Cloud Native Computing Foundation Serverless Practitioners Summit 2019 (San Diego), 24 Nov 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Getting Beyond FaaS: The PLONK Stack for Kubernetes Developers - Alex Ellis, OpenFaaS LtdSDCC

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

Getting Beyond FaaS: The PLONK Stack for Kubernetes Developers - Alex Ellis, OpenFaaS LtdSDCC

The term “Serverless” was popularised by managed FaaS offerings in the cloud. We are now seeing strong offerings becoming available on Kubernetes such as Knative and OpenFaaS that also includes APIs, static sites, and microservices. That definition of “Serverless as FaaS” is now causing confusion for some developers who feel that they may need to learn Kubernetes inside out before they can move onto something more advanced.

The opposite is true. Serverless when applied to Kubernetes brings a PaaS/FaaS style experience with easy packaging, scaling, metrics, eventing built-in using familiar CNCF projects.

The PLONK stack includes Prometheus, Linkerd, OpenFaaS, NATS and Kubernetes for a complete event-driven platform that is scalable and trusted by dozens of companies in production. Attendees will come away with an understanding of how Serverless on Kubernetes is really about developer experience and it can be easier than maintaining a mountain
of YAML files.

A

I want to talk to you about a slightly different perspective. We've heard a lot about architecture and components of different frameworks. I want to see how could we actually use these to build something, and what do we need to be able to do that? Well, I think the word fires and serverless can confuse some people, and you might have seen this diagram before this picture. I guess today, most of us are probably on the right-hand side, but there may be people here they're on the left-hand side of this diagram and I.

A

Don't know if you've ever had this conversation with somebody where you explain about a project that is fires and they say we're not ready for service. Yet so what you're building we have containers and kubernetes and micro services, and so I think there's some kind of disambiguation that we need to go through to show you that there's more to server less than fat right.

A

So we'll start with that, and then I'll show you the plank stack. So these are cloud native projects we'll look at what each one is and how it can be used. Some case studies of what people are doing with it today and then there's a few ways that you can respond, so it's always I always find it helpful to kind of go back to this example, particularly because it's quite simple, we have a huge monolithic application and a database and a client.

A

The reason why we've probably moved away from this is because the change domain here is potentially unbounded every time we want to make a change to application. We could have a knock-on effect, anything else, that's within it and when I worked to ADP producing payroll applications, we could take up to six months to Kerr entire release or a patch might take two weeks to test the river problems that docker helped to solve. Where we were able to package our code in containers and know, do you have the same code in every location and so I?

A

Think? If we look forward from here, we end up splitting things down and those responsibilities become smaller. We don't have to worry about the change domain. Quite so much so in the pet store. We now have a purchase function and a search function that might even be a micro service. The authentication doesn't have to be something that we've written.

A

It can be something that we buy in or a service that we make use of like AWS Neto and we can even have an API gate waivers before we might have had to have relied on iis or building that ourselves. Now, if you squint at this cube, Connor I could show you this and you might see that's in Cuba Nettie's architecture and if I showed you this, maybe at AWS re-invent, you say: that's a lambda architecture and that's what I want to try and get across.

A

Is that they're so similar that I think we can do very similar things with them.

A

So keep cuban at ease is probably undoubtedly where we want to be running this kind of workload because we're at Keep Calm, but the popularity of kubernetes of developers has really peaked. I think we've got around 10,000 to 11,000 people at this conference alone, but that's not the only interest actually there's top-down interest as well. So you imagine us here, maybe the developers and PMS the bottom-up interest.

A

We also have the c-suite that's heard about it and for whatever reason, whether it's from analyst relations or maybe they even quite technical- they used to code they're like okay, adopt kubernetes, let's be cloud native, it will fix all our problems and it turns out that it's not always that simple, there's some advantages you can get from listening, shifting lifting and shifting. So let's look at it from both perspectives from the developers side. We've got this declarative, API, really extensible, and it's so extensible that the communities extended beyond what people Ness is even designed.

