Cloud Native Computing Foundation KubeCon + CloudNativeCon China 2019 (Shanghai), 5 Jul 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Porter - An Open Source Load Balancer for Bare Metal Kubernetes - Xuetao Song & Fang Du

Description

Join us for Kubernetes Forums Seoul, Sydney, Bengaluru and Delhi - learn more at kubecon.io

Don't miss KubeCon + CloudNativeCon 2020 events in Amsterdam March 30 - April 2, Shanghai July 28-30 and Boston November 17-20! Learn more at kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects

Porter - An Open Source Load Balancer for Bare Metal Kubernetes - Xuetao Song & Fang Du

As we know, the backend workload can be exposed externally using service of type "LoadBalancer" in Kubernetes cluster. Cloud vendors often provide cloud LB plugins for Kubernetes which requires the cluster to be deployed on a specific IaaS platform. However, many enterprise users usually deploy Kubernetes clusters on bare meta especially for production use. For the on-premise bare meta clusters, Kubernetes does not provide LB implementation. Porter, an open-source project, is the right solution for such issue. This session will focus on the network technologies to help expose service and EIP management for bare meta Kubernetes

https://sched.co/Nrnw

A

Good morning, ladies.

B

C

B

A

Actually soldier Tao from the Beijing unified technology, company limited and I'm an engineer there and, as for my colleague, do found he's ill today, so I will have to deliver the speech by my own, and today I would like to talk about Porter. Then it's actually an open source, no balancer for their medical bananas.

A

Introduction, first, why do we need to use porter when we use k8s if we want to access the.

B

Service there we'll.

A

Have to use service.

B

Lightly but as for service, IP.

C

A

Actually, a virtual IP.

C

A

It can be reflecting.

B

In the following.

A

Aspects, first of all, for service I think it's actually managed by should be proxy. You know through IP psy PPS.

B

And the service IP.

A

B

Only be obsessed in.

A

Cluster and also.

A

The service IP is not bound to any device. There's no service IP any of the Gateway.

B

A

Wanted to look assess the service or IP. Therefore, k8s has enabled us who access the service.

B

First of all, the.

A

First, one is not important: it's actually the commonly.

C

Mostly commonly used to.

A

Expose service to external and just to do a mapping walk to the terminals and through the hosts IP as well as the terminals. We do the access of the service. But it's really hard to remember because it combines hostile. He, as well as the other dress and.

B

A

Really hard to name that and also secondly, the single point as well as the bottleneck problem, can be very random. It can only be operated in the test environment in this way and we are going to access the port. Of course, we have a host IP, but if the house is down, then the user can accept the service anymore, I'm gonna, even F CKS.

B

A

Still enable the normal service, but externally, when the host is down, there's no way for you to again access the service anymore.

B

A

Secondly, stress- and it's also the frequently used method to.

B

A

His service, first of all, I, would say we'll deploy the service by domains in this way very understandable, but they still got armies for ingress to work properly. We do need to expose ingress to external and there's no ingress. Unfortunately,.

B

For the ingress controller and secondly,.

A

It's actually for.

A

B

A

Use no pot or housework to external to expose that external again, it's using.

B

The host to obsessive service again, if.

A

The host is down, the ingress is down, but still this service is up and rounding still it's actually maintained by developers, and it can still be used in this way and.

B

Thirdly, for ingress.

A

We can only use.

B

A

B

Ingress use engineers.

A

There is smart allure support, called stream, which has four layers, but it's in the test period is not a stable, therefore, for HTTP can be used to expose the ingress to external and as well out there kind of service. We soda music ingress.

B

A

Called load balancer for a little balancer.

B

A

Used by cloud technology promoter, we use the.

B

Ip to bound to our service.

A

All we need to do is to deploy a k8s. Then we can use the low balancer to expose the service to external, it's very convenient and user-friendly, and also you can attach any domain to the IP, and it can be that's very easy to remember.

B

A

Again, we use the.

A

Provider, therefore, there will be no single point.

C

A

As for the traffic, it will.

B

Be distributed.

A

Evenly to the notes by the technology provider, therefore, there's no single point. Thank you and.

