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

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Observability in Service Mesh Powered by Envoy and Apache SkyWalking - Sheng Wu & Lizan Zhou

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

Observability in Service Mesh Powered by Envoy and Apache SkyWalking - Sheng Wu & Lizan Zhou, Tetrate

Service Mesh provides a new angle to provide observability, no matter the architecture or language. In the traditional way, we needed language agents or SDK to observe the application server status. Since service mesh provides full control of RPC, observability is much easier to be added without language specific technology. In this session, we will demonstrate an open source integration solution based on Envoy and SkyWalking. Without code injection technology or Istio Mixer, we could build telemetry from Envoy and analysis in SkyWalking, with good performance. The user could get the service topology map, metrics graph, request detail and error message, with a very nice visualization.

https://sched.co/NroB

A

B

Can start now, and today we are going to talk about the mixer.

B

C

B

Another perspective, apart from without.

D

A mixer server how.

B

D

B

C

B

Service manager, company.

B

E

And to tell you what kind of services and sticks.

B

Why, using often more and I would just tell you about how can we achieve their point, some abilities regarding it.

B

I work in in touch with I was born, so I'm not accustomed to dinner a speech in Chinese. This is my second time so this is my second time doing. The speech.

E

B

So, first of all, I'd like to introduce what is envoy I think. Maybe some of you are.

C

B

With one, it's a mismatch, and what is that boy talk.

C

B

Observability of Amoy and also I, will show well just to tell you how to connect my in stable to Sky walking. First, what is service mesh.

C

B

C

B

Connecting the network and to make it transparent to applications so many of the common functions needed in the network, including the connection.

C

B

E

B

Can be done, young boy and someone can just replace and many of the following applications to achieve the function needed in the network. So.

C

B

For the general service mesh.

B

Well, a common thing is that the SEC works to use the otherwise test set car proxy and to put in every.

C

B

Service and applications through these kind of envoy service proxy to communicate.

B

C

B

Is another process.

C

B

I know we should ensure C++ and to achieve the high performance with low latency code base. It has many future architectures and also support with HTTP l7 filter architecture and also, as for the service, discovery and kind of observability. What is the emphasis we gonna put today.

C

B

Well, as the authentication, authorization and edge proxy.

C

B

These are the general architecture of avoid.

C

B

I see it in other places,.

C

B

Listen, the future.

C

B

What kind of filters we need in this, and this can provide more information to it,.

C

B

The kind of the encoded theater and to do it in different ways to encode this kind of linkage and we will use TCP.

C

B

Our support service in.

C

B

B

Our seven repeaters, so actually, once you understand the l7.

C

B

Will provide more observability, for example, how many requests comment responses and that login will also be much.

C

More improved and.

B

These are one character upon what that is done, API, which is a control plane and to provide a different configuration for my listen discovery, service cluster discover service under root, discover.

E

C

And in terms of observability.

C

B

To so is it one is metric service and another one is access log service. Also, we have the trace service.

C

C

B

Fill in we haven't fully finished it.

C

B

Today, I will mainly talk about the metric service and access log service.

C

B

Achieve the runway in the sky, walking connection I.

C

B

Observables in now, let's first look at the metric service, actually which is owari in this service just an end by one- and this is that restrain, and we can see these metrics.

B

Using structure from primitives.

C

B

Means that which I'm one in which cluster? Is this a more engine and fighter.

C

B

For the exit lock service is very easy to is similar.

C

B

Just HTTP and access log entries as an example which we.

C

B

C

B

Can add some of them a.

C

B

C

B

Also to add one and in the fire, this is.

C

B

Shrim introduced lock service.

C

B

I named sky walking, we have Emma metric receiver.

C

B

C

B

A function of sky, walking.

B

C

B

Support the metric service and access log service, so we want to connect enter into the sky.

C

B

You just have to send to these service access.

C

D

B

I've mentioned and access box Evans mainly in the HTTP.

C

B

He ever listener.

C

B

Connection manager used.

C

B

Be realized the nights I'm like doing why into.

C

D

C

B

How we can receive these data nice guy, walking.

B

We can simply find them we're just going to take one Easter and Y connection.