A

I was listening to Tim Hopkins yesterday telling us about ingress route, and these plans the VMware and hectar, have had to kind of rethink about and make it more multi-tenant it's built to scale and there's a very rich ecosystem. You see a picture of the cloud native landscape and it can be a bit off-putting, but those are all projects that we can actually use to build our stuff now, rather than having to do it all ourselves. That's before now, businesses can, if it from the open governance.

A

So this project was originally created at Google as we know, but now it's a place where people can be held to account, and that is very important for some companies it's portable, so we can run that in different environments.

A

Gartner recently changed their multi cloud advice from deploying to multiple clouds at the same time to using docker containers because they give you a level of portability and because it's used in production. That means there's companies that are ready to stand by you and support you. So there's a hundred is fees at the moment, it's probably more as of this week, and you can pay them to help. You support your software in production, okay, so great reasons on both sides, but I think it's still a little bit lacking in the middle.

A

So it's a great for a platform engineers great for the business but the application developers. Perhaps the background that I had at ADP was creating applications on a daily basis. They probably don't want to be getting into the low-level details.

A

Don't know if anybody's ever used. The lamp stack. Do you remember what it stands for: yeah, Linux, Apache, my sequel, PHP yep, and when I first used that it was great because I think there was an installer. There was one thing that I put on my computer and I was cutting code very very quickly and I could take that to a VPS and effectively running in production.

A

We've got some slightly more modern stacks now lamp and a sari mean and jam, but I want to look at a stack that can allow you to write in any programming language a bit like what Eduardo was showing us. So this plunk is an acronym. It includes Prometheus so that we know what's going on with our applications. We can observe them, and these are all cloud and cloud native projects apart from open passes, are not part of the foundation, but the others are linked.

A

Adid is great for when you're working with micro services and the certain properties or behaviors that you need from the network again SEO is an alternative that you may use there, but if you're not ready for that, we can still. We don't have to be pedantic. We could just swap that for Linux. So that's why that's up there open files gives us a platform at restful, api c RDS on kubernetes, where we can go in deploy micro services and functions very quickly.

A

Nats is the current default asynchronous processing bus, and that means that you don't have to design your own queue system. Nobody should have to do that in 2019 and then Kuban active. We all know why. That's there so I'll give you a real example of an application that we wrote. I did this with a contributor in Japan and it took us about a day asynchronously and it records the activity of a github repository of your organization. What we're looking at is the issues that have been opened and comments that have been made.

A

You may say that you could measure a github activity through commits and pull requests, but this is trying to look at it from a slightly softer angle, something that shows a level of engagement, so the architecture for this, if you think to the pet store that I showed you and how we broke everything out, it's very similar again, I put the far right. We've got the user who would come and try to get to the dashboard and simply was serving up a static, HTML page he's using VJ s.

A

That will make your request back to a function called leaderboard, which is, of course us execute social procedure, and we've got a relational database that we're not having to host someone else is hosting it for us, the bit that I, like is the event based part at the bottom, so the contributor, when they make the action and get up, has a pub sub relationship and it will send to a micro service, the payload.

A

We check that it's actually from github and then we do the insert, and so you it's an obviously not actually real time, but it hasn't been very close to that. We can be tracking the activity and we can build a system out like this. Okay.

A

So that's a real example of the plank stack look at each project. Prometheus is something that you can use, whether you're interested in service or not. You probably will you have a good idea about what it does. It's a time series database and we want it, so it can observe our system and be able to ask questions about how it's doing it's very important. So you've got these containers spinning up and down and many services as an example of prom QR, which is the language you can write in.

A

This is also available in other time series databases like in flux, DB. It can be quite difficult to get your head round when you begin, but we're just looking at the rate of invitations over a minute that that went. Okay, okay, but the other thing that you can do because this is a time series is, you can ask a question in it in time and you can even look at where we are today or you can go back over I.

A

Think the retention period is one two three months and see how things were going, and so that might be useful if you're rolling out a new version of your code, or perhaps you just want to know how things are going now. I was chatting to somebody came up to me this morning from optin and he said: we've been using Open plasma production and we basically built these webhook receivers as functions, and so when we have various prometheus alerts that go out of band, so we may be getting too many errors.

