Red Hat OpenShift Commons Briefings 2018 | OpenShift Commons, 23 Nov 2018

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: OpenShift Commons Briefing: Serverless, Kubernetes & Knative OpenShift Cloud Functions Natale Vinto

Description

Serverless in Kubernetes: Knative and OpenShift Cloud Function
Speaker: Natale Vinto (Red Hat)

Serverless paradigm is becoming a new methodology for developing and deploying applications without worrying about configuring any server to run them on. Hear the latest updates from Red Hat serverless open source solution based on OpenShift and K-Native, discussing all use cases matching serverless context for container-based workloads.

Natale Vinto gave an update on State of Serverless in Kubernetes: Knative and OpenShift.

Serverless, Kubernetes,Knative, OpenShift Cloud Functions

A

Welcome everybody to another openshift Commons briefing this time as one of my thoughts and things that I want to learn more about serverless and Kay native and how it all works with OpenShift and what the implications are for people using open shift and kubernetes I. Think a lot of folks have heard the terms Natali Vento who's with Red Hat out of our Italian front, is going to give us a talk about that. We have a couple of other folks that are on the line as well.

A

Who can answer your question, including Michael, has a blast so I'm gonna, let Nathalie introduce some stuff and get started.

A

B

I'm working read that does open shift, especially solution architect, so my role is really to bring to our customer to their container journey. We've always shift and I'm recently focusing in selling computing model server less computing model and to show that opposite can also be the platform for several mass for micro services or anything. The modern IT needs today.

B

So this will be an update on how to do service on upper shift and kubernetes, and lots of things are changing fast in this world, and it's good that we can also have an open source ocean solution in this new paradigm of service, which is really thrilling. The IT world, so just I would like just to share with you those news that we have from red dot and from the community. So if it's ok, I will proceed with with the slide yeah.

B

You know, of course, what is open shift, but I, just wouldn't I just wanted to focus again on the fact that almond shift it's a community. It's a group of open-source project that the the major project is kubernetes, of course, and okd, which is our upstream version of our ships container platform and around this. There is an ecosystem of a many open-source project that build and contribute to the OpenShift ecosystem.

B

And yes, it's also because it's open source and what the most important thing is that that we are talking about standards, because our shift, it's about the cloud, medical computing foundation and open initiative, of course under standard. This is very important in technology and we join very helpfully the Linux Foundation with building those standards. This is the ecosystem, actual ecosystem of a kubernetes and OpenShift. And yes, this is a serverless data center.

B

It is that just a joke from our own code, but it gives the sense of what people think about when what they think about fabulous. They may think that the service is something that doesn't exist, but actually is a it's. Not it's not like that. If we follow the Wikipedia definition is a computing execution model. So really, if you look at the gap diagram on the right, it's real some pipeline over some atom and trigger some action and give some rest so we've with several s computing model.

B

You have always despite blind this process and actually a lots of definition medal. The most famous is from marching forward. Yes, is something that lets. You run your server-side logic. Instead, less containers even trigger it, and if you look again on the diagram, is it's always event. Action result with some trigger, and this is called add, also function as a service and the most important thing that there, the cloud man, the cloud netting computing foundation, made a what paper are around several s, so the around several is.

B

There is also standard definition, and you find that the several is computing like concept to. Let you build and run your application without refining any server management, or you can write your function, deployed your function and execute the scale in somewhere. So what is a function as a service? And what is serverless function is a programming model turbulence like more looks like more are dealing model. So what we need? We need together because we need the something to run function, but also something to matter in this function. This diagram explain good.

B

How is the relationship between function as a service and service? So service is really the engine that lets you run and scale your function in this paradigm, which is function as a service. This is there are. There is an editor architectural revolution around this. You may have seen those texts also to read that Sam it in San Francisco.

B

There is this evolution from service. Then we went to micro services and now we are on function. Actually, the micro services conversion is still ongoing. We are still a lot of a model it's and legacy application, but here we have some much finer definition, which are thoughts, function which are single action, ephemeral and related to some context. So if you look on this day of this line with function, you have more predicative, but you lose control because you don't care who is gonna run your function.

B

You only get the fact that your function is gonna run somewhere, and this coop is gonna have scale up very fast.