B

Regular information, as for the XX log.

A

B

Information we usually use ISO to.

A

Manage to kind.

B

Of orchestration of so what is tracer our matrix or our many topology by many of the producers.

A

B

The end of last year and the beginning of this year, in somewhat open source community to post these adapter, even to do the mixing that we're going to do the implicit adapter you will find about. Actually instant team well not clearly receive these P art. They want to tell you among you to achieve five percent of what is five has adapted wood.

E

B

Still, one filter to send data to the mixer server and, according to your information, to.

D

C

B

At least to bypass so actually sky walking is the only one doing there to bypass. It's got adapter and open source.

D

We mixed a servo doesn't do anything it's kind of more like just route.

D

Most people, the performance of makes adapter.

A

D

C

D

Performance concerns instance to NYU filter cost. It has more CPU, it needs more CPU of invoice and if you actually look at this so most of complicated.

D

Information inside.

D

What are the sources and destinations and in the mixture, so the match right? You could actually see you would receive the information about the source and reservation.

D

A

D

Another problem is latency of our PC. We need to actually have this issue would be hyper procedures and latency.

E

D

Cannot be avoided.

A

D

Elisa actually has down a lot more than zipping and in terms of the tracing in a mixer is more complicated. You can see that the elementary it collects more data.

A

D

Concern is that one is still one story: users.

A

D

Use to bypass, but there would be unnecessary cost when using bypass.

A

D

It's just a little a bit a little more cost than the.

A

People would still consider it.

D

Unnecessary. The third concern is a steel filter, server.

A

D

Kind of like a protocol is quite a complex anger. We have these complex, attribute, URI port protocol protocol.

A

D

Estelle's team is trying to you know where the complexity, but.

A

It's still not user-friendly.

D

We say that we don't want mix them and what a steal thanks down, we need to do in terms of the ability and visibility is like in court matrix or not, and the tracing.

A

D

Tracing this, it's not mysterious, so we all know how to do it. I know when you tube, you have SDK and then I need to have agent, and then we go.

A

A

D

There are whether it's Beach, Prometheus or other systems. It means not mirror so the matrix of notes analyzed and then we also have topology so mixer actually provides much more powerful probability in terms of the.

A

D

So, on the basis of premiere, actually, you could get all the antibiotics and your application can get you to a PM system, including Prometheus or other systems, but if we do not want those and how can we match if we don't change anything, how can we get a topology that we want and all the values of the matrix want.

A

D

If we take out mixer what are the other options.

A

D

Let's go walking, there's.

A

D

Source project and you can plug in for your tennis insight kind of with Yama what yum is going to do, and also through tracing and analyzing tracing.

D

Foundation until today it can do or the tracing and details of tracing- and this is just open-source.

A

Apm system: that's the first option. Second,.

D

So we can actually implement a steal, attribute protocol, which is quite a complex I, mentioned.

A

D

D

It's just relative, but not mission impossible. So.

A

D

Up front ago, an array and then if we can implement that is to attribute protocol the bution, it's still working.

A

D

The East else of future, probably.

A

Cannot be interested and.

D

The impact on latency and set for us till then the third option is LS.

A

D

A

D

If whether you are being like, if you are familiar and boy or other gateways,.

A

D

Log is actually the long service.

A

D

That's not a fire, so it has access log and accessible.

A

Service actually.

D

They're saying.

D

E

D

So amboy, so you can see they actually answer. You.

E

D

Have two different rows.

A

D

Ingress those two different.

D

Topology and decide call is the application behind it and it's kind of like an agent and then, if it's great way, I'm going itself is to service want to show the topology they're. The party that you want to have is totally.

A

Different from you.

D

To solve is is to tell me that it's.

E

D

Or sidecar but.

E

Anne Tuohy II actually doesn't know, but.

D

E

We still have any steel, it.

D

Stai was a pilot when they named every invoice. They have quite a.

A

Regular pattern so.

D

For instance, the pattern includes name of the world and actually they contain very important.

A

D

If it's ingress.

A

D

Is named the rotor if it's citecar, it's named like it's like Carl mm-hmm, so you can see the naming pattern here and that you realize what actually what it is and.

