GitHub GitHub Satellite India 2021 - DevOps: Day 1, 28 Mar 2021

Previous Meeting

Next Meeting

⏯

youtube image

►

From YouTube: GitHub Satellite India 2021 - Building event driven apps on Kubernetes with Fission

Description

Presented by Vishal Biyani, CTO @ InfraCloud Technologies

FaaS abstracts us from the underlying mechanism and makes it easy & fast to build and deploy applications. The open source project Fission enables FaaS on Kubernetes and makes it easy to build and deploy event-driven applications on Kubernetes - without having to understand Kubernetes concepts in depth. In this session, Vishal Biyani, a maintainer of the Fission project walks us through the uses of FaaS, how Fission enables FaaS on Kubernetes and the journey of contributing to open source.

As always, feel free to leave us a comment below and don't forget to subscribe: http://bit.ly/subgithub

Thanks!

Connect with us.
Facebook: http://fb.com/github
Twitter: http://twitter.com/github
LinkedIn: http://linkedin.com/company/github

About GitHub
GitHub is the best place to share code with friends, co-workers, classmates, and complete strangers. Millions of people use GitHub to build amazing things together. For more info, go to http://github.com

A

uh Remember audience: please engage with him on discussions.

B

Hi everyone, uh my name- is vishal biani, I'm cto and forward infected technologies, I'm also a fashion maintainer uh and I'm pretty active organizer of kubernetes pooling so we'd love to connect you there cool. So today we are going to talk about building event different applications on kubernetes with fission.

B

Now before when we go there, I think the first question that comes to your mind may be why care about function as a service platform on top of kubernetes. When we have cloud provider, you know managed uh function platforms.

B

I think there are some use cases and there are some scenarios where I think the fast and kubernetes really shines and we'll walk through some of those scenarios. So today, if you look at developers, they develop functions, they develop micro services, they develop source code, they also build docker images.

B

uh Will it not be nice to have a consistent way to deploy and not worry about the underlying runtime or not worry about what platform I'm deploying to all of these artifacts?

B

So today I think the developer's choice, pretty much is clear. We want to deploy all these things onto kubernetes in some form or fashion, and this kubernetes could be sitting on top of other cloud. One cloud, or you know on you know on premise bare metal machines, but as a developer, I don't want to worry about what is the underlying platform? I want to worry about the code. I want to worry about how much value I can deliver.

B

You know with my applications, so one thing is consistency of being able to deploy different types of workloads right from functions to microservices and then not compromising on the underlying things like scale or operability, or you know being able to scale as the demand grows and then being able to scale down when the demand kind of vanishes.

B

The second factor, I believe, when we develop all these different applications. You know whether it is services or functions, or you know, using docker images or so forth. I want to integrate with them. You know with a bunch of systems, it could be a message queue system, it could be a database, uh it could be a monitoring system and I want to write all these integrations as a developer without having to worry about. How do I connect to this? How do I make it work? That is one part.

B

The second part is, I don't want my workers to sit idle when there are no messages or no events happening in my message. Queue. For example, I want the workers to be spinned up only when there is some event consume. Those events, once the consume thing is done, processing is done, shut down the number of processors back to zero, maybe right so I think something like that would be really ideal. For you know developers apart from these two key factors, uh I think a lot of times. You also want the control of the underlying infrastructure.

B

You know what is the operating system? What are the patches running? Are there any security issues and stuff like that?

B

You also want a fine grained scaling control, for example, being able to scale out or scaling as required, and you don't want any time out or you know any other limits on to how do you run these functions or microservices? You know whether on cloud on-premise with a similar experience- and I think one key point you might have noticed in all of these is a developer productivity.

B

How do we make sure as a developer, we are productive? We are able to deploy things, build things fast without having to worry about the underlying infrastructure and just scale out. You know on on demand.

B

Now we talked about why fast and kubernetes matters, but I think there are some scenarios where you shouldn't probably think about fast and kubernetes, and let me call them out very clearly so, for example, if you're doing only a few hundred invocations, you know of a function per day, it doesn't make sense for you to maintain the kubernetes cluster, the rest of the machinery to make that happen. On top of that, so anything at a very small scale. I would really not use fission or any fast platform.

B

You know on top of kubernetes, I would simply go with cloud provider, uh provided you know, managed like function services. uh Also, there are a lot of companies which start small, let's say a couple of engineers together, you start with a single cloud. You don't really worry about multi-cloud and those kind of things right. That is not a point at which you should be thinking about a fast platform on top of kubernetes.

