1 of 9

​

Kubernetes Scaling SIG

Update for Community Meeting

Feb 8, 2018

​

2 of 9

Logistics Recap

https://github.com/kubernetes/community/blob/master/sig-list.md

https://github.com/kubernetes/community/tree/master/sig-scalability

Weekly meeting Thursdays 9:30 am Pacific

  • We shifted to biweekly
  • Bumped later by 30 minutes to avoid overlap with sig-architecture

Link to notes and agenda above.

#sig-scale on kubernetes slack

There is a group, but barely used (the action is on slack): https://groups.google.com/forum/#!forum/kubernetes-sig-scale

​

3 of 9

Informal Commentary

  • Small but very consistent attendee list
  • Topics are much more about “How do we build bigger single clusters” to “How do we ensure big single clusters are stable with varying loads and densities” at the 5000 node level
  • Mostly about fighting regressions
  • Community Feedback Question: Is there compelling reason to push beyond 5000 nodes in 2018?
    • If you have a burning need for this, please attend!
    • Ongoing discussion for very large installations about multi-cluster or “federated”
    • Very large installations can be driven by failure domain considerations, not number of nodes per cluster
    • Note - firmament scheduler indicates some efforts in this direction

4 of 9

New Charter

New charter as yet unmerged:

  • https://github.com/kubernetes/community/pull/1607
  • The SIG is fine with it, need ratification or any concerns from outside the SIG
  • Shout out to Marcin Owsiany (porridge) for driving the charter rework�

​

5 of 9

Who Are We (New Charter)

“sig-scalability helps define kubernetes scalability goals, and makes sure that every kubernetes release meets them by measuring performance/scalability indicators and publishing the results. We also coordinate and contribute to general system-wide scalability/performance improvements (ones which do not fall into the charter of another individual SIG).”

​

We provide performance measurement tools, processes for ensuring continued scalability, support and consultation to kubernetes developers and sig-release members in order to meet the above goals.”

​

6 of 9

Our Values (New Charter)

​

We are not firefighters, we are fire-prevention specialists.

​

We promote deep technical understanding of the kubernetes system and our tools.

​

We strive to eliminate toil

https://landing.google.com/sre/book/chapters/eliminating-toil.html

​

We work towards building a scalable kubernetes even in face of superlinear growth of number of contributions.

​

7 of 9

On Fighting Regressions

In the coming weeks, going to be discussing tooling improvements to help with regression debugging.

Example work:

Add apiserver profiling to our scalability tests #59352

https://github.com/kubernetes/kubernetes/pull/59352

�

8 of 9

Final Thought / Question

At the last Kubecon Dev summit we planned for a f2f SIG meeting….

Instead we were deluged with users (not devs) asking all kinds of scaling questions.

Shout out to Shyam JVS (shyamjvs) for an awesome job talking about how we do scale testing. (Hoping he comes to Copenhagen!)

Do we need to do a better explanation/communication job to the user community?

​

9 of 9

Thank you!

Wojciech Tyczynski (Google) wojtek-t

Bob Wise (AWS) countspongebob

​