B

It's for the con of that it can.

A

Be only available in that provider only, for example, if there's I mean they're metal, there's no way for you to use the balancer.

B

A

A nutshell: it's for testing, but in grass it's convenient but again.

B

We need to find another way to expose.

A

That to external, as for no port, really sorry, no balancer, if they'd say private clouds and there's no way for you to use that, and that's actually these. The finish of that. That is the reason why we news we need to use porter. Judging from its name, we know, is to some water.

B

A

Its it means packaging.

B

Engineering aids, english translation.

A

Its balancer on, without being relegated to the fair amount of kubernetes, it has basically three functions as you can see on the screen.

B

A

Traffic load balance, access.

B

Management and open.

A

Source and some.

B

Fundamentals about that we use ecmp, it's actually a part of the function of linux.

A

Judging from its name, it has the same destination, enhance different pathways, but all the perhaps are actually equal to each other, that is to say well we're going to the same end point destination. We might have different paths leading.

B

A

Same destination and basically that's a logic behind and it can be able to see that from the picture below, for example, there's ruler. And if we want to go to this elimination.

B

There's two three.

A

Paths, one point one point: one point one point one point one to include one point: one point: what would three three V and although this.

B

Kelly do the same. In fact, any of the.

A

Failure of any note it would not really interact or misusing higher service with a CMP in place. We.

B

A

B

Use PGP to.

A

B

A

Called order gateway protocol and.

B

I would say it's.

A

Actually, a cool active, different properties, for example. It was support multipath.

B

A

General Reuters: it will support that in the form of multipath. It's actually the same with a CMP. It will allow us to.

B

A

Rotors interchanges to have multi path, leading.

B

To the same destination as.

A

For BGP it will choose one of them and to achieve the load balance.

A

Oh and this for this picture demonstrates.

B

How BGP really works.

A

B

A tiny Network, for example our user. It's.

A

Out of the cloud and you want to access the one point, one point one point one and you can see different layers of Reuters, for example a border and leave a layer.

B

A

Two BGP: how are we going to access the surface? First of all, you can see the service at the left, hand, corner or bar at the bottom. That's actually the servers you can use. For example, zero point two is reporting there and it's relief. It will deliver the information to the next level and again they are two rotors. They will be able to get the information to access.

A

One point: one foot, one foot one you'll be able to go through at least one and look and again the spine, the information to the border, and if you want to access one foot, one foot number one and then you can just go through this point. One spun two, and in this we can see the XS to one point. One point: one point one can be cleared in its constructor in this way and we'll be able to do the connectivity connectivity here and you can see altogether.

A

There is two times two times two times two or together sixteen pathways here and you will be able to see not even for the two or three service in one or two layers, or even some of them are actually down. It will not really interact or seriously affect the entire network.

B

That is to say any.

A

Failure off the server or brother will not really affect or family damage. The success.

B

Eventually and now, let's see how the two rationals can.

A

Be used to do the constructive work first of all for Porter, it has two components: the first most evil set. It has BGP server inside for deficit, it's actually the brain, it's basically dispatching booters as well as IP management.

A

Incidentally, we got agent as well. We have a diamond set method to do the deployment to indicate us cluster. For example, if there's a new service that needs to be exposed to the external, then we do need to maintain some of the rules locally and it's for these agents. They would be responsible for the making draws and it's really lightweight.

A

And it's for the porter: we have this er D plus controller, this kind of standardized k8s.

A

B

Be actualized and.

A

We can use this watching page. This kind of made of Kate has not thought to do the job and the shiny point with a pinch.

B

Of that is that the ETS can entrust the IP.

A

And if the user.

B

Want to access.

A

The porter they only need to coat some gold language and then just to access the suburb and also to come control, can use the order line to control the porter.

A

A

For a load balancer actualization, we got a front-end load.

B

Balancer in.

A

At the back end, we have the not part of the host or not part of different posts to achieve a low balance, but in a difficult but not world, it has a large remark. That is.

B

A

High this source IP that.

B

Is to say, do not.

A

Except on the port, wouldn't be able to know the.

B

Source IP brother.

A

It will be able to only see the NOS.