B

I would really use as many as possible cloud managed services and scale my business, you know early on and not think about all these factors very early in my business also workloads which are very static or workloads which are not driven driven at all.

B

I think fast on communities is not a great candidate and- and I would rather use managed services from cloud providers great now, having looked at a fast and kubernetes, why it matters and why you should worry about it, and you know why, when you should not really use it, let's talk about fission, so fishing is a fast platform on top of kubernetes. uh It's a simple and you know easy to use platform for developers.

B

uh You can write code or you can provide docker images either ways the concepts of docker and kubernetes can be abstracted from the developers. You can expose them if you want, but you know, as a general rule, if you don't want to worry about docker or kubernetes or building docker files, all that is abstracted out from you.

B

We maintain something called as a warm pool. It adds a latency of 100 milliseconds, but it gives you the benefits of you know not being able to or not consuming the entire infrastructure when there is no load on infrastructure, so to speak, and there are a bunch of other things right from. You know how it is integrating with even sources and bunch of uh things like that. We will look at in a couple of slides.

B

So, while we are here here is a framework which makes developers job super easy, I would love if you can go to the github vision and start the project and definitely follow us.

B

Cool, so let's understand what fission can do for you as a developer. How does it work? What all it supports? You know what you can do with fission, so the first thing is: friction fits on top of kubernetes, so kubernetes is the platform you absolutely need for running fashion. We also very heavily use a project called cada, it's a kubernetes event, driven auto scaling. That is a long form and it allows you to integrate with a bunch of you know, sources, message, queues, databases and whatnot.

B

Now, as a developer on fusion, you can write functions. You can also write micro services, you can provide source code or you can provide docker images, irrespective of what you provide the way you build, deploy and then execute and run. That is consistent from your point of view.

B

Now. It also supports a bunch of languages right from java to golang to python rubypull.net. So you can write in any language that you want and the experience is more or less consistent across all these language platforms.

B

Now, once you've written a function or a microservice in any of these languages, you ideally would want to call them right, so you can call them using http or you can call them as in cron job kind of parameters, so you can say every hour or every minute you know this function or this microservice gets called, but you can also call them using bunch of event driven. uh You know sources, so you can use any of you know: amazon, kinesis or nats or azure event hub or kafka or redis anytime.

B

Something happens in any of these data systems. You can say a function should get triggered or invoked basically, but I think one of the most different factors is when you're operating a complex platform like fiction and kubernetes.

B

You need absolute details as to what is happening if something goes wrong and that's where we really, you know, integrate well with all the underlying observability systems, whether it's from ethios or elastic or jager, for distributed tracing or graphing for visualization, and you know, alerting basically cool now having looked at, you know why fast matters uh having talked about what fiction does, let's dive into the code- and you know let's, let's get some stuff running, so I'm gonna do quick, two demos, uh the first one is a simple hello world, so you get a sense of how fashion works, what it does, and you know what a hello world looks like.

B

So, let's, let's go right. There.

B

So, first of all, I'm going to create an environment which is nothing but the underlying runtime for any function.

B

Secondly, I'm going to use a simple hello.js piece of code and create a function now. What does this hello.js? Do? Let's, let's take a quick look, so hello.js is nothing but a simple nodejs function. It is returning a hello world and a status code of 200..

B

It follows a specific syntax when it comes to function, but you know it's relatively simple in terms of the logic good. So now we have the function. We have the amount running, let's test it out, so I'm just gonna do efficient function, test and hello.js and, let's see how it works. Great here is our hello world. Now, of course, hello world is great, but you know we need something more concrete, something more realistic, so to speak right, so I'm gonna do another small demo.

B

Let's say I have a producer function, uh it is producing messages. This could be anything. This could be your iot device. This could be your. You know. uh Pos applications which are generating events and those events are getting dropped into a kafka topic.

B

Now, as soon as there is a message in the kafka topic, I want another function to be invoked called consumer. This consumer function is gonna, invoke messages or consume messages from your topic and process them. If it is able to successfully process them, it will put the response back into another kafka queue called response, but for any reason it is not able to process. There is some error. There is some exception. It is going to put that into another queue called error. So you can, you know separately process them later now.

B

One good thing you know I want to call it here again is when you're listening to kafka messages coming into kafka topic. I don't want x number of workers to be always running and listening to kafka. What I really want is there is no power running or no application running. Only when messages arrive, those parts are horizontally, scaled out and actually delivering messages. So, let's go, and you know see how this is going to work.