A

We execute that code and we try and remediate it automatically, and this is an example of what you can do with alert manager. That's part of Prometheus now we're using it for something slightly similar in open fire, which is to scale the functions or services. You can also use HPA from cuba Nettie's if you prefer that, but this is just turned on by default.

A

Let's take it a step further being able to integrate the metrics from Prometheus directly into the API such that we can tell you how things are going and you see other projects do that link of the ships with its own version of Prometheus, for instance, and they use those metrics in their dashboard.

A

So if you're wanting to use this, whether you're interested in service or not, you add a metrics, endpoint or application, you then be able to get HTTP data very, very quickly. They've actually got code in an SDK for that we can start listening to your own kind of messages. So perhaps we wanted to see the github organization as a counter and see who is actually committing code or commenting and then can trigger alerts.

A

So that's what alert, manager and Prometheus can do fee, and this is a graduated project, which is the high status in a foundation.

A

If you ever, one that I wanted to mention was linker D and again, the guy from optimum came up to me this morning and said well, we using hiss do with open fire. Both both are fair game and what you might want to do with this is to shift traffic between versions. I think this already came up earlier today, and the basic idea is that you may not want to have a new version of code and have always traffic go to it once so.

A

Without changing the open files API, we were able to use this and just create simply two functions and then another one with a primary name, and we can change using SMI spec how much traffic is going to A or B, which is really useful. Mutual TLS isn't something that necessarily everybody is going to need up front, but originally in the 90s it was okay to have our security up to Perimeter up to the firewall.

A

If we want to start looking at zero trust networking, perhaps we're doing payment processing when you think about encrypting and traffic over the wire- and this again is something that you can turn on very easily.

A

Now the insights we had from Prometheus we're good and they're built-in, but we can get more than that. We can start to look at the TCP connection level and link it. Ii can help you debug that if you want to know a bit more about how the two projects can work together, there's a talk on Wednesday with Charles from buoyant I'll show you a link at the end, so open files, which is the O started in 2016 and I.

A

Think much like the story that we heard from Asia I had been using lambda and basically cloud functions and wanted away being able to run this on containers. That was my basic goal, and lots of things have changed since then, and the project at the moment, I think, is still the most popular by github stars on the landscape.

A

But the reason for that I think is down to the values of the project and the community, which are to try and put developers first so that it's very easy to create new functions to deploy them there'll be a CLI, that's intuitive command line and a UI to be friendly for operators so that, yes, perhaps you will want to use kathcart in production and that's going to be your use case, but maybe not everybody needs that. Maybe there's something we can do to get.

A

You started quickly or allow best practices to be embedded within the code. So there's a ham chart with lease privileges and many of the things that we can do so that when you try it out, it's a good experience and a positive one. The community is really really important. There's 250 contributors at the moment and 2,000 people on slack and they're.

A

Pretty active somebody joined the community recently and completed three huge features over the course of a week, and so that just shows how important the community is always doing things to try and involve new people now part of the reason that I created this originally was to try and move away from some of the constraints that I'd seen building applications in enterprise and then coming to service in the cloud and finding things were different.

A

So if I was an Amazon customer, then I'd be building a zip file out and trying to somehow compile my dependencies if they were native against the version of Linux. That Amazon were running in the cloud. If I moved to Google I had to completely change that process, I might have different regions available. Different maximum file sizes just didn't, really seem to fit very well with what I knew using containers and also it's very difficult to lift and shift a micro service into this pattern. I've seen the enterprise we can actually get something.

A

That's more portable.

A

Now I do I've been pulling this service, 2.0 and I. Think I've heard a few other people uses term as well. We think of that cloud service, the SAS purely hosted where you can have an argument whether it's really serve a list. If it's, if it's a paid product or not any code, means that you can have this new kind of white light rather than the zip file. That's specifically for Amazon or specifically for Google. We can have a new type of white load that we can run anything with it.

A

We can take it to any cloud because of kubernetes and then scale down to zero if we need to or scale up as necessary, and the only way we can do that is by inventing it something new, which is a runtime contract. So before the runtime contract was specific to each cloud, and now it becomes a docker image that is stateless and listens to HTTP traffic on a given port.

