Knative Knative use cases - Community videos, 7 May 2020

Previous Meeting

⏯

youtube image

►

From YouTube: WEBINAR | Behind the scene of FaaS at Scaleway | Intermediate

Description

🚀 Behind the scene of FaaS at Scaleway: When the magic happens between Kubernetes and our Serverless products
🚀 Level : Intermediate
🚀 Broadcasted: May 7th 2020

🚀 SUMMARY:
0:00:20 - Scaleway's Presentation
0:11:36 - Behind the scene of FaaS at Scaleway
0:37:54 - Demonstration's repository
0:51:07 - Q&A

🚀 WHAT YOU CAN TAKE AWAY FROM THIS VIDEO?
- Behind the scenes of Serverless at Scaleway
- How Scaleway builds its Paas/FaaS experiences for its users?
- The integration of tools such as Kubernetes or Knative to manage groups of shared resources ?
- Finally, the use of Object Storage or Load balancer to distribute your loads as well as possible

All Scaleway teams 🚀

________________________________________________________________________________________

🚀 PRACTICAL INFORMATION:

- Checkout our integration to Serverless Framework to easily deploy your Serverless applications on our Platform 👉 https://github.com/scaleway/serverless-scaleway-functions

- For more information on Scaleway Object Storage 👉 https://www.scaleway.com/fr/object-storage/

- About Scaleway Container Registry 👉 https://www.scaleway.com/fr/container-registry/

- About Tekton Pipelines 👉 https://github.com/tektoncd/pipeline

- More information about Kaniko 👉 https://github.com/GoogleContainerTools/kaniko

- Checkout Scaleway Kubernetes Kapsule 👉 https://www.scaleway.com/fr/kubernetes-kapsule

- About Scaleway Load Balancers 👉 https://www.scaleway.com/fr/load-balancer

- To learn more about Knative Serving, the core of our Serverless Solution 👉 https://github.com/knative/serving

- Here's the demonstration's repository 👉 https://github.com/AmFlint/serverless-scaleway-apollo-typescript

- Subscribe to beta for Scaleway Serverless (FAAS and CAAS) 👉https://www.scaleway.com/en/betas/#serverless

- Working at Scaleway 👉 https://careers.scaleway.com

________________________________________________________________________________

Stay tuned by subscribing to our channel! Do not hesitate to check our website.
○ EN: https://www.scaleway.com/en/
○ FR: https://www.scaleway.com/fr/

Follow us on social media:
○ LinkedIn: https://www.linkedin.com/company/scaleway/
○ Twitter: https://twitter.com/scaleway
○ Facebook: https://www.facebook.com/Scaleway

For doubts & queries about Scaleway, feel free to ask on Scaleway Community:
○ https://scaleway-community.slack.com/

Scaleway, a leading European infrastructure cloud provider covers a full spectrum of services for professionals: public cloud services with Scaleway Elements, private infrastructures and colocation with Scaleway Datacenter and dedicated servers with Scaleway Dedibox.

© 2020 - Scaleway

A

So I'm glad to be here today. My name is Louie my joining scalloway about six months ago, originally as Euclid social architect and then recently, I joined the marketing team as a product marketing manager and this model introduction what lasts more than 10 minutes. We'll talk about skill way, who we are, what we do and then we'll cease. How do we reach those several s? Concepts if you need more information about scale away, feel free to have a look at scale. Wait!

A

Welcome website, you'll, find more information about our other products, even links to some documentation, testimonials, etc.

A

Scale weigh scale is the cloud branch of the group Illiad for the French ones. You probably know Iliad it's the motor brand of a free free, mobile free box, and we are the the ninth client provider in the world and the second one in Europe, and we are mostly a cloud service provider and shown your impairment.

A

We currently have more than 250 collaborators split into two different offices, one in Paris the headquarters and another one in Lille. Those headquarter are. Those offices are called Lamaism in French and the home English, because everything is made to fill you like at home. It's a great for experience and by the way we are hiring so feel free to have a look at the opening position and feel free to join us. Our some of our collaborators as well are in our data centers.

A

We currently have I have five data centers and four in the Ile de France, in Paris and one in Amsterdam and in the year 2020 we will open a new one in Holland Universal.

A

We also have a many international clients and we have presence in over 160 countries.

A

So that's scale weigh scale is the main branch, the umbrella branch of 3 sub brands, which are scale weight, data center scan, wait, 30, bucks and scan web element so scale. Y data center is mostly focusing on private infrastructure. Basically, what we do is we provide space will provide electricity network aircon for you to bring your own servers in our infrastructure scale. Wait. 80 bucks on this side is mostly focused on dedicated servers. We have more than 150 references depending on your need in CPU, bandwidth, RAM, etc.