B

Now, if you look at the bottom half of my screen, there are two functions I have one is the consumer function, which is what I was talking about, the consumer side of story and then there's a producer function which is actually going to produce these messages.

B

What connects both of these and the kafka topic is something called as a mqtrigger a message q trigger. So the name of the message to trigger is kafka test. It is going to invoke a function called consumer and the type of message queue is kafka.

B

It is going to listen on the request topic for any messages arriving and it is going to send all the messages which are processed successfully to response topic and any errors that happen they are going to get sent to error topic.

B

It's also gonna try three times before you know it decides to put anything in the error, queue and- and you can look at also. Lastly, the content, type of application or or the messages basically now before we go- and you know actually execute this, and I show you the you know: processing of functions. Let me actually go and show you the code of you know all this so productive function now predictor function called simple protocol.

B

It uses a handler method, uh it's a go program by the way, uh and you know it has a certain uh argument that it needs all it is doing, is connecting to your kafka message queue and you know, processing those messages, uh and there are about thousand messages. We are gonna produce and put into a kafka queue, and once it is done it is, you know it is going to exit the function is going to exit and then it shop.

B

Basically now once the messages have arrived in kafka queue, it is going to get picked up by another. You know component, which is going to call the consumer.js function and contributing not a lot. It is just picking up the message that it received from the kafka queue and it is printing it out that hey. This is the response you know, so to speak. That's all there is in terms of functions.

B

Now. What I'm going to do is I'm going to invoke the function producer, so it's going to produce a thousand messages, but before I go there, I want to show you something important.

B

So, as you can see right now in the default in space, I don't have any parts once I start producing messages. What I'm expecting is there are parts which are created on the fly, which will actually start consuming messages from my kafka queue. So let's put them on the watch more in the second window. Now I'm going to call my producer function.

B

So this particular gonna produce now about thousand order messages and and we'll see, two things happening. One thing is that more parts will be treated here as the messages start arriving now, as you can see, one part has been created for consuming these messages.

B

And it is running in state and if I look at the logs of that part, I can see it is producing about or it is consuming. You know the 20th or 21st messages now right now it is only one part running and it is consuming the early set of messages, but as it realizes there are more messages than a single part can handle.

B

This is actually going to auto scale.

B

So right now we see only part one part and we'll see in a couple of you know a second or maybe a minute, a couple more parts getting spinned up and consuming all the messages that are getting. You know uh written in the by the producer in our in our case, all right. So that's that's a very quick and simple demo.

B

uh What we can do is we can we can let the auto scaling happen here and we can come back and watch it a little later and we'll continue for rest of the presentation and you know talk through other things right another example. You know that is again available uh under on the github repository of session. uh We actually use a six set of functions.

B

uh One of them is a kafka producer very similar to our producer that we just saw right now that produces messages put them in a kafka cube and then there are about you know six other functions. One of them is the consumer. The consumer basically picks up messages from kafka. That's the messaging of the data and put them into redis. The other four functions really read: redis data and you know, provide it as an api endpoint and the last function is the web function. It basically queries the api in points and show you all.

B

You know nice statistics. Now, I'm not going to run this it'll probably take another hour for us to try it out and show entire code and working, but you can go and check it out on this repository and there is also a screenshot and a demo of you know. This working in in you know, live basically cool, let's go and see if there is more messages or more parts getting you know.

A

B

Right, I think it is still consuming about 235 order messages uh if it is if it finds more than one consumer growth, it will probably scale out, uh but if it doesn't, it may not scale out- uh and I don't know which way how configured right now, but but you get the gist of this once these thousand messages are consumed. The part that is running calf cut is, is gonna get shut down, and you know it is gonna get back to zero state.

B

So that is a truly elastic infrastructure that you can scale out and scale in as per your requirements and and still able to you know, build event driven applications which react when there is an event, and we don't react when there is no event so to speak great now. One thing I would like to talk about is contributing uh not just to vision to any open source project, and I want to bring a personal anecdote. You know from my journey of open source, so I really started my career into the enterprise software.

B

uh I was a mechanical engineer by education. So most of what I did was you know, learning on the job. In my early years of you know, software engineering career, I worked with enterprise companies, not a lot of open source. uh Of course I did use open source like java and other things, but most of the things that I did or that I work on were closed source and you know driven by commercial companies.

B

I think around 2010 11, I discovered a project called cloud foundry. It was a past platform uh very promising, and you know very interesting, probably ahead of its time, and I discovered the value of you know open source software. I could download the entire stack. I could run it on my machine. Try it out fix the bug, fix the documentation, send something back upstream. Why join github, you know, as any developer would do back in the day.

