c87ac5ef2e
layer-nvidia-cuda does the hardware detection and sets a state that the worker can react to. When gpu is available, worker updates config and restarts kubelet to enable gpu mode. Worker then notifies master that it's in gpu mode via the kube-control relation. When master sees that a worker is in gpu mode, it updates to privileged mode and restarts kube-apiserver. The kube-control interface has subsumed the kube-dns interface functionality. An 'allow-privileged' config option has been added to both worker and master charms. The gpu enablement respects the value of this option; i.e., we can't enable gpu mode if the operator has set allow-privileged="false".
Kubernetes-master
Kubernetes is an open source system for managing application containers across a cluster of hosts. The Kubernetes project was started by Google in 2014, combining the experience of running production workloads combined with best practices from the community.
The Kubernetes project defines some new terms that may be unfamiliar to users or operators. For more information please refer to the concept guide in the getting started guide.
This charm is an encapsulation of the Kubernetes master processes and the operations to run on any cloud for the entire lifecycle of the cluster.
This charm is built from other charm layers using the Juju reactive framework. The other layers focus on specific subset of operations making this layer specific to operations of Kubernetes master processes.
Deployment
This charm is not fully functional when deployed by itself. It requires other charms to model a complete Kubernetes cluster. A Kubernetes cluster needs a distributed key value store such as Etcd and the kubernetes-worker charm which delivers the Kubernetes node services. A cluster requires a Software Defined Network (SDN) and Transport Layer Security (TLS) so the components in a cluster communicate securely.
Please take a look at the Canonical Distribution of Kubernetes or the Kubernetes core bundles for examples of complete models of Kubernetes clusters.
Resources
The kubernetes-master charm takes advantage of the Juju Resources feature to deliver the Kubernetes software.
In deployments on public clouds the Charm Store provides the resource to the charm automatically with no user intervention. Some environments with strict firewall rules may not be able to contact the Charm Store. In these network restricted environments the resource can be uploaded to the model by the Juju operator.
Configuration
This charm supports some configuration options to set up a Kubernetes cluster that works in your environment:
dns_domain
The domain name to use for the Kubernetes cluster for DNS.
enable-dashboard-addons
Enables the installation of Kubernetes dashboard, Heapster, Grafana, and InfluxDB.
DNS for the cluster
The DNS add-on allows the pods to have a DNS names in addition to IP addresses. The Kubernetes cluster DNS server (based off the SkyDNS library) supports forward lookups (A records), service lookups (SRV records) and reverse IP address lookups (PTR records). More information about the DNS can be obtained from the Kubernetes DNS admin guide.
Actions
The kubernetes-master charm models a few one time operations called Juju actions that can be run by Juju users.
create-rbd-pv
This action creates RADOS Block Device (RBD) in Ceph and defines a Persistent Volume in Kubernetes so the containers can use durable storage. This action requires a relation to the ceph-mon charm before it can create the volume.
restart
This action restarts the master processes kube-apiserver,
kube-controller-manager, and kube-scheduler when the user needs a restart.
More information
Contact
The kubernetes-master charm is free and open source operations created by the containers team at Canonical.
Canonical also offers enterprise support and customization services. Please refer to the Kubernetes product page for more details.