A

External application that requires auditing it's really inconvenient because they can only see the not.

B

Write P of a lot of noise, but.

A

For aks it has such a mechanism do.

B

A

B

If you want to randomly access an R.

A

And s for the not can be accessed by node, but there's no love, then there's a dead end and it will be a timeout without any response.

B

In it's not good, and in.

A

Difficult we're.

B

A

Used not to randomly to show you the information to the court with and with the board, we'll be able to open up this service.

C

And to do the setup.

A

And use no count to ensure that locally. You always have your point and that's where this function can be controlled by point and it's report will watch all the end points and at.

B

Some points if.

A

There's any sweep about that then, for example, from 1 to 2, then the what once Reuter will be deleted, and then there will be a second Buddha. In this way the front-end load, balancer will always have access with, and service will be able to set up such a property, and in this way we might only have one less routine. At the same time, you will be able to see the real IP and as for the application that requires modeling I feel like it's really good and.

B

A

B

A

Ip management for load balancer, we have external IP plus.

B

Cos I piece mapping and there's no way for clients.

A

To access host IP and we have the external IP.

B

To access the service and that's.

A

The control gap for second service- they will be able to.

B

Access these external IP IP, it.

A

Should have to be.

B

Controller manager and as.

A

For the entire LAN, it will be spreaded and there's no way for us to could an IP in the LAN randomly it will in this way, leech conflict of IP address or even chaos. It will be a.

B

Huge to inserting display.

A

For the air and therefore will have these kind of function.

B

Theory points to.

A

Use dynamic or static setup of IP, for example. If you want to have the access to the service.

B

You can do this that.

A

Can do that in a static manner, of course, and do the England embankment with dynamic and linear smoke? You can just have the pool in place to tell the IP, not all the IPS in this pool can be available, and then you just do them in ten and Square.

B

A

And as long as the service use our port reporter, then they will be able to get one of the IP out of the pool and secondly, for CN space in.

B

C

To use all those ideas in the domain, which is similar to chemical assembly,.

B

C

B

We can expose the availability and member of the resources that can be used externally. Maybe hey check.

C

B

C

Ability, as well as the traffic.

B

C

For the other part, this is the one that we are going to develop and it is now in our plan.

C

Complete working.

A

For the first service.

B

Would be start from service a.

C

Balancer or not other to.

C

Monitor the balancer, a circle and indiscipline p.m. from that I don't want to.

B

Be her first check whether is the take your.

C

Dynamic, if not, we can acquire.

B

C

From the pool to live.

B

To the service.

A

Since you update.

C

The service Oh once.

A

Because the IP we.

C

B

We don't be serviced at being sure.

C

That the internal I'm here yogurt the routine, can go higher order.

C

A

C

My know which marketing the whole services and the location of the portal.

B

That expose VG service. You arguing that the passed.

C

Through all those not.

C

B

They will be broadcast.

C

B

C

The network in.

B

C

Way the network's would know there.

B

C

That is maintained by PG mean.

C

B

Sam or agent well do some more over.

C

Here any to the external routine in issue of a leave.

B

To the internal, because.

C

It's not connected with subtle, hold fast. The whole.

A

C

Allow me to join two men. We need your from Mary table. This is an approach for all the table. We have to open it. How.

A

It, although this package.

C

Is not value being still need? This.

B

C

And by a cop Rossi, we were hi know it later, and the st. Louis is no video art and figure. It's also.

B

C

For the typical fingers its function is dutiful, PGP is, although.

B

As a barcode, it's a fraction of the four there.

C

Are many whose PD really horrible many features each even.

B

Configure we need, we can dynamically.

C

Configure all of this one with you and watch us to see Rd and missus erd. If there are changes we can dynamically update the server. If we you want to add in future events or the.

A

Video server and we.

B

Can a brother can't stop certain things we can keep.

C

This by this kind of.

B

C

B

Achieve the 0.9, the.

C

Portal is always in operation will do not need to stop it to maintained.

C

We use the water quite easy, don't know this download.

B

The Yahoo get hard and.

C

B

A

Ugly, because it's not that I'm very good after.