B

um I think the early years of my github and general open source activity was still you know watching from the sidelines. You know see what is happening, maybe look at a few projects. Maybe there is a few issues. If you try out something, maybe it is a small documentation? Patch, maybe fix one word here, one word there, but I was really happy. You know I could contribute in my own small ways in my own tiny ways to open source over time. You know I took small baby steps, uh so I created a solid stack workshop.

B

Basically, I learned salt stack as part of my work. Salt stack is again, you know another open source configuration management tool and I realized there is not a single demo out there where you can spin up. You know three boxes of vagrant on your own machine and play around. You know with salt. uh The way you would do with a very realistic kind of you know application or demo. So I built a small utility. uh You know put that out in the open source.

B

I talked about it a couple of my colleagues, my friends, and they were really happy that you know they could actually clone it up it and you know they could. They could actually start playing with salt. So that's a great great satisfaction. uh You know from giving back to community and you know making somebody's learning easy and around 2017 I started with fishing. I didn't know golang, you know, for example, when I started with christian, I learned it on the job I started continuing to solve features small documentation, patches, uh small bugs over time.

B

You know I started converting, bigger and bigger features, and eventually you know I'm maintaining the fission project as of today. So I want to say you know, don't don't think you know that you're to start big, don't think that you'll do everything you know right from the beginning, very big start small starts helping in smaller ways. You can. Eventually you will reach a point. You know where you'll start contributing in bigger and bigger ways.

B

Now, after expression is concerned, you can go to fashion, you know dodge.fashion.ios docs, contributing and you know build code on your machine and see how it is working. If you run into any shoes, please join slack. Ask questions to you, know contributors other community members, they're very, you know happy and you know friendly to help. Others uh also there's another project. You know that I'm kind of practically quite involved into is it's called what cube it's a chatbot for kubernetes works.

B

On top of you know, microsoft teams, matter modes, slack bunch of chat platforms and again you know very nice project to start contributing, make small changes understand. You know, infrastructure, understand systems and so on and so forth cool. uh Let's see and go you know if our parts have scaled out. Our messages are still consuming. Okay, so we are about 715.

B

I don't think it did scale out because there's probably only one consumer group, but if I had more than you know one two three four consumer groups, then it would have scaled like three four parts uh and consumed all the messages and you can see by probably in couple more minutes a thousand messages will be consumed and done and- and the part will be- uh you know, shut down the one that was consuming messages.

B

So great, uh that's a simple demo. uh If you have any questions, I'm happy to take them on the github discussions as well. uh But you know thanks for listening to me uh patiently.

A

A

You have brought in a lot of energy a lot of shine in our head of satellite event. Thank you. So much for that.

B

Thank you thanks.

C

C

Celebrating and a big shout out to all our community in north of india folks out there make some noise share with us on twitter, github satellite. Thank you. The session was really awesome. Let's jump to the questions now.

A

Yes, vishal record questions for you and I'll take the first one. How does one build a chain of even functions with fission.

B

Yeah, that's a great question. uh There is a project called fashion workflows which allows you to do that to some extent, uh but also our users. You know they have actually built out their own custom logic to build a chain of functions. They basically use message queue as a way of origin stating so one function produces something it you know puts into a message queue and there is some other function listening to it and that just builds a chain of functions. You know that you can execute.

C

Basically awesome um when it comes to lambda or azure functions like when are those better to use as compared to fission, for example,.

B

Absolutely I think, uh for anybody who is new uh in general to computing or let's say you know fast, I would highly recommend them using lambda or you know, azure functions they're a great starting point. They give you a lot of value without having to understand all of the underlying you know. Nuts and bolts of you know infrastructure.

B

So I think they are they're a great starting point uh and in fact I would say even for smaller businesses early stage businesses, they are great value. Add they give you so much power without having to worry about the infrastructure.

B

Only beyond a certain point, when you start scaling out, I would really think about fast and kubernetes as a strategy, or you know if it is going to give me something more than what I get from azure functions or lambda, for example,.

A

That's a great advice for how one should look at transitioning from one stack to another and we got another one and uh you shared a lot about the projects uh where you are contributing michelle, and I think that's like to this. So audience is asking here: how do I get involved in fishing and start contributing to it.