A

I'm, pretty sure you will find what you need with kuwaiti box and on the last part scale. Web element. Scary element is our public cloud ecosystem and is composed of six main bricks. We will see them in the next slide, so the first one is compute. Then storage, network, IOT, a I and obviously the tools to manage everything. So concerning the compute break and we have virtual instances whether they are for development purposes or general purposes and also GPU instances mostly made for AI training, algorithms or video, including those kind of use cases.

A

Then we have the pyramidal bare-metal our dedicated servers, but in the cloud and they enjoy the whole club ecosystem.

A

After that, still on the compute side, we have tribal entity capture, which is the control manager, control, plane, the container registry and what we are going to talk more in more today. The server list, and several s is speaking to different sub products, which are the fast the functions as a service and the caste compare ourselves to manager. On occasion on the storage part, we have object, storage, which the protocol is interoperable between s3 object, storage.

A

The block storage are made for piste persistent data. If you wanna make some data persistence on your instances, then the cold storage stands for C 14 C 14, so it stands for carbon 14, and this is made for long-term archiving.

A

Then in the database, which are post post grade and trail and MySQL has just been released in early access feel free to register. If you wanna, if you want to test it, then we have the network. Iot I won't go much into details, the other sub bricks and, finally, the tools where you can manage everything, if even if either it's with the console.

A

So if you want to create your service manually or using our API, CLI or or even tariff, or for you to automate your deployments and we have many clients, large clients on, as well as small clients, free lengths, etc. In our main clients, we have Peters Lagoon qua a dub, adopt homage, won't, give a winner max, etc. Now, maybe, let's dive more into the ceremonies concepts? How did we reach several s so originally?

A

When you want to build an application, you need to buy physical server, so either you were buying it or you could range its monthly. But at this point what you had to do is manage all your hardware, but you had the full control of your of your server. Then came the virtualization the virtualization is made like well, you could split your actual physical servers into different virtual machine, but you had to take care of the virtualization yourself, but in the end you could focus on your application part on one virtual machine.

A

Then the cloud computing came and at this point you were giving only one virtual machine, and maybe one of the concept that is different between those two is that some cloud provider made those beautiful machine build how many so that's a huge step so that you can switch off your virtual machine when you don't need.

A

It then arrive the corners, which are great, because you can only focus on your application side, but if you want a manager, corners environments that have the power up in auto scaleable, etc, it can be a bit complex and then the crowd came to service. So the carousel several eyes are either contours of the service. So everything under your concern is managed and you only have to focus on your application, all the functions as a service, and in this case you could split your application into a different task, and this task ran independently.

A

I won't take more money a lot of time into those those diagrams, but in the end, what you want to you. So that is your application. So for the physical servers you have the full access and you have to manage every steps which at some point can be great. If you want to control everything, then on the cloud computing of virtualization, you have only your virtual machine, but you still have to manage your operating system, update your dependencies, etc, but in and in the end your vacation can be can be up.

A

When you arrive in corner ization, it can be a bit complex if you wanna auto scale, and if you want to manage everything about what what is under your container. I won't explain this this diagram, but what you need to focus is you: can you can host your application and this application is a hotel, Hill, auto scallop.

A

And then the servers, the platforms as a service or the function is a service and they lets you deploy an application without worrying about your own, fresh, real infrastructure and scaling or for this functional service. Without worrying you still don't. We need to worry about your infrastructure in scanning, but here you deploy a single function.

A

Let's maybe take a use case so that it's going to be clear. Let's imagine your SAS company, you are a great company and you develop a SAS software. So maybe some of the tasks that you wanna split this may be ran some AI prediction or send different reports, and in this case each different teams can develop their own tasks with their own language. Some other example could be with the user management.

A

If you want to change this profile picture or if you want to check the token token validity or even if a user, when I recover all these data to make sure we're not over exploiting it. This is a single task and that can be split into one specific function, so the main advantage that you only pay when the function is executed each code, each team can coding in a given the language, few watts and test.

A

For me, the clever point is that you need you: don't need to worry about your infrastructure and you only need to focus on your application or your task and maybe.

B

The main drawback is.

A

That it's not applicable for every application. The company need to be cloud mature enough to be able to use these new concepts.

A

If you wanna, if you might try it feel free to register on the private beta of a server list, and you can you can test it to give us our your feedbacks. Thank you for listening. I will now pass the pass the mic to Othman and he will get dive more into into the technical part.

A

Thank you. Thank you.

C

Hello I'm going to share my screen now and then we can start the presentation. Okay. So thank you reason Tatian now on deeper into the technical parts of this service platform, one scare away, so my name is Duncan. Maslow I am developed inside the server listing and I'm going to talk about, what's happening on our service platforms on both products, functional service and continual service at scare away. So this talk is called behind the scenes of fictional service at scare way when the magic happens between kubernetes and several s projects.

