Cloud Native Computing Foundation KubeCon + CloudNativeCon North America 2022, 11 Nov 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Lightning Talk: Where Did All My IPs Go? - Cynthia Thomas, Google

Description

Don’t miss out! Join us at our upcoming event: KubeCon + CloudNativeCon Europe in Amsterdam, The Netherlands from April 17-21, 2023. Learn more at https://kubecon.io The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects.

Lightning Talk: Where Did All My IPs Go? - Cynthia Thomas, Google

Kubernetes cluster planning requires quite a few things to get started. What about IPs? Common IP management hurdles with Kubernetes clusters include IP assignments when building a cluster and challenges faced when deploying in a multi-faceted environment. Kubernetes Admins often need to use IP addressing handed out by Network Admins juggling other non-k8s workload IP assignments and IP exhaustion. In this talk, Cynthia will discuss new and existing KEPs that SIG-network has implemented to help mitigate IP challenges. Such features include discontiguous cluster CIDRs and the journey to IPv6. Cynthia will also discuss how the best practices for Kubernetes IP management are changing with these new capabilities to help scale and grow instead of rebuild.

A

All right start the timer five minutes all right thanks everyone for joining me today. My name is Cynthia Thomas I'm part of the kubernetes networking team at Google. So let's talk about. Where did all my IPS go when we start building kubernetes clusters, um a lot of you're gonna, rather Love Me, Or, Hate Me after this? Oh, my slides, look different. We like yellow eh, okay, so why do I need IPS?

A

um First of all, kubernetes use Internet Protocol for addressing for network communication and that's for two fundamental networking reasons uh that kubernetes clusters dictate so the first being Paws need to communicate with other pods in the cluster without using that or network address, translation and also agents on a node need to be able to communicate with pods on the same node. So we use you know IP, for that.

A

There are several implementations of Ip addressing depending on different architectures and those can be attributed to different networking models flat air gaps, Island modes, but that could be a whole other talk. So let's talk about. Where did all my IPS go when I started building my kubernetes clusters? So what are those IP requirements for a cluster? Well, there's at least three big IP blocks that are required when you're building a kubernetes cluster, the first four pods also for nodes and then finally for services.

A

So, since the early days of kubernetes, we, when we build a cluster, we assign large blocks for these kubernetes constructs I have a great way to adjust cluster-wide allocations back in the day, but this actually in turn caused organizations to face challenges when they assign networking IP ranges and such to kubernetes operators, because private adjusting can be a great.

A

You know, fragmentation problem within their organization, taking into account various types of existing workloads that already exist in consuming private, addressing and finally, when you think about it, what happens when you outgrow those blocks that you've assigned to a cluster? What's a good way to move and and grow your workload? Do you have to build a new cluster and migrate over? This is something that's not easy for kubernetes operators to handle.

A

So how have we been making it easier in the kubernetes community? Well, the first is single stack, IPv6 and ipv4 only support but, namely IPv6 only support to tap into a whole new addressing scheme. Ip address family, that's mostly untouched minimizes IP conflict, uh but what, if your workloads aren't ready to be V6 only it was difficult to have. You know a migration path to get there. So next and most recently we have introduced dual stack, ipv4 and IPv6 support for kubernetes clusters.

A

So this helps you know the story of migrating to V6 only ensure you can reach V4, only Services still and make sure your application is ready to move in that direction.

A

Finally, we've also introduced multiple cluster ciders support for node ipam. What that means is you don't no longer need one large contiguous block for your pod ciders. You can add discontinuous blocks and continue to grow as your your application, as your workloads grow on your cluster, so woohoo so which caps are on the horizon that are going to help us. We have the first being multiple service ciders. So what does that mean? Similarly to the multiple cluster ciders? Now we can add and actually tune the allocation of service.

A

Ciders, so we can change that cider size. We can actually tune the IPv6 default so to meet some of the best practices on the IPv6 side, so something along the lines of what I expressed earlier with the podsiders.

A

uh The second is actually reserving static and dynamic allocation for the service. Ip ranges. So this comes in the scenario of minimizing IP conflicts when you might have specific services that need a static or specific service IP. So this new cap is actually helping Define. What that service range looks like in terms of balancing you know the fraction that it's reserved for static versus Dynamic application allocations.

A

So, where will all my IPS go well with those changes that are I've mentioned that we've introduced and are up and coming? We can think about the way we move forward for Best Practices when building kubernetes clusters. So what does that mean? Let's think about maybe starting with a smaller application allocation for these IP address, ciders and slowly, building and adding dynamically new ciders for the the relevant ranges and then build up?

A

uh The second point is yeah. Kubernetes is actually making that push for IPv6. This is what we're actually seeing more and more organizations want to move there, uh and this might just push what we've been talking about for decades in the networking Industries moving to IPv6, which is really exciting and I'm out of time, but before I take off I want you to get your thoughts on what do you think about mixed mode uh service, so a mixed mode service?

A

We want to hear back at the networking Community whether you'd like to mix single stack and dual stack for a service, so you know that's a limitation to do so. We love your feedback on that and that's my time. So, thank you very much.