B

This is a guy IT guy, before the micro services were, and now several eyes. We looks like we're in the teleporting in a Star Trek because it yeah a new world. How does it work? We have always that pipeline. If you remember some haven't fired some function, some string or some haven't, and this function going in output, something that need to be consuming to somewhere. Something else to be a micro services could be another function, and this is their force.

B

If you look here, this is the function. The lifecycle define it in the standard. So in this function, life cycle, you have the definition of your function, your code that need to be created from some circle and then virtually before been publish it. So you have many version of your function and you may decide to update or just to go through the revision of this factor. So this is in this time. The question is: is cerberus open source, of course, because we are talking about always about standard standard standards.

B

We are talking about flower netting computing foundation and cloud Amin's, which is the framework around the several has made about vanity computing foundation and all, of course, read that and limbs foundation are following this very well. So if you look in this slide here, we have a landscape of project around several s, so also the several s. Landscape is really rich, like the kubernetes one that we seen before. There are lots of open source project, but also a lot of public cloud implementation service like AWS lambda as a function or Google+ function.

B

We may know fabulous more for those public routes, but now it's possible to have also in our Bible 12 solution with open source implementation. If you look at this Cerberus timeline, you see this evolution of Cerberus, starting from AWS we are now Dante native, is the kubernetes implementation that is the baseline for surplice. What does it mean?

B

Yeah I just wanted to show you some data around several s, so what about organizationally interested in server list? What about technologies for server list? We see that there is some interest for backends web application process, automation and expectation around third server list, because the most important thing is that people are invited by this fact that service can reduce the scaling cost in the complexity and the operational cost.

B

So if you remember the previous slide, we lose control, but we we scale up very fast and what about the risk around several s if we lose control, we may be concerned about a security issue about testing, and so those are the main concern that are the. This is this report that read that made around several s technology region and what people expect service security and applicate way, one example of this and the technology integration. It's a really common landscape for HTTP web books and API integration nice-nice also to have it in the kubernetes events.

B

So if we want to collect from kubernetes and often shift events around odd slots, I'll check it. It's a good use case that we can have it in place and languages languages. There is a lots of interest around Java, nodejs and use cases, use cases about several s. For instance, in your monitoring, if you want to monitor something in order to trigger some action, if you remember the pipeline, it was event action and then the result.

B

So if you want to monitor something like verifying some water that can trigger some feeling, server monitoring is a good use case for serverless, but also storaged, for instance, if you want to reach to get some Evan from the storage and you want to make some manipulation or trigger some time action from your storage. Another good use case, of course, with api's, because is the fact that can scale a lot remember.

B

The services are always stateless, though, when you run your function, they don't have so much logic inside, but they can scale fast and go much in parallel, so the web api is also good use case and, of course, IO IO, T and Sand source, because you could have a multiple source. Multiple definition of multiple devices, lots of devices that need to trigger some action and service is the best Arabic for this use cases very common. As we said, the web books tasks like a cron, PDF generation or image manipulation.

B

All around this world and everything is a synchronous, concurrent and easy to parallel. Light is a very good fit on serverless. Well, what about not to put in server list, because it looks the panacea for everything, but it's not actually because you, if you have a some real-time application that require ultra-low latency spec for networking it may not a good place or a long-running task. That cannot be split into steps.

B

If you look at the implementation of service, there's always a timeout that you need to set up so a function according to run more than this timeout. If you go to something like AWS llama, it's about five minutes. If you go to some implementation like in Connecticut number shift, would be it's always five minutes. So you shouldn't violate the part of adding of some function that ran for a long time. You should have a little function that make little tasks and then maybe you call another function to continue the work.

B

So if you have advanced or complex topic is not a good fit and if you require lots of our specific memories, if you that's, because service is fast at first, so you run those function, you run multiple function and you collect the output for them, but you don't really care about. If any functional fails, it may happen. So you have to be aware that failure may happen in this world and let's go to the most interesting part. What is this? How to do service in my open source server list?

B

Well, it's possible we've open shift and OpenShift ecosystem of components. If you, if you look at how it's made of a shift of course, there is the operating system, there is kubernetes on top or kubernetes.

