This week we have a special issue, summarizing last week’s Contributor Summit. Expect LWKD to be spotty over the holidays.
Contributor Summit Summary
KubeCon Seattle last week included the 2018 Contributor Summit, which had dozens of discussions around how we build Kubernetes. Watch the videos to get a full understanding of these. Here’s a few highlights of plans the project has for the upcoming year, based some of the sessions at the summit:
Stability and reliability will be big themes for the upcoming year. This includes making sure that the tests are constantly passing, so that we’re always ready for release, and SIG-Release held a “test deflaking session” to attack some of the problems.
SIG Architecture is introducing a new subproject, Code Organization, which will take ownership of staging repos, vendoring, and code origin. They also want to “factor out” more components, and make Policies an abstract concept so that we can unify it across Node, Pod, Scheduler, and Network.
SIG Network also plans a lot of changes: refactoring Endpoints and Ingress to work better with external discovery and service meshes, full support for IPv6 and dual-stack, and eventually full multinetwork support. Node-local services (“connect to the logger on this machine”) should come soon, as should multicast support. Most of this is being developed outside the core repository to avoid wholesale breakage.
SIG Cluster Lifecycle announced that Kubeadm was now GA, and HA Kubeadm will be alpha soon. Other new tech include: KIND, for running a cluster inside docker for testing; etcadm for managing etcd clusters across Kubeadm, Kubespray and KOPS; and the ClusterAPI in alpha for managing multiple clusters. ComponentConfig (soon to be a WG) is their effort to make the various components work in a consistent way, especially having YAML configs instead of flags, and a kubernetes/component toolkit to help you build them “right.”
If you missed this one, there will be more Summits in Barcelona, Shanghai, and San Diego.
Next Deadline: 1.14 cycle starts, January 2nd
The 1.14 cycle has not begun yet, and the 1.14 team is still being selected.
New patch releases have come out recently to include a patch against a low-impact golang CVE, including v1.11.6 and v1.12.4. Version 1.10.12 will come out soon (as soon as tests pass), and v1.13.2 will supersede v1.13.1, released last Thursday, later today.
A clearly critical feature for Prow (the tool behind our beloved
k8s-ci-robot) is displaying memes, and this week we got a double-header of ponies!
/pony in a Kubernetes ticket will display an image of a random pony, and
/pony gif will level you up to an animated pony. All ponies graciously provided by https://theponyapi.com/.
A small patch, but a big impact for all of us! The NodeLease system will be activated by default for 1.14. This moves the node status system away from expensive timed health checks and on to a dedicated API using long-lived leases. This dramatically reduces both DB size and I/O usage for etcd in large clusters. A very welcome improvement for the future!
Another small but useful change, this time improving the security for etcd with better file permissions on the data directories. If you run your own etcd, maybe double check your current file perms and tweak them if need be.
And finally a more substantial PR to look over, the last remaining bits of the
scheduler/cache package have been moved, either to
scheduler/internal/cache or to the new node info package. Overall it looks like 1.14 will be a banner release for the scheduler, big congrats to the whole sig-scheduling team!