Kong Kong Summit 2021, 17 Dec 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Apache Kafka, API Gateway and Service Mesh for Cloud Connectivity

Description

In this video, @ViktorGamov illustrates the differences between an API gateway and service mesh — and when to use one or the other pragmatically and objectively. He also discuss the similarities and differences between the communication layer provided by gateways, service mesh and Apache Kafka. Finally, you will learn a few ways to use Apache Kafka within a service mesh architecture.

Watch the full recording from Kong Summit 2021 here: https://konghq.com/videos/proxies-gateways-and-meshes

▬▬▬▬▬▬ TIMECODES ▬▬▬▬▬▬

0:08 From Proxy to Gateway
4:43 Road to Service Mesh
11:00 What About Apache Kafka?
15:55 Let's Talk Patterns
19:35 Conclusion

▬▬▬▬▬▬ LINKS ▬▬▬▬▬▬

-Learn more about Kong Gateway: https://konghq.com/kong/
-Learn more about Kuma: https://kuma.io/
-Follow along and grab the links in the blog version: https://konghq.com/blog/apache-kafka-api-gateway/
-Watch more recordings from Kong Summit 2021 here: https://konghq.com/kong-summit/2021-videos

#Kafka #APIgateway #ServiceMesh #KongGateway #KumaMesh #KongSummit

A

I will teach you some of those patterns and some of those communication, things uh from um very inspirational, sci-fi tv show right. It's a star trek and we're gonna be talking about the all the things I will try to explain all the things through the lens of star trek, john lupicar will be representing our service. We're writing some of the application where we need to bring this application to production so think about this. This application boldly go in where uh no one men have been before. So.

A

In this case, um our micro server is going into production and we do have a couple things apart from the solving the business task. Apart from the solving the problem that the business um wanted us to solve, um this can be like order um service. It's gonna be billing service. It's gonna be shipping service, something that we need to solve uh like as a as a developers.

A

We need to implement this, and, um apart from that, many developers constantly forgetting about the data responsibility, um it's good that we're trying to educate many people these days in terms of uh ways of devops and how developers need to appear to communicate with operations.

A

People in terms of understand the responsibilities, what kind of stuff needs to be implemented in this application, and we started talking about things authorization notification, so I want things to prevent, uh like uh denial of service attacks, some of these things to prevent um work against the legacy apis, some of the apis that you implemented your system will be obsolete soon. You need to update things and how you will support all these clients in the future.

A

You need to think about this, how you can do proper monitoring, end-to-end tracing some of the um logging functionality, maybe even caching, of some of the data uh where you know you need to do this and how you can do this so um and here we're gonna. I will try to bring this metaphor and see how far I can go with this metaphor.

A

Now, there's a neat part you don't have to. If you don't know how to do this, you really don't have to, because there is a set of program, components that allow you to implement this without any problem. So, essentially, how I see how I see the gateway api gateway as a system that provides you with some additional implementation. If you remember in the star trek um the the borg collective, they were pursuing perfection in terms of how these species will communicate.

A

So in this case, if you will have some of the things that or you don't have some of the things, those things will be provided to you at no cost. If you want to have a fancy augmentation for your face plate, you will get this. If you want to have this like a fancy augmentation for your arm, you will get this from for free, for the borg. Collective will provide this to you. So I think that you, you think about this.

A

When you become in a drone in the board collective you getting all these super powers of the gateway uh for free, so essentially with your micro services, um with the with the additional mutation becoming a like very super, powerful micro services right, so you will be with the clone gateway. You can enable all this functionality on the side of the gateway, so your external clients will talk to gateway. They will be talking to you.

A

They will be talking to. Let me show you: where are we standing with this?

A

So your customers, your play, api, calls, would be engaging with your gateway and the gateway will translate specific uh specific call to your service that we'll talk in grpc rest. Maybe even some database calls. uh We will talk a little bit about some of the event driven patterns with uh with the thing called kafka. You've probably heard about this and we're going to talk about this more today.

A

So this is where your gateway brings more superpowers to your um to your system, and you never need to implement those there's a hundreds of plugins available that can implement notification, authorization, integration, third parties etc. So that's it.

A

Next thing is that we're going to talk what what your services uh we'll be doing in future, so I started this conversation with small module, we're starting this with the small small module that we will be um augmenting with some of the functionality that provided by by gateway. But you have multiple services. You need to communicate through the services, and the next thing is that we're going to um provide the custom or common ways how the microservices will be communicating. This talk is all about microservices, communicating and the ways how they communicate.

A

You can actually put the gateway on top of your microservice and go talk to other microservices that has a gateway in front of this.

A

It's you know some of the things that we're doing years- maybe not so often right now, but you might be hurt about some of the things when you go into production, some people, operation, people say: oh, we need to put load balancer and fraud, that's essentially load balancer and uh very uh very uh crippled version of the that thinks api gateway can do apart from the doing load balance now we're talking to how we can.