B

We have automated operation that we will the operators kubernetes operators, we have history of our service mesh and on top, we have k native kennedy's is the base for delivering service in kubernetes, and it's supposed to be something like the baseline that has to be used for from some Tyner implementation of cephalus, like if we think about cupola or if apache open, which is another service implementation. But it's more complete doesn't fit very well, take a metaphor or a camel k which is of 4k native.

B

So the concept here is really to provide the pace of some finer seems service implementation that could invoke a native to run your function inside kubernetes. So this is really important because it's really integrated into in kubernetes and you every visit stack to build your function as a service on top of.

B

Servus mesh with this, of course, is used by connective in order to make high security routes policies and also make a split around traffic. In your thoughts and operator framework, it can be used also to optimize the maintenance or the components around KNAT, but at the end, what is Canada play as we say it is an extension to kubernetes and as a building blocks for container bases application that can run everywhere.

B

If you follow up the, if we follow the definition on the kinetic project, but just to remember this, and it is really the base foundation for a surveillance arrived in kubernetes and offer ship to provide those building blocks to run the function and how does it with three major components? One is the field is a sum is a component in kinetics that can build container from your source code. So you you deploy your function.

B

The build is in charge to compile the source code put in a container and put in that air, putting it in the platform so putting in the read registry internal register shift, for instance, and serving the serving is the decor of the serverless implementation.

B

Around Kennedy, because it's the one that lets you scale your application and is the one that responds to heavens is avin driven model, so if there is any trigger to be fired, serve be native serving is the one that found your container containing your function and is in charge, of course, of maintaining the revision and around your function and in the bearing events, is another important, very important component around kinetic, because is the one that make the the max the pipeline between events around Toobin ad and the events that your function need to be too logical.

B

If we go in depth in the kinetic build a component, we have this object. Of course we are here in our shift and kubernetes, so everything is I can impede the clarity by infrastructure as a code. So how does it works?

B

It starts from your source code when there is your function, so you have to tell what is your function in this example is on github or somewhere, and you have to also to define which step to compile your function, so you have to tell which container to start to the planet, to compile your function and which image to use to build your French right and there is a native serving soak. A native serving is the really the core of this world. You have another terminology here, which is the service this this service is not take.

B

A community service is a top level controller that manages the route and the configuration. So you have the service which is which is a high level representation of your function. This service manage the route. The route is actually some route. That is an address that resolve to the top running your function and the configuration the configuration is the there is, the layer is the representation of your revision of your fraction.

B

So if you remember the first diagram in the specification about the functional lifecycle here, the function life cycle, it defined the in terms of a configuration and provision, so you define some configuration and then you, your function version is an era vision. You can decide to split the traffic across the rectus provision, keeping the function alive or just let the connective spawning a new function when it's needed. So this is the very powerful mechanism that makes the Surrealists implementation, responsive and, of course, on-demand. You don't need to keep the function.

B

The function will be scale it automatically up or scale it down the in terms of some trigger and the eventing component of kinetic, which is the one that connect inside kinetic internal event and external events. It's a kind of complex implementation that may require also some streaming or messaging system like Kafka. In rezoned implementation. We have Kafka, we have a reference in 410 AD by bending. This works is really under going. Is it's really improv and is following the specification from cloud events?

B

So everything that you see here is some specification is some some standard that has the first declared and then implemented as open source. The question is: is kinetic enough for service, like we've, seen like we said before, tentative is a building block to run a function as a service in our platform. So the idea here is yes is, can run service, but some other related implementation should invoke tentative and take care. Maybe you'll make their life easier to developers to simplify the creation of environment, of programming, language environment, for instance, or just there.

B

This execution of yourself for offer for function so is not complete along with Kennedy's it's enough, but not complete. We need some light runner on top that can do the function. Assess the capabilities of tentative are really.

B

Really adamant to standards, if you know the twelfth factor up, is a definition of.

B

Writing some application best practice to follow to run application and the kinetic custom resource definition for for this. I really have some one-to-one mapping to the 12 factor, 1/2 definition. So the component that we seen before about yeah.

B

Configuration revision.

B

A

Everybody, if you're, you did I'm trying to mute everybody natality, sorry, there's a person just just came in named Bob I'm, not able to mute sorry about.