A

If you can provide a container like that, it turns out most code can do that, then you can host it in open fires, inque native and even on cloud on, and so what I'd like to see and what I'm hoping to see is that more of the projects that we've talked about, like open, whisk and cater will start to adopt this format and we'll be able to actually take this, and if we want to move away from that idea of fires, we can just run this on Cuba 90s anyway, but it's actually pretty hard for a developer.

A

So what would we have to do to get hello world?

A

Well, probably going to start with a index J s. If we're using Java Script, then we want to put a middlewares in there package.json. You always have to have a package.json. You then want your handler service definition for keeping at ease. This is getting away with it lightly. This is a deployment Yama file with some health checks in it, and then we've got our ingress and we can't forget about that docker file. Now, after all of that work, you can say hello.

A

World on port 8080 on cuban at ease, right and I think there's probably some things we can learn from what we've done before some people have compared serverless to cgi-bin right, but we just do a bit differently. Now.

A

The object-oriented programming has four pillars: those are encapsulation abstraction, inheritance and polyfill autism through the function store in open fires, were able to take a container that YouthBuild our for your pipeline, whether that's an API or machine learning model make that discoverable people can then take that and catch that you'd function run it locally and make use of it in their code. You can even have your own store at work.

A

Abstraction allows us to take and I'll show you what this will actually turn out, like all of those Yammer files and the docker files and entry point, and make a very simple template that someone can pull in or write their own I guess the closest thing we may be seen as a build pack before cloud native, build pack and then with inheritance, we're able to take perhaps a docker layers and make sure that we're only shipping. The last part of the code that we've changed in terms of polymorphism.

A

This is that service definition, where you have a an API, static website and micro service or a function, but because they all conform to that same definition. We can treat them like an interface and that allows us a portability. So this is what you would have instead right, so we had all of those files I think they count up to 300 lines of text, which is probably okay if you've got one, but if you have ten or hundred it's quite hard to maintain.

A

This is what you get instead and it's very similar to the Azure definition that we saw and there's a very small Yama file within that you can set environmental variables secrets that are predefined memory limits and labels, and that's really all there is to it. So we can go from zero to production in probably five minutes. We take this code and it finds our IP address is written in nodejs. We could just deploy that to Kuban at ease now, when you start to install these projects, even if they're relatively easy to do one.

A

It turns out that, because we don't want to reinvent the wheel of each project, we want to use what's out there. If open fires want to use something like nginx cert manager, open fast itself and then have a certificate. That was about five pages of instructions, even though it was super simple, so created a CLI called catch up and it allows you to just wrap home charts and pass in strongly type variables. So we could run this and five minutes. Probably exaggeration would have a production-ready system with a TLS certificate. I.

A

Do you want to try and show you the basic experience for bringing up her, bringing up a function.

A

So we've deployed a cluster earlier using those settings and what I'll do is I'll set my open phase URL. This is the gateway. I'll, then create a function and copy and paste this text in you can see here that I've done Farsi line new and put the language that I want. That brings it in from the template store and we have this code that will enter in and then there's simply one command.

A

Once we've installed our dependencies, which is up okay and remember, we're using containers, so we see container built- hopefully that's not surprising, and then the layered sister means that we're only shipping, the deltas so I built this earlier and we'll have that endpoint up and running and we'll be able to go and visit. It.

A

Once it gets pulled down from the internet, and so we've got our IP address and status taco right and if we wanted to iterate on that, it would be pretty easy to do so. So we've this ketchup tool. You can now automate the installation of this and perhaps we'll add something for SEO as well, and then you can just simply run the faz up like I did there and you have a production endpoint you can give to somebody now.

A

If you want to create your own, you would do first year Lai language, whatever node twelve goaline middleware. If you want to use docker file, that's fine as well. This will generate your code, scaffold it, but all of those files we saw earlier those big files.

A

They still are there, but you don't have to worry about them, because the template is maintaining it for you, and so the handler I eat a bit on the far side will be combined with the template at Build time and then we'll get a docket image that we can put in a registry. We can deploy that and that's where you'll have a URL.

A

Now, the other way that you can deploy stuff is through CRD, but you still need to build your function once you built it and pushed it somewhere. You have a pre-existing docker image that conforms to the service definition. We can simply go first Feli generate and that will give us a cid.