A

We have like thousands of services now um and we start thinking about the things not from perspective of pets, but from perspective of cattle, so we want to deploy microservices every day. We don't want to um them to be like a super difficult to communicate. We don't need to know their address. We just know how we can discover those, and this is how we're going into mesh and from perspective of work. That's called the simulation. You become a bore, collective and you will be connected to the hive mind.

A

So this is how your services start communicating. Once you become a part of the board collective, your services will join the big work collective and there would be easy to communicate between those. um You always have a connection.

A

You will get unified set of tools, how you would communicate and you will communicate through this with um with other components and this our microservices will communicate to each other through this, um the bigger mesh that allow them to uh to communicate, and this is how it would look like your application. Will um all these components will just like it will be plug and play um some of these meshes actually. Can uh can be communicated together, and this is what the kuma can do.

A

Kuma is open source, the service mesh implementation that supports um deployments across multiple different um platforms, including hydrogenous platforms, so you can deploy um your one mesh in one data center, another mesh, another data center third mesh in another- um even maybe I don't know you have the virtual machines and uh the only thing that nash will have kind of uh slightly the way, the way how the mesh will be getting its signals or getting this configuration instead of thing called control plane.

A

So each work, uh collective member will behave as a data plane, so data plane will receive some of the comments from um uh from uh control plane. So in this case it's gonna be a borg queen sent all these requests to this, and if there's communication would be locked between uh between the queen and between the particular drone. um For those of you who are not the sci-fi gig drawn is the individual member of the war collective.

A

So with with this, um your um your drone still continue to operating and it will be working based on the uh some of the commands that were sent before the network was destroyed.

A

So this is how it looks like from perspective of you know: let's, let's get back, let's bring into uh to the normal diagrams right, so we're bringing this into um into world of the architectures that, where the diagrams architects law of the draw diagrams so the control plane, this is the brain of operation and the control plane can operate over multiple zones or multiple bore cubes.

A

We have a q1 and we have a cube2 and communication. Communications happened in the real world by a small thing called um proxy. So remember when we were talking about the gateway gateway proxy, you can use some of those terms as people using this as a synonyms. In this particular case. This small instance of this proxy provides a similar set of functionalities that I outlined previously when I was talking about gateway and through this, this is communication is actually happening. So your service, as a you know, new member of collective.

A

You don't need to know anything about the board collective before, so everything would be provided. As I said, uh you will get the fancy augmentations. In this case, you will enable observability in your service. You will enable authentication, authorization, um zero trust, security, some advanced policies. In order to um doing some some some communication, uh you can have a sub meshes. You can establish communication with an individual like cohorts. You can create the small armies, like small uh squads, of your uh drones, and this is everything can happen through control plane.

A

So as a as a as a borg queen, we can control this one. So that's how how it looks like from perspective of this application- and this is where we go into space and go into different quadrants with the single mesh. We can create one big mesh that is going to be gcp.

A

It's going to be aws, it's going to be azure and your application will communicate to each other. The drones from one cube can communicate to drone to another cube without directly communicating to this. They will be communicating through proxy through this medium, that your board collective, provides all right.

A

And this is how it looks like from perspective of the real world where you can have deployments in the different places, and you will see this in the ui. Now we learn a little bit about a um gateway in proxy. We will talk about mesh and how the things are are communicating I hopefully, hopefully you will uh learning something from this.

A

um Do not forget to throw some of the questions right now, because we will have some time to talk about some of those patterns now, all right so victor, but uh like it's all cool cool and nice. It's everything is great. However, like where's, the kafka comes into play.

A

It turns out- and I spent significant time in my life talking about this- it turns out that uh sometimes this uh um unified approach to communicating services um or like a too too simplified, because your services rarely uh operate or communicate to each other without having a context without having some of the knowledge about what kind of data is going there, and it turns out that many people use um kafka as a bus to communicate those services and those services can communicate in.

A

I can say in the data aware mode right because they communicating through the certain pattern that we'll talk in a few few minutes um and uh obviously, when I'm talking about data, I cannot. I cannot miss this opportunity to provide this fun pawn to data aware mesh. So data was also simulated at some point during the uh the first contact uh movie. So that's, that's you have it like some people who are expecting to have a data reference in the in in a work talk.

A

You have it now, okay, so um my friend and my former colleague gwen shapiro, uh if you follow folks doing anything with kafka and you're, not following where shapiro are you doing things wrong? When shapiro awesome speaker awesome author of this um definitive guide kafka for definitive guide? um Second version is coming soon, so she did the talk where she also talking about the how service mesh would be comparing to kafka and the one of the things when they started talking.

A

Many people started using microservices as a um in the communicated through api and very soon they find out these different microservices to tangle to each other, and one of the approaches is to entangle this communication by providing this unified bus specifically like. Let me show you the example of this log and one of the things what we can see here in this log. We see there's some of the events happening and uh based on those events based on the things that we have in uh in our log.

