Cloud Native Computing Foundation KubeCon + CloudNativeCon Europe 2020 - Virtual, 4 Sep 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Keynote: Kubernetes Project Update - Vicki Cheung, KubeCon + CloudNativeCon Europe 2020 Co-Chair

Description

Don’t miss out! Join us at our upcoming events: EnvoyCon Virtual on October 15 and KubeCon + CloudNativeCon North America 2020 Virtual from November 17-20. Learn more at https://kubecon.io. The conferences feature presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects.

Keynote: Kubernetes Project Update - Vicki Cheung, KubeCon + CloudNativeCon Europe 2020 Co-Chair & Engineering Manager, Lyft

https://sched.co/ZfDC

A

Hi, I'm vicki, and I want to talk to you about what's coming up in new versions of kubernetes, so as I'm recording it today, 118 is the current newest version, but around kubecon time we're also looking at a coming up. 119 release so I'll do a little sneak peek as well.

A

So in 118 we saw a lot of features moving to stable again, so this is really uh as the team calls it a fit and finish release. We also see 11 features moving to beta and 12 new alpha features.

A

A couple of things I want to highlight. There are a lot of exciting storage features that are released, and there were scheduling, features as well that I will talk about in a little bit.

A

Two other new features, I'm quite excited about are the horizontal pod, auto scaler controls, as well as a coupe cuddle debug, which we talked a little bit about in a keynote last year. That is finally happening this year.

A

Okay, so storage, I think one big thing that I am excited about is the raw block device support graduating to stable, so if you're running any more sophisticated, stateful workloads, if you're running your own database, that has high demand on performance and latency on, I o that might have its own storage format.

A

This is the feature for you.

A

We also saw cloning graduating to stable and a new feature called generic data populators that allows you to get pre-populated volumes.

A

um This is super useful if you're doing backup or restoring, um and we also see immutable secrets and config maps for any sort of uh safeguarding of uh the data that your application reads um and also csi support on windows.

A

In terms of scheduling, we have a lot of more sophisticated controls. I would say so right now, or it's always been the case, that you can run multiple schedulers on kubernetes um and you could run into a problem where um your different schedulers are out of sync.

A

Essentially, their views of the world are different and you can run into race conditions where two schedulers maybe try to schedule things onto the same node at the same time and so to overcome that now you can say, uh run one scheduler but have multiple profiles, so you can still satisfy needs of different types of workloads that need to be scheduled differently, but you can have one scheduler that takes care of the scheduling, so you can avoid these types of issues with race conditions.

A

We also see taint-based eviction graduating to stable. I had to look this one up actually, because I thought that I've been using it forever and it was a feature that was released in 1.6, but it is finally going into ga and last but not least, we see pod topology spread graduating to beta.

A

This feature allows you to control how your workloads are spread across your different um topologies. So, for example, if you have a multi-availability zone cluster, um you can say how you want to spread your deployment across the different availability zones.

A

And if you have a anything, that's slightly more sophisticated deployed on your cluster, that is auto scaling.

A

You might find yourself uh sort of wanting for more finer grain control over how your work real workloads uh scaled up and down. So now you can take advantage of this.

A

Horizontal part of the scalar control that is now an alpha and here's an example uh on the right where you can see the new yaml structure for uh defining the behavior for how you want to scale up and scale down, and you can really dig into you know the rate of scaling if you want to scale at different rates based on the current replica size- um and you know, define multiple policies if there are different rates for different stages of scaling.

A

These are all definable now.

A

And coop cuddle debug. This was a feature that I was super excited about last year when we released ephemeral containers, so thermal containers were added in 116 and just to catch up on the context. So when you have a pod, it's immutable, so you you define the list of containers on in the pod and you deploy it to the cluster, and um you know if later you just you're like oh there's, an area in my pod, and I want to like debug it.

A

You can't just go in and um you know and install something new and be like. Oh, I need a different container that has more tools to you know if you want to like tcp dump or something and you're running an alpine image like.

A

Oh, you gotta, like redeploy your your pod again um so now you can have ephemeral containers that essentially are containers that you add on later to a running pod. um So here's an example.

A

If I want to have a shell in my pot and none of the containers in my pot have a shell installed you can have. You can attach, for example, a debian ephemeral container into the pod and um and then run the exec again, so so the coop cutoff debug command really just like wraps that up for you, so you can just say like. Oh, I have a pod. um Let me run code demand deep cuddle, debug and then it creates the thermal container attaches it, and you know you have a nicely.

A

Have a nice debug workflow for anything, that's already running in your cluster.

A

And also very excited about the controls you can get over how the api server handles requests, not super proud to say it, but it's possible that my team had previously been dosed by one of our own applications spamming the api survey. um So now we can prevent that by defining this flow schema.

A

You can say so right now you have controls over uh how the api server is protected from requests generally. But with this new feature you can say uh give different priorities to different types of requests. So, instead of saying when you're dost, you protect your api api servers from going down by rate limiting all requests, you can say: okay, user requests should go through so that admins don't lose access, but maybe some other types of lower priority requests. Maybe from your applications, get rate limited.

A

Another feature that uh graduated to beta is the note topology manager, which again very excited about from uh running machine learning, workloads, point of view. um So a lot of times. You want to control how your workloads are scheduled on the hardware itself.

A

uh In this example, there are eight gpus on the server um some of the pairs of gpus have higher bandwidth connection and they're faster if you schedule on those two together.

A

um So if you have a gpu workload that uses two gpus now the node knows how to optimize and say, give you a pair of faster gpus rather than two that are maybe crossing um crossing the cpu boundary and using a lower bandwidth connection.

A

There are a lot of other features I didn't cover. Ipv6 graduated to beta, a lot of people are starting to use the certificate. Signing requests, api um and also you can now run dry, runs on your on your request, so it used to be, uh you would run them and the the diff would be done on the client side. With this feature, you can actually do the diff on the server side, so um you get a more accurate view.

A

And, of course, without our amazing community, none of this would have been able to be shipped and realized. So thank you so much to all of our contributors, and it's so amazing to see the trend of uh the number of contributors happening, release over release and the community is just so active and so wholesome.

A

A

And then one last thing for the 119 sneak peek. I want to call out a few features that, as I was scrolling through the beta release notes I got super excited about.

A

One is generic ephemeral volumes so now you have more flexibility than just using empty dirt and there's ipv6 dual stack: support on windows and then even more to the coupe cuddle, debug workflow. You can debug notes by running in the notes, host namespaces and last but not least, there will be support for a secret v2, so everyone who's been complaining about it. Until now you can relax now.

A

Thank you. So much.