B

To come back to our concept here, so we have a one to one month between the 12, 5-12 factor, 1/2 definition and a native customer service definition in terms of configuration in the 12 factor up there is the one one point is to keep the coda configuration separate and configuration erudition are the way to go to implement it, so each configuration change trigger decoration of new relation is an is about this topic again and with routing and services. We have this possibility, though this is an important another.

B

Another time we see that we have a standard and we have an implementation that followed this tampered. These standards in the in the open source way a little demo to show a native capabilities around Cerberus. This is the bitly URL that I prepared. If you want to follow in the world, but I think we can, if we can go directly.

B

So if you go in this in the link in that link that bitly, you will learn to my repository here that contains how to run the connective on on leadership.

B

This exam in this example here you will run mini shift with Kennedy and you will follows this example. Those example are inspired by a commercial comfort example and his blog post around Kennedy van efficient, real good. I really suggest you to go there.

B

I've already is example to build this demo and once you configure mini shift with some some some option, like memory cpu on disk and also additional controller hooks mandatory to install a native and east you, oh you will you have to configure some security content constraints because ii still need it, and and then you you're gonna, deploy a kinetic, can use as we've seen before easter. So you have to deploy this in in this case and then you will deploy Canetti service energy serving them with deploy also their components, the energy field and kinetic.

B

So you have this tiny religion, tiny release of native that you you can you can install in this example. You have also to to add this role, and then you can try your your application, so you you can try your first server les implementation in this example.

B

Let you go through configuration and revision in order to to explain you.

B

So if you go in in this section for, for instance- and you follow the prerequisite, then you will be able to run your first server less your face function in the opposite, ok native basis with the kinetic field, I will just go through my mini shift here. So you see here, I have my mini shift with my k/d is a mini shift 311 and it's empty for the moment.

B

So I will, let you see not GS function that prints something and to radiation of this function expressed the in terms of configuration.

B

So we I just you- have just to clone this repository and follow the primary site and follow these steps. So here I, don't know if you see good I'm going to create a template template to tell to kinetic builder how to build my my function and then I'm going to create my configuration. I will just show you those files once I created it so once I create my configuration, I need also to define the route diagram before we have a configuration we have about.

B

We have a service and in our case we have also builder if we follow our.

B

If we follow our mini shift here, we can follow what is happening under the hood in the Evans initiative, because if you go to the to the heavens and look for look for what was going under the hood, you see that to some point some part is not being created.

B

Some configuration has been created, some route has been created, so you can follow what is going to handle the hood and in the while our thought finish is finishing. That is a compilation, because what is happening now is that my not Jes function is going to become put inside a container. We build that is our tool to build and create container that the substitute yes, which is a more over compliant to authorized standards and consortium and once is built. You have a new pot in your in your in your deployment.

B

You have a new deployment representing your function and this deployment will create running your front. So here you see, this pot has been created and one is ready. We will have here our for running and there and answering with our function. So just it's ready, but I just wanted to show also because in the demo we we have to invoke the function by common line, but I will I will like also to show the source code of course of this function.

B

So the function is really and not yes, function that just print out hello, openshift, and if you go, if you keep some environment variable in print, so it's really simple and there is the build, a script that will let you build your.

A

B

A

B

Your function and and then you have let them plate that define it. So, in this case, I will show that I'm going to call this script, which is ready here for us just to invoke our function. In this case with K native, we need to invoke our function using the internal IP address on the platform because of how it's done before. So, if you just execute the script here, I'm gonna show you in this case.

B

If you just do it, you will have the function which is answering so this is the function running in our automatically created for us by Fayed. It's the built with bilder. This is really powerful and I want to show you the definition, because in the example we use it with template. So we have. These are not built that template that what we from some tilde image in this case yarn from the guise of a no Jes and make.

A

B

So it prepared environment to in terms of steps to put our function inside, and one thing that you have to look for also is the configuration. So our first revision was about this configuration here, and this configuration is based on the is properties repository on github is using just this repository and is finding that not Jes code because I'm giving the context here. So it's not really guessing it's just giving the context still and telling it that you need to use also this template to build our my function.