A

You could use flux, orga Topps, basically to be able to deploy that, and if you are creative developer and you want to try using the user experience of the templates you can just with generate and you can get a cane a teaser vein definition as well. I wanted to give you a quick case study. This is my friend Edie. Someone said that people aren't using serverless within financial processing, Ed's working at a payments processing company in the UK, and he actually likes this thing called faster payments and I.

A

Don't know that we have it in the US. This is where I can send you $100 and you have it pretty much instantly in your account, and companies like Monza and revelry are pushing the boundaries on this. You also liked platform engineering. You know likes going you like to github as well, but it doesn't like the change advisory board and what this means is. There has to be a quorum of stakeholders to prove any change into production, and unfortunately, JIRA is also on that list. So today what happens?

A

Is this a very sort of bread-and-butter, workflow pio product owner comes up with a new requirement: developer picks it off the backlog, works on it. Great we've got an artifact and then we need to deploy to production, and the current process was to use JIRA. Have those stakeholders login and then approve as necessary, but it was very difficult to see if it was ready. It could be error-prone and you had two disparate systems, because the pull request was on get help. Now.

A

What we did is use plunk and they're running this in production now and so all of the approvers come along and they simply approve the PR through the github API. Each one sends a web hacker to the Gateway and to his code, and when he's had a quorum amount, they will allow it to be deployed in production. So we were able to get rid of JIRA, for at least for that part.

A

The Evoque case that he wanted to look at is from Channing Connor over in the front row. Thank you for writing. This we'd be able to generate a montage from a video, and this is a kind of use case. You might see it a streaming company such as Netflix, for instance, you may be wondering well, did he write it in no js'? How do you put this together? Well, actually, he broke some bash and I thought this was great. He has this bash code pulled the template.

A

The COI up I know he's got bash over HTTP, that's auto-scaling with metrics, and you see in a minute that you can do even even call a stuff with it like connecting it to you. Events now add form free. They use a able to assess us full of their eventing, and so they built a event connector. We call it and you can build your own event connector to what you do is put together. Your function, you're, CRD or in your stack Yama file and just add an annotation for a topic you care about.

A

So in this example, we want to email, a customer. If we've got a topic called on-boarded, that's fired off, you just deploy a function to play a connector, and then they will invoke the function. That's necessary.

A

Another example case study is FIBA and we have mark peek in the back from vmware and some other folks that join recently. This is my friend, michael gash, and what he's been able to do is extend the functionality of a closed source proprietary product called vCenter that manages VMs such that when a vm is powered on the event can be observed and trigger function.

A

The use case that nvidia had for this was they had 12 different servers that were coming on online and provisioning VMS and they wanted to tag them and so using the vCenter connector and open fires they're able to achieve that worth. Lee, there's just an example of what you can do once you've got a micro service or an API.

A

Now the other product in plunk is Nats and again. This is an optional one. At the moment it runs at a synchronous bus- and this is great if you want to do messaging between micro services or you're. Looking at streaming data. The new version of mats has a global network that you can tap into, which is really really pretty cool, and then you can use it for command and control with IOT when we apply it to a use case. Let's take vision, Benko from paraguay.

A

They generate customer banking statements as PDFs the request comes in through the Gateway. It will then read the data from Postgres and upload the results of s3, pretty simple, but it can take up to two minutes, and so the customer will be sat there with a spinner.

A

What we're able to do is very simply without changing any code have that one asynchronously and so using a different route. The code will be ingested into a queue, a queue work or pick it up when this capacity and then the thing that's particularly useful here is being able to pass in a callback URL. So you can say when you're done I want to know here what the result was, and so then we can go off an email, a customer for instance, so you don't have to change a code at all.

A

You just changed the route and whether that's a micro service or an API or ffmpeg, it will work the same.

A

So some of the plunks, like users, have talks from coop con and you can go and find the videos on YouTube one of the companies at the moment. I think it's interesting is DX out of Norway. They have a large ticketing system in across the whole of Scandinavia and they're selling tickets. For events they want it to lift and shift a monolithic application written in laravel and PHP, and they are looking to use open fast for that.