A

We can restore history of of of the system, so in this particular case each event will have a time. So we know when it happened this event and this event um will be stored somewhere, usually it's event store, um and that also will be.

A

um You know all these events will be ordered by time and the things happen here, so there's client doing something, string, client doing something consumer doing something, and this is how the time is growing. It's the time right time, so the same concept of log that you're using in your application development, was taken as a data structure and implemented as a distributed lock inside the kafka. Kafka's event store that captures. All events happened in the system, so one um one service will create some event and post this into kafka.

A

So other systems who will be interested in this event will be responsible to consume it and the way how it looks like yeah. Some some people are saying victor. Are you talking about like a the block for g now? Are we talking about log as the data structure, and this is how uh you can entangle by providing kind of like a um direct communication with your consumers of your service? So, for example, service one publishes into topic one, um and uh there will be multiple consumers who will be interested in this.

A

So instead of this service going here and getting data from service one, they are not. They are communicating through the topic and the good thing they they become decoupled and they don't need to know about existence of each other um through uh through this topic. So in this case, this is a totally disconnected communication, same thing here, like service two published here and the service three and four uh service, three three or three, four, it's called five and service five will consume this data from other sources.

A

It can be consuming data from other services, for example, these five also consumer services, from three and four. So in this case you might argue in terms of like yeah, we don't really fix the detangling, but I think that's um also allows the persistent, uh persistent messaging uh of of your of your messages um and also to bring this idea back. uh The kraft think about kafka is kind of a delayed queue for your emails and again a reference to charlie from it's only scientist philadelphia.

A

You have a this delayed um communication between the services services can be down, while uh one of the things will go up. So let's bring it back to our things. This is basically three most important slides of this talk. Hopefully you get a little bit understanding from perspective like how one thing will be connected to another thing. So, let's talk about three patterns like what you will take from this talk. So first pattern like if you're doing everything with kafka forget about this gateway forget about measures.

A

This is something that, if you kafka guy this is you will do like all day long. You will be using this pattern called um uh service uh event coordination. So essentially you have a message. Bus that will storing these events and different services will be providing the data to the different topics. This is what you saw in the previous slides and, uh and that's it they will be disconnected. They will do anything together, so everything would be nice and fine and these services can be down.

A

This can be happened in the in a different time, because this service, um this event store, will store this data forever for them, um and this is how you will be doing this. uh The microservice communication, if you don't have anything else, but um we're coming into the world of uh where you need to have a like overall control and your you cannot enforce some of the uh communication between the system.

A

So if uh you still have a situation where you are um your system talking to each other, you kind of doing this, something like this. You are using a gateway, api gateway um and uh like a kong as external as an external, um uh the conduit proxy, something that and allows things to enter your system after that, inside your application ecosystem using mesh to communicate applications, and also um you can run this like a kafka where systems you in in a mesh. uh There are the filters that allow this proxy communicate.

A

They understand kafka protocol, they can rely on also service discovery and these type of things. So your cafe, your data, awareness of the services also will be running inside the mesh, and this is kind of like this is what you will be doing if you implementing these services today, for example, for me, um this, this type of thing would be a no-brainer back in my uh kafka days. So that's what I will be doing all day long without using third-party tools.

A

However, it is something where you know you don't go in for your adventure without the friends. So that's why you bring in friends like your service mesh like a gateway to you, know, to have observable and reliable connectivity, and the third thing that's also people um kind of forgetting that this is also very valid pattern. uh This pattern, known as event gateway, so the kafka communication uh also can be um kind of like this side thing to your to your to your uh to your system and the kong, for example.

A

With this new announcement of the kafka plugin can produce to the kafka topic and the systems who are maybe not developed even yet so, but in future they will be interested in this data that was produced through this event gateway.

A

They will be able to read this data from the beginning of time from from event store, and we will be able to processing things so the links there. um You can find and read a little bit more about this um also. I would like to learn a little bit more um if you're interested in this type of subject. So maybe I will go and do a deeper dive into this topic during our uh kong builders live streams that we do in the kong youtube channel subscribe if you're watching uh subscribe to our youtube channel now.

A

The last thing something that you need to remember gateway is for admitting gateway is the is your? Is your big nice fancy door to your to your application functionality? So it brings traffic to you from your clients to your application. So mesh connects your application together, so it is provided the mesh provides communication, a reliable communication, some of the uh traffic routings like circuit, breaking pattern inside between your microservices and kafka. Sorry, for this one kafka streams, cuff streams.

A

Your data uh kafka is good for asynchronous communication for use cases where you don't know who will be consuming the data and the same thing with this. Like you don't know, you don't need to expose some of the um some of the your api. You will just publish this data, publish this into kafka.