B

So he is a word that these are not yes, because I'm gonna tell which template to use in this case and I'm gonna tell we. Where is the function which, in my case, is this this function? If we come back on on the demo I just to show you that we can go through another revision, so this is the second revision in the second revision. The defeat of this second revision of what I do is just changing the environment here and I'm writing Chow, which is.

B

If you just go through that second revision in your configuration you're applying this as Yama file again, you will see in the mini shift again that your function is created. So you have a new function which now is answering to your request and if you see is changing, the text is not any moral.

B

Overshift is Chow open shift because I have a new revision of my code, I'm using a new environment variable in this case, but if I change the function in this case, I just need to make a new revision that we build a new container for me, a new image and that would be put in a view for the running function.

B

Another cool things to see here it's how to manage revision so I can. How can I go through those revisions? In this example, I I just went to the give to that this route that first revision all the traffic sent to this road. In fact, if you, if you look here what I've done after the point of first religion is tell that the same and 100% of all the traffic will go to these services.

B

But if we want to just give some 50/50 traffic to them, so if you see here is referencing the revision name, the addition name is one thing: this is a current with the definition. Also you can see in the deployment. So here I have two deployments if I want to split the traffic across to deployment which are to revision in the K native world, I can create a route and Express this traffic splitting in term percentage in the yellow 5.

B

So if we just follow the example in the in the in the red meat again, if we want just to split the traffic- and we do like this yeah, we can try to invoke our function again and see that the traffic is really something like 50% with symmetric that they get inside so and we can also, if you follow the example here, you can define multiple traffic splitting 90% and 10%, so they're really them the metric is the same. The logic is the same. You define a route.

B

You mentioned some revision and you give some person to do this revision in order to have this traffic splitting another question that the another thing is interesting to know is: those function exists only until a certain address off, so those function by default exist for five minutes. If you don't fire them, they exist only for five minutes. When they go down. The creative autoscaler will create a new thought that contained that that the latest good revision that you deploy- and if you see this is possible, is the changeable.

B

If you go to the train and observing which let you also show that k native up there because does a component, which is the activator, they are two scalar, so the other scalar is in charge of doing this and it works with a config map.

B

So if you want to change those parameter in terms of how much time you need to want to let your function scale- and this case I change this case- if you want just a lower and a high on that thing, you change the content map over here, and you have the change of this. Of course, you don't have to put one hour mark because you are violating their paradigm, so it's not any more sterilized something something else, and another thing cool to see is the Easter.

B

So when you start a native, you also have East your inside and east, your English gateway and egress gateway are the one that you deal when you want to invoke your function. Of course, in this example, you are invoking by the in terminal IP address, but since this is under development, work and developing is going really fast and really in progress. You can. You will also be able to define your route and Ike level level and invert your function and.

B

Say yeah: this is: was there this little demo? So if you want to check out it, you have just this.

B

Function so oven shift. Con function is really effort to bring open source server less to your libraries cloud. So if you want to run your function as a service in your private club with not only on the public cloud but your private cloud, if you want to run as open source implementation, you are to extend, contribute and work on function.

B

You can do it with a red dot front function, which will be our products around this top. So if you want to work with community project, there is always ok, D&K native implementation. If you want the products and repress product, there will be read that function, giving support on writing function and just also executing in the same or read that way: community first and open source. First.

B

Wouldn't be any more that the empty place, but maybe a rocking and bruise represent when you can you have you can around your function.

A

Yeah, that's awesome, Natalie I, love that last gift and it wonderful this. This has been really good. I'm and there's been a couple.

A

A little bit of chatter Michael, hasn't lost is been answering a few questions mostly for me, I, think in the chat and so I'm just going to unmute him quickly to and we'll see, he's gonna back, and so there Michael, you should be there, but I'm a little um I think I'd, like a little more clarification on the relationship or the dependency on sto for K native I mean how, and maybe there was an earlier slide or something that I that just went by too fast, but that that is still a little bit confusing for me, and maybe you could tell us, you know why you have to install sto in order to use K native, or is that just me missing a point here.

B

Very complex also is still under development, so something is they find it really in a few times. So it's it's complex, really new, but it still is using the bike, a native because of a service mesh. So when you want to go across those revision, those traffic splitting you need something that lets you attach your service dynamically.

