VMware Lightboards - Intro to Kubernetes, 12 May 2017

Previous Meeting

⏯

youtube image

►

From YouTube: Networking with Kubernetes

Description

In this lightboard talk, Nathan Ness walks through the networking options in Kubernetes and explains the types of communications through Kubernetes services.

For more information, please visit the VMware’s Cloud-Native Apps website:
https://cloud.vmware.com/cloud-native-apps

A

Hi, I'm nathan, nuts, from the cloud native applications business unit and today I'm going to talk to you about networking within kubernetes and your different options. So first thing we're going to draw is the is network.

A

We're just going to draw a simple topology here with a physical router and a switch or a VLAN hanging off of it. Okay,.

A

And then we'll draw some kubernetes constructs up here, so Cooper naze consists of clusters, and so a cluster consists of a master, node or nodes. This master node provides an API. It provides a scheduler and there's other constructs in here, but I'm not going to draw them for the sake of time, and so after you have a master, the cluster itself is actually built up of what are called worker nodes. So we'll name this one node one and we'll name this one node two.

A

Okay, these nodes have Ethernet zero attached to them. This can be again physical or virtual.

A

And so we'll connect these nodes via to the infrastructure network. So the requirement here from a node network or RI as network is all nodes within the cluster have connectivity. This connectivity can be via l2, l3 or and overlay just so, just as so as long as they can communicate with among each other. Ok, so after that, we go down to what's called node networking, and so this is where we assign a sitter block.

A

And for the purposes of this talk, I'm going to use one 80216, 8 X, dot X, so the first two octets and my node networks are going to be one 92168, and so you can really think of each of the worker nodes as an IP router, that's assigned. So this one will choose 10, dot, 0, / 24, and then this one we're going to choose. 20.0, /, 24, okay,.

A

Connected to this is the container bridge.

A

So this is acting as our switch.

A

All right, and so we'll play some IPS up here, just for the sake of providing some clarity up here for when we get into the overlay topologies. So will IP this 50.2 and we'll set the Gateway up here to 50.1, and then this one will be 50.3.

A

Okay, so kubernetes works a little bit differently than docker is in kubernetes that you have a construct called pods instead of containers, and what a pod consists of is a grouping of one or more containers.

A

Okay, so we'll just name this pod, one and pods share the same IP address. They have share the same localhost and then they also share IPC. So all of that's contained within a set of pods, so I can have pod one pod two over here with containing only one container.

A

This can be pod two right and we'll draw a couple of pods over here pod three and Todd four.

A

Got some containers in them.

A

All right, and so how do these pods communicate with each other? So a couple of different things that we can do here is the first one is. We can have a flat routed, topology.

A

Right and so in this particular case, these node networks are actually going to need to be physically advertised across the physical network and so in this physical router. In my routing table.

A

I'll have 10.0 / 24 going to 52 and then I'll have 20.0 / 24 going to 53 and that will be installed routes inside my physical router and across my physical fabric right. So this is going to be over l3 I need all of these routes, reflected inside my physical network, okay, and so that's the first type of topology.

A

So when pods communicate across this they're going to actually use your physical network to communicate back and forth, so there's no complexities of port mapping and NAT for pods to talk to other pods right, and so the second type of topology we could deploy is an overlay technology to have pod, 2, pod communication. And so what I'm? Talking about these two constructs here, I'm talking about pod, the pod communication we'll get into how we access these pods externally here at in a bit, and so when we have an overlay technology.

A

These are not necessarily required anymore. So now we're actually going to tunnel over our physical network.

A

And so, when we're looking at this, then we just develop this tunnel here between our worker nodes, and this is going to be our overlay Network, that's going to tunnel the pod traffic right, and so now, when pod one talks to pod 3, node 1 is actually going to encapsulate that packet into this overlay, and so the source of the packet is going to be and the destination of the pack it's going to be 53, and so what takes the place of actually providing this functionality in the mapping instead of the routing table, is actually a key value store called EDD.

A

Okay. So now we can have pod pod communication. We don't have to worry about static routes across our physical network or when we stand up new worker nodes, we don't have to make sure that being advertised across the physical network. Again, our simple requirement now is node 2, node communication and not necessarily all of these node networks across our physical network. So now let's talk about communication and so there's two types of communication and that's done through a kubernetes service.

A

So there's two types of kubernetes services: the first one we'll talk about is cluster IP.

A

So this is going to be the type the next thing that's in here is going to be an endpoint or endpoints, and endpoints is just basically going to give me a listing of pods I.

A

Also have ports what ports I want to expose, so I can do 80 and 443, and so there's other constructs within the service. But this is kind of the basic idea. Is this kubernetes service is going to allow pod pod communication. So now, when we're talking to pods, it's actually going to pod one's going to talk to a cluster IP, and so now we get native load balancing within kubernetes, east-west because when I talk to the cluster IP, the cluster IP is going to give it to one of these endpoints.

A

And then this endpoint is going to be either go through the overlay Network or it's going to be routed physically across my physical network, depending watch which topology I'm in and so that's how pod. The pod communication works within a cluster there's another type of service called node port.

A

So no port really consists of the same constructs as the cluster IP.

A

We have our list of endpoints here, which is going to be pod, one pod, two so on and so forth, and then I have ports that are going to be exposed for that particular pod and I can map these to any port that I want or an arbitrary port here. Maybe this is going to be one thousand twenty two, and so, when I come in externally to any of these 52 or 53 via ten 0.52 : 1022.

A

This is going to route me to my appropriate endpoint and I can go to these endpoints or worker nodes of whether they have the pod that I'm looking for or not, and so what's going to happen here is this is going to get installed in iptables and regardless of which IP that I go to for a worker node perspective I'll always get my endpoint service, so this is going to be distributed across our entire cluster and I can hit any of the worker node clusters inside of that and access my particular service that I want, and so that's how external connectivity happens as well and so again a little recap here we have kubernetes service from a cluster IP level, a cluster IP where we can actually communicate east-west and then accrue Burnett ease from a node port.

A

Another option there would be load balancer where I can actually come in externally and access my kubernetes service and then for networking topology options. I can either do an overlay network which provides some simplicity when I'm standing up or scaling out clusters, and then also I can do a flat routed Network, where I'm maintaining the routes of the node network inside of my physical network and so on and so forth, and so those are basically the kubernetes networking options that we see today. Thank you.