C

It is in operation, you can added the bgp collaboration. Did you pick a figure.

B

C

Are some certain kinds of formats.

C

B

C

C

Porter and here at all, not to mention that normally.

C

You mechanical.

B

Also use bt bewdley, it.

C

B

7-9 connector.

C

B

For some, it only.

A

C

The end-to-end.

B

179 service, only this one, all of what I will why.

C

Are some major.

B

Principles do guides and.

C

Shifted the porter and achieve the ghost you to be very competitive and in operation with other religious.

B

Because we didn't use the standard 179, we need to tell the porter again use the standard, while in.

B

C

To connect the exchange machine.

B

During function set.

C

For the address of the net gate.

C

B

Is telling it we need to force and not about.

C

B

A

Service, it would only receive one route.

C

It will be goes to one destination, they feel this was a.

B

Permissible routine, yes to.

C

Achieve the function.

B

On the right eye,.

C

The Cisco configuration because.

B

Routines of different.

C

Manufacturers and.

B

C

A

Can be found on my github.

C

Including other collaboration.

B

From other manufacturers.

A

B

The figuration is completed.

C

The portal can be input into operation.

C

B

Order to put it into.

C

Operation, we need to add that map into the pool.

B

Under the control and.

C

Address too sappy address at.

B

This I'm here is an address.

A

C

The I came home to ensure that it's unique game in this local network.

B

Operation create a service.

C

We sanitation, because you see, on occasion a notation to town system to.

B

Use the plural for.

C

Loading balancing to tell it no other service.

C

Is to be suitable for well balanced in this way,.

C

For seconds later, we are happy because it takes very few seconds to achieve all those functions.

C

A

C

B

Enterprises, they have very strict requirements for the control.

C

Need to go through the net gateway and we would.

B

Like to provide more.

C

Flexible IP, for you.

B

C

Would like you to use other bottles or.

C

B

For example, daddies I do not support.

C

Ebgp, what we need PG be flexible in configure.

B

The policy every time we need to send the routine to the three exchanged machines here is the same complex. That is the best several.

C

Resources and also use.

B

C

In this kind of and.

B

Because domestic enterprises.

C

B

Misunderstanding Yahoo BGP.

C

On dates, talking to do, BGP is safe to use well, I'm, not sure why they are thinking about and.

B

C

That we can support speech in a policy to control resume advertisement, so I.

B

C

B

This is my address, a key topic.

C

You are interested in.

B

C

Welcome you to try interviews and give us feedback hello.

B

C

B

C

In village, thank you.

C

We have any questions.

C

I would like to ask one thing:.

C

C

Nation algorithm: it's.

B

C

Algorithm, do you support for some specific scenarios.

B

C

Didn't handle those are complex. Business means that one service, if you use the pool order, we need to know IP because it cannot be built in the internal cluster. In the sub principles we receive a package and normally destination is with a certain time. You just drop this course. You will make this process easier.

B

C

There are many backend it needs to select one right, then it will readily be distributed to one.

C

Mission, what has to.

B

C

B

Me that should be configured to have the Rooter.

C

You need to come, make the configuration under notice. Normally, the BGP have this kind of features.

B

C

When choosing the.

B

C

How to there was the consistency.

B

Of this process.

C

So we go drug-free from La. Ruta is our.

B

Or no balancer.

C

We use the rotor as the balance, so everything should be configured in the rotor and also because the audience didn't is not using the microphone. The interpreter cannot hear clearly.

C

It's that evening.

C

And I think I understands this question.

C

And any other questions.

C

B

C

Your work once BP and second is second layer Network. So one awesome.

B

C

Reporter is, after.

B

C

In our team, we have this research online to see whether they.

B

Are similar competitors- and we found out this tactical- your.

C

Biggest difference.

B

Is that first there they'll be? They have all the servers. All the video servers.

C

B

Video supports very few reaches functions, although they have the server a whole other nodes. Is you very happy so good hobbies.

C

Native, we are based on.

B

Cr deed, there are.

C

Based on a finger map.

C

A

That configure Merrill first.

C

That is the biggest difference.

B

C

Are no questions I'd like to thank you again for coming to this session.