide the inventory
- self deployment of the node from inside a container #321
Provisionning and cloud providers
- Terraform to provision instances on GCE, AWS, Openstack, Digital Ocean, Azure
- On AWS autoscaling, multi AZ
- On Azure autoscaling, create loadbalancer #297
- On GCE be able to create a loadbalancer automatically (IAM ?) #280
- TLS boostrap support for kubelet #234
(related issues: https://github.com/kubernetes/kubernetes/pull/20439
https://github.com/kubernetes/kubernetes/issues/18112)
Tests
- Run kubernetes e2e tests
- migrate to jenkins (a test is currently a deployment on a 3 node cluste, testing k8s api, ping between 2 pods)
- Full tests on GCE per day (All OS's, all network plugins)
- trigger a single test per pull request
single test with the Ansible version n-1 per day- Test idempotency on on single OS but for all network plugins/container engines
- single test on AWS per day
- test different achitectures : - 3 instances, 3 are members of the etcd cluster, 2 of them acting as master and node, 1 as node - 5 instances, 3 are etcd and nodes, 2 are masters only - 7 instances, 3 etcd only, 2 masters, 2 nodes
- test scale up cluster: +1 etcd, +1 master, +1 node
Lifecycle
- Adopt the kubeadm tool by delegating CM tasks it is capable to accomplish well #553
- Drain worker node when upgrading k8s components in a worker node. #154
- Drain worker node when shutting down/deleting an instance
- Upgrade granularity: select components to upgrade and skip others
Networking
- romana.io support #160
- Configure network policy for Calico. #159
- Opencontrail
- Canal
- Cloud Provider native networking (instead of our network plugins)
High availability
- (to be discussed) option to set a loadbalancer for the apiservers like ucarp/packemaker/keepalived While waiting for the issue kubernetes/kubernetes#18174 to be fixed.
Kubespray-cli
- Delete instances
kubespray vagrantto setup a test cluster locallykubespray azurefor Microsoft Azure support- switch to Terraform instead of Ansible for provisionning
- update $HOME/.kube/config when a cluster is deployed. Optionally switch to this context
Kubespray API
- Perform all actions through an API
- Store inventories / configurations of mulltiple clusters
- make sure that state of cluster is completely saved in no more than one config file beyond hosts inventory
Addons (with kpm)
Include optionals deployments to init the cluster:
Monitoring
- Heapster / Grafana ....
- Prometheus
Others
Dashboards:
- kubernetes-dashboard
- Fabric8
- Tectonic
- Cockpit
Paas like
- Openshift Origin
- Openstack
- Deis Workflow