B

You need something that need. Let you have some connection between those services and one thing: I didn't show you here: maybe I can show you.

B

Anymore, but in any container in any container representing your function, there is East your sidecar here because he still inject those site cards in order to let your revision be third shift in production dynamically as a service inside the overshift, though the relationship is really related on the fact that is to enable fast the creation of new addition and the splitting of traffic across the strip. That's why history is used year and is a mandatory I mean it's some summer requirement by kinetic.

B

A

That that makes this for me and then I'm looking to see if there's there's other questions in here, it seems to me that that K native is is very early days, so I mean I, haven't seen or heard anybody yet using in production. In any way, are you seeing that people thinking about using in production now or is it? Is it really still very early days? It's.

B

Really is really really it's really. Development is really ongoing progress now, for instance, the creative eventing part is really a draft, so they are just following the CN CF and the cloud events specification and implementing right now that implementing it right now so is not complete. I wouldn't suggest it for production for sure, but is the future around the running, Cerberus paradigm and computing model in? So, if you think about one years, it would be major enough major and there will be also many implementation on top of plan on kinetic planning.

B

In order to let you run the function, I can say is like east your one year ago. So can it even right now is like is your one year ago is something that is a really new cool is really needed by everyone, but it's not ready for production for sure and also for testing and need some some time, because some components are not ready or they are just in definition right now, but.

A

That makes sense, so thanks thanks for that, I was I, hear all of the new projects come out, the CN, CF and stuff that people are working on at Red, Hat and I always see this stuff come by and then the K native stuff is mean there I think there are like three or four talks at coop, con, North, America and Seattle coming up and it's you know one of those things that popped up and came up really quick, so I'm, looking forward to going to the coop con and sitting in on those sessions.

A

This is going to this session. That's alec has really helped me personally understand a bit better. The you know how it all works together. So thank you very much for that. I will upload this on youtube. People have been asking that as well probably take me until tomorrow morning to upload it and get it formatted for YouTube, along with natales slides and if you are coming to coop con. Please consider joining us for the open ship commons gathering the day before koukin on December 10th.

A

We're gonna have an all-day festival of all things, open shift and upstream there, so that this is a really good point place to also meet some of the folks who are working on these upstream projects as well.

A

So a number of them will be in the room and we'll probably have at least one or two more talks on ke native and Michael who's having technical difficulties with his bandwidth in his hotel room wherever they have them stashed today, will probably get you to do another one and going on a bit of a deeper dive on high performance applications and deploying those so there's lots of good content.

A

Coming, like you said, Natali that this is all very much new early days, but it is definitely something we at Red Hat see as a necessity a necessary piece of the the cloud native ecosystem. So what we're looking forward to working with the community on it and making sure that it's in a year's time, maybe sooner ready for production workloads, though yeah.

B

I think we are in a good part, I think by the way, connective is the standard now the fact of our kubernetes for running surplus workloads. So in the future we really hope would be enough, mature and stable tool to run those workloads in production, and we are working as rather than or just community to do it and I would like to thank you for giving my giving me this lot to talk about the Canadian appreciate and thank you all. The people in this ending servant I know, of course, available.

B

If you want to get more about this or just even more information to come in in Seattle, there would be a live demo for from our rendered view that we'll also do some them about a native events. They are working on it, so it's very interesting to attend on those that they will have and, of course, I'm looking forward to join you guys at Commons in yah, Oh, awesome.

A

Natalia, if you can send me the link to the session or the time when that demo is going to be done, I'll include that in the blog post on open ship comm that contains the video and the resources links to and along with your slides, we'll get that all up and out there in the next day or so so, yes again, and if anyone is coming to coop con.

A

Please do join us for the open ship Commons gathering the day before, if you are already registered for coop con reach out to me and I can send you a promo code to get you in for free. If you need to, and we've got that lots of good people on on the agenda- and this is going to be probably one of the many topics that we cover off on that day as well.

A

So looking forward to it all the thanks again, Natalia I know it's later in your day over there in Italy and I really appreciate you taking the time to do this and yeah again this. We started with this conversation in Helsinki and trying to get you there, but I'm really glad that we finally got got you on on the line and able to give this presentation. So thanks a lot.

B

See you sound and talk to you soon all.

A