B

Sure sure uh I think the the first and foremost thing I would do you know, is go check out the project and try it out forget about contributing, but also understand what it does right, build a simple function. Deploy it see if it works right. If you again run into any issues, join the slack people are happy to help. You know and guide you once you understood what the project does, then I would go and look at the good first issues on fiction or any repository.

B

I think most of the qualities do follow the standard and find out a very small issue in the area that you are comfortable with, maybe with a message queue, maybe with a database, maybe something you know simple as linting code, for example, right and start contributing.

B

What that will do is that will give you an idea of the project structure, how to build how to deploy once my first, you know, bfx simple bug is done, then I would go, and you know, figure out more complex, bugs smaller features and get more and more involved into the project.

C

Thank you, that's a great advice, any special messaging for say our student community out there and early developers anything different. You would want to suggest them.

B

uh Yeah, I think I I do advise you know a few of my family and your friends were students, I think uh start basic start small start with a simple language, but definitely go to depth of it and have a lot of attention to detail. And today you know we are lucky in 2021. We have so many tools whether it is a browser-based ide or you know, github github actions.

B

We can do so many things without actually having to know, and you know, build everything on your own machine, uh so I would say pick up one language get great at it and beyond that, explore your interests. You know whether it is intro systems or ai or machine learning or big data. You know whatever it may be, and slowly really you know progress with open source projects, because there is so much out there.

A

Thanks thanks for sharing that- and I wish all of this was available for me as well as I was growing up so vishal- uh you mentioned something very interesting about the pune communities and cncf meetup would love to know more on how it started and where it is today. What is your future plan.

B

Absolutely absolutely yeah, that's a topic close to my heart. uh I think when kubernetes was very early and still the dominant orchestrator to speak, I thought missiles will take over, but uh I realized kubernetes a lot of power. I started the meetup because uh there were a few people that we knew about kubernetes and we thought it's a good idea to share with broader community, and hopefully more people will join us.

B

So when the first meetup we did, it was just eight people, including myself, and over time you know it has grown to now almost 2000 people actively in the in the meetup group, but also 50 60 joining us every month. uh We talk about you, know all the basic topics and some advanced topics as well and and uh learn from each other.

B

Basically I mean that's the goal, and- and that was the original idea that you know what I know if I can share- and you know others share from their learnings- the mutual learnings will be met together more than you know what we can do individually.

C

uh I'm sure folks who are listening in would want to also know how can they join in.

B

Absolutely so here's how to go to meetup.com, kubernetes, pune and and if you're, already a member of meetup, you just join the group. If you're not just sign up for meetup.com and join, you know kubernetes and poolyonmeetup.com.

A

Great great, that's that's great, and uh this question has been coming up for almost every speaker out here, so there it is for you. What's your favorite set of features at github and what do you use the most.

B

Yeah, absolutely absolutely, if you ask me what is my most favorite one today, uh it definitely has to be. You know the vs code on github, where you can open a browser. You know ide and start developing.

B

I think the other one that I really like is the github actions, because it gives you a truly integrated, workflow right from source code management to writing code to ci cd, all in a browser window. So so I think the vs code in browser is my most favorite feature. I must say.

C

Wow, so it's actually coming really close most like code spaces and actions and that's what uh we've been hearing from folks as well.

B

C

When it comes to other projects that you contribute to, is this usually in the same area or do you do you prefer like again from an advice to others as well? Would you prefer to deepen in one area or explore multiple.

B

Sure uh I think in past I tried to do multiple, uh but I think at least personally from my perspective, what happens is then you spread yourself too thin into too many. You know areas and you don't get depth. I prefer to focus on one area. For example, even if you look at just you know, infrastructure, there are probably hundreds of projects out there and I don't think you can again contribute to all hundreds of them anyways.

B

So I would really focus on one area uh get depth into that and then choose a few projects that you can contribute to. You know consistently. I believe the consistency over time, you know, is a lot more important than contributing to x number of projects uh if you're early in your career or your college student, who still don't know what you want to do, whether it's system, design or big data or machine learning, it's okay, to explore it's okay to be a little open.

B

But at some point you have to take a call and say: hey, I'm gonna focus on this one area and consistently spend my time over. Let's say you know next couple of years.

A

That's great that's great advice and definitely very very useful for all our audience and vishal with that. We'll wrap up the question and answers here, but for our audience, you're gonna hang around in the discussions room with them for next 30 minutes, so everybody out there do catch him for all the amazing work that he's doing, learn from him. His experiential knowledge, especially about how to contribute and getting started with the open source projects with that bye vishal, and we will see you in the get up actions room leave you there with all our audience.