A

Also, we have different requests and.

D

We have inbound and outbound and quite inside that.

A

Car and it's also.

A

D

Still is controlling em boy, that's like if we just run a normal Embry. It doesn't look like this.

A

D

The inbound advance is that, all and like we tell em boy which is out then, which is employed.

D

Actually, we can activate it anyway,.

A

D

Then, if we want to actually make that a visibility work, we need to have like this mash and follow all the rules of Israel or the controlling rose and the Winston's naming and everything you can not have your own invoice that you can assume that it would work if doesn't work that way. I.

A

D

Skip the previous slides, okay.

A

D

A lot of things are specific on the Cuban, a test platform that you see.

A

A

D

Have one a very special thing: is that something that you would.

A

Normally see so this is the data.

A

D

That you could see.

A

D

Have the concept.

A

A

D

What you need to do is that you were slow, IP, n and a for the same service, mash and kubernetes.

E

D

The features so it is like.

E

D

No specific autonomy, so that would not be shoe. Ports sharing one IP, so every IP is specific, is distinctive and you can, if you can have.

E

D

A

D

Not what we are looking for, so if we look at any monitoring system, we can only receive IP of the service and.

A

D

Don't know the progress of that.

D

Actually, the IP is the most important to think we need to get the IP to locate what is the original service and like, for instance, the data of.

A

D

When we try to pour out of the data.

A

D

On each topology match the.

A

Data is generated.

D

A

D

To tell you specifically that there are different cases in Christ, so it has a special row for every incoming request. It will have a specific LS, so in honor topology.

E

You will have two lines.

D

Ingress, it's an opposite service. You need to have this kind of like you to visualize its incoming and outgoing, to make it to make sure that it is an independent service and then server-side cycle. So it's service side, matrix.

D

We can show that who is pulling out.

A

D

Was collected in the day so this something specific that when you.

E

Use ingress to pull up this like car ingress.

D

Is it will actually go through.

A

The ID of the address and you.

D

Cannot generate this mind and on this line, this will only have a ingress for the target service and the service cannot trigger the ingress. So this is from only.

A

Can we achieve the.

D

One side and the client desires like car is the same.

A

D

We just talked about how we identify those it might abstractive. Then we also have this kind of like open source about sky walking. So you can see the project we both here right, visiting this website and order matrix and website.

D

We want to share.

A

D

A

D

Topics that's related or any classroom, so you want to ask: please: don't have any to raise your hand.

C

D

Actually I do have a question. The first one.

C

D

E

D

D

Can you also cover the observability of the nodes and can you yes.

A

D

It's actually the.

C

D

The second question I have is that we cannot have an like just a standard. Employee you'd still need hold.

C

D

C

A

D

A

You don't need to filter. Actually you.

D

C

D

For isto itself, we still proxy have this m'boy and also the mixer shorter, and we don't need shelter for that part, as we should mention for this, the naming rules and Patchin and everything the identifier. We need to give it to you, the m'boy, and if you use other m'boy, you will have you don't have a consistent naming patterns or rules. So we want to use a palette to identify us to achieve the result. So.

C

C

D

Need to replace it, the current and boy that I am using to the standard Android. Was that like that right? So in theory, it works, but we haven't tested so, but I think you don't need to change it into employ because there's no a steal filters.

D

C

Can I avoid the cost or lower the cost of filters.

C

D

In the configuration there's, if there's no, it's your future.

A

D

You can just turn off and it will not generate any cost. However,.

A

C

D

A mixer is quite slow, so.

C

D

We are if I turn off the filter, so actually we haven't done any large-scale testing. So we don't have this kind of height application.

E

B

D

So for a single node house, if we.

A

D

Have testers you single-note.

A

D

Performance so, but we maybe thought we were have an extra 10 to 13 extra cost. That's what we needed.

D

So actually it has a lot of things to do with the traffic as well.

D

I'm sure the cost will have an extra cost of CPU about 10 to 12 percent and there was no impact on the latency and.

E

D

It's just not only about the mixer, but also about the agency. So thank you very much.

C

B