A

They also have a pattern called the strangler pattern that they're implementing, where they're cutting pieces of the code off and rewriting it in c-sharp and that's one end of the spectrum. At the other end, we have vision, banker Shoji but then also live person and live person have integrated open fires into their conversational chat product and they allow customers to extend their their chat experience with customer raps while being able to write code. So remember the Prometheus alert that fired off.

A

If we had too many errors, they can actually fire an email off to your manager if you had a bad rating from a customer and what I think this shows is that we can use kubernetes for both sides of these use cases right and it can be great great for both. We don't have to pigeonhole it too fast when it comes to CIN CD.

A

If we may run three or four minutes into lunch, I don't want you to feel bad if you use Jenkins actually think that sometimes coming to coupon can be like a bit of an echo chamber, you feel like you're, not on the latest technology. All you need is to build a docker image and if that makes sense inside your business or your corporation, then go ahead.

A

But one of the things that we found is that teams like BT and Citrix and vision bank when they adopted open files, they went about building their own pipeline and that was actually stuff that we thought we could do for them, and so new users can just come in like DX and immediately have a pipeline ready up and running and not have to build that owl and be so repetitive. I'll show you what it looks like this is called open, first cloud and what we'll have is a completely managed system.

A

So if you're particularly technical, but maybe your team, isn't you guys can benefit from having access to this sort of platform. What I've done is I've created a quick and nodejs function and one using the docker file, because I said we can use docker files.

A

All we do is a git commit and get push and then, within a few moments over on my github page.

A

For my cubicle repo, we'll see, the code has been pushed up and there's a bunch of commit statuses that have appeared, and we hit details we'll see that those functions have already been built and deployed that entire process of pushing and putting into production has run. We've got container logs that have come back and we'll have live endpoints, so the one that runs in docker is giving us a calendar in bash. That's what I asked it to do and then we have the able one written in and node.

A

So you do not have to be an expert in Cuba Nettie's. You don't have to build your own pipeline. You can actually get started very quickly. There's somebody else touched on this earlier about kubernetes not being for developers. I think it is for developers, but there's probably two categories: there's the application developer and the platform engineer for application developers. It probably isn't suited to them, because it's so highly configurable it can be confusing to know what to do.

A

The declarative, API doesn't quite work. How we used to using systems as an API developer. We used to calling stripe and registering a new client, for instance, not setting a destination or a compass track track point, and the eventual consistency can be really confusing to write.

A

This idea that you may get the events out of order, so there are sharp edges and there is a steep learning curve when I learn this and I, don't know where you are in your Cuban eighties journey I did struggle with all of this I think a little bit better with it now after a few years, but I remind of remember, it was like so just to sum up the layers that we end up having our container registry because we're building containers, so that is the same everywhere, and so we have that portability of the service contract we can run in KNAT.

A

If we can run on cloud run, kubernetes allows us to go between clouds very easily. We get the same experience on our laptop as in the cloud on top of that. We have open files in the gateway so that we can build these services out and deploy them, whether through crd or rest and even autoscale them using prometheus and gnats can allow us to run things in the background.

A

So I think that plunk is actually easier than at least for application developers and doing all that on your own, there's actually UNIX. That I think UNIX. That set this precedent. It can convention over configuration, rain, kubernetes sort of went in the opposite way. It doesn't mean we all have to suffer with that same defaults can actually be really useful and getting you up and running so you probably do want an on route user. You probably do on a read-only file system.

A

We can set that up ahead of time and then, if you need to customize it, then it's up to you when you need to pop the bonnet that doesn't make things more maintainable using templates rather than having to have those three hundred lines of yeah more for every service. We can abstract and use those ropey principles and then, finally, this is something that has got a great developer experience because of being built by an open-source community and it is used in production by the companies are shady.

A

So here's a few ways you can get connected and I won't keep you from lunch any longer. You can join slack. You can try a free workshop where you can see what I showed you and go a bit further with it and then there's another cube concessioners afternoon with Ramiro he's going to show you how you can accelerate building those containers make them even faster than then it was when I did it and then on Wednesday come and see the link addy session it'd be great to have you there. Thank you very much.