Kubernetes 1.6 is forecast to be released this week. Major themes include new capabilities for Daemon Sets, the beta release of Kubernetes federation and new scheduling features, and new networking capabilities. You can get an in-depth look at all of the new features in the Kubernetes 1.6 release notes, but let's get a quick overview here.
DaemonSet rolling updates
You're probably used to dealing with Kubernetes in terms of creating a Deployment or a ReplicationController and having it manage your pods, making certain that you always have a particular number of instances spread among the nodes that are available. DaemonSets, on the other hand, look at things from the opposite perspective.With DaemonSets, you specify the nodes to run a particular set of containers, and Kubernetes will make certain that any nodes that satisfy those requirements will run those pods. With Kubernetes 1.6, you now have the option to update those DaemonSets with a new image or other information. (For more information on DaemonSets, you can see this article,which explains how and why to use them.)
As Kubernetes takes hold, the likelihood of running into situations in which users have multiple large clusters to deal with increases. Federation enables you to create an infrastructure in which users can use, say, the closest cluster to them, or the one that has the most spare capacity.Now in beta, kubefed "supports hosting federation on on-prem clusters, [and] automatically configures kube-dns in joining clusters and allows passing arguments to federation components."
Authentication and access control improvements
Role-Based Access Control (RBAC), which makes it possible to define roles for control plane, node, and controller components, is now in the beta phase. (It also defines default roles for these components.) There are numerous changes from the alpha version (such as a change from using * for all users to using system:authenticated or system:unauthenticated) so make sure to check out the release notes for all the details. Attribute-Based Access Control (ABAC) also been tweaked, with wild cards defaulting to authenticated users. The kube-apiserver and the authentication API have also seen a number of improvements.
Now in beta is the ability to have multiple schedulers, with each controlling a different set of pods. You can also set the scheduler you want for a particular pod on the pod sec, rather than as an annotation, as in the alpha version.Also in beta are node and pod affinity/anti-affinity. This capability enables you to intelligently schedule pods that should, or shouldn't be, on the same piece of hardware. For example, if you have a web application that talks to a database, you might wat them on the same pod. If, on the other hand, you have a pod that needs to be highly available, you might want to spread different instances over different nodes as a safeguard against failure. You can specify the affinity field on the PodSpec.Kubernetes 1.6 also includes the beta release of taints and tolerations, and some improvements to that functionality from the alpha version. Taints enable you to dedicate a node to a particular kind of pod, similar to the way in which you might flavors in OpenStack. Unlike OpenStack, however, you can tell Kubernetes to try to avoid scheduling pods that aren't explicitly allowed (read: tolerated) to that node, but if it has no choice, it can go ahead. This functionality also enables to you specify a period of time a mod might run on this node before being "evicted."And speaking of being evicted, Kubernetes 1.6 now enables you to override the default 5 minute period during which a pod remains bound to a node if there are problems,s o you can specify that a pod either finds another node more quickly, or is more patient and waits even longer.
The Container Runtime Interface is now the default
While it's natural to assume that containers running on Kubernetes are Docker containers, that's not always true. Kubernetes also supports rkt containers, and in fact the goal is to enable Kubernetes to orchestrate any container runtime. Up until now, that's been difficult, because the container runtimes were coded into the kubelet component that runs the actual containers.Now, with Kubernetes 1.6, the beta version of the Docker Container Runtime Interface is enabled by default -- you can turn it off with --enable-cri=false -- it will be easier to add new runtimes. The old non-runtime architecture is deprecated in 1.6 and is scheduled for remove in Kubernetes 1.7.
Kubernetes 1.6 includes the general availability release of StorageClasses, which enable you to specify a particular type of storage resource for users without exposing them to the details. (This is also similar to flavors in OpenStack.)Also now in GA are the ability to populate environment variables from a configmap or a secret, as well as support for writing and running your own dynamic PersistentVolume provisioners.Note that StorageClasses will change the behaviors of PersistentVolumeClaim objects on existing clouds, so be sure to read the Release Notes.
You now have added control over DNS; Kubernetes 1.6 enables you to set stubDomains, which define the nameservers used for specific domains (such as *.mycompany.local), and to specify what upstreamNameservers you want to use, overriding resolve.conf.Digging deeper, the Container Network Interface (CNI) is now integrated with the Container Runtime Interface (CRI) by default, and the standard bridge plugin has been validated with the combination.
Kubernetes 1.6 includes a huge number of changes and improvements, some of which will only be of interest to operators, as opposed to end users, but all of which are important. Some of these changes include:
By default, etcd v3 is enabled, enabling clusters up to 5000 nodes
The ability to know via the API whether a Deployment is blocked
Easier logging access
Improvements to the Horizontal Pod Autoscaler
The ability to add third party resources and extension API servers with the edit command
New commands for creating roles, as well as determining whether you can perform an action