C

So before diving into the technical part, we will just do a short introduction about what Savalas is so, as we said before, the the main points about server list is that you get to delegate your operations. So in this talk we are mainly going to focus on the function. L service pods. Although we'll talk a bit about the container one as well, so delegating your operations, it means that, as a provider, we are going to provide you with some features such as scalability and automatic availability for your operations. So for your application?

C

Sorry, so you can focus on building the code for your functions and containers and we get to manage the runtime of these applications. We also get you manage the ad where of the application and the service ad host levels, so it allows you to focus on your business, so you're going to save money because you only pay when your fashion is being executed. So, for example, at folk a in the morning when your function is not being executed, you're not going to pay for resources because you're not using any any resources.

C

You're also going to save money on the DevOps system inside as you're not going to have to pay someone to maintain your infrastructure. Usually we try to products to provide easy to use products to help you get started and deploy your applications and publications very easily. This allows you to focus on your business, which is a code of your application.

C

Also, there is the common use case that we're going to take for this talk as a web developer. I want to deploy my server list application so that I I'll have to go through multiple steps to do so. The first one is to develop the application to provide the code for this application. Then I'll have to upload the code, so my provider can take and then take the code and deploy it into the production environments and then manage the runtime of the application when I'm, for example, executing the code of this application.

C

So now we are going to to deep dive into the behind the scenes of the products before that's just a little bit of terminologies. So what is a function on our platform? So as ice as the title of the talk implies, we are using kubernetes to run our users workloads. So in order to deploy your application on the on kubernetes, we need to transform your code into docker images.

C

So one function is equal to one docker image and then, when we deploy its application into kubernetes, we need to tend to replicate this application into multiple instances to manage the incoming traffic. So each function instance is going to be one or the inside kubernetes. So we are going to use the scheduler capsule projects, a fully managed community cell service products provided by Skagway to host our applications.

C

So this is the classical lifecycle of a service application. So there are multiple steps here that we split between the responsibilities involved. There is the school, the rectangle purple, one which is involving the responsibility of the developers. So as a developer, I'll have to code occasion to provide the code for this application to the provider by uploading it somewhere. Then my provider will have to build this application, so I'm going to talk about it later on and then, as a user.

C

My final responsibility is going to be triggering the application, so, for example, I can trigger my application by sending events like HTTP requests, former from Ted client, for example, and on the other side there is to provide a responsibility. So when the user finished, you quote the application and he's going to upload the code on a private space, then, as a provider, I have to transform this code into a useable binary. So in our case we are using kubernetes. So the binary that we need to deploy is going to be a docker image.

C

Then I'll have to scale the application instances once the user was to trigger an event and I'll have to execute the code inside this these instances.

C

So we're going to go through each of these steps, one one by one and describe how it's happening on our service platform, so the first one as a user I want to code my application, but how do I do it as a functional service provider? We are going to provide you with multiple runtimes, which means that we are going to wrap. You transform your code, your function code with some transport logic to to fit our whole infrastructure.

C

So, in order to do this, we're going to provide you with different languages like nodejs version, 8 and 10 python version 2 and 3, and go wrong as version one that 11 plus, which means that you get you used to go modules for your applications.

C

We made sure to build compatible mode runtimes with Amazon lambda, which means that, as a user of Amazon lambda, you can develop your application. The same way. You would with square away, which means that you get to test our square way several platform with your already existing workloads and it's way easier for you to get started with this platform. So we're going to talk about that inside the demonstration later on. So here is a function. So when we deploy a function, our service, we we we want some.

C

We we have to know that one function running in production is basically one function inside our code, so it is very lightweight code. So, in this function it is really is a simple function. I'm going to research, you receive HTTP. Events is going to trigger the code, so, for example, HTTP calls and I can send back HTTP responses. So here inside these functions we can see two parameters. The first one is going to be the event, though the event object, contains the data about the event that triggered application.

C

So in our example, it's going to be a HTTP request. So when I send you will request to my function. We are going to create an event, object and pass it to your function code, and then you can get access to things like HTTP body of the request, Corie parameters, paths and methods, and all that and the second objective pass as parameter is going to be the context. So it contains data about the execution kin context of the function. So, for example, it contains information about the name of the function.

C

The runtime used resources that you are located to the function for some. So in this simple function, we are just taking in an HTTP request and we are sending back HTTP response with the body of JSON object with a message and the status code of two hundreds. So this is how you would do it on Amazon lambda, and this is exactly the same way to do it on the square wave service platform.

C

So in order to make it easy to deploy your application, we developed a plugin for the 7s framework, so the 7s framework is a CLI tool, open source that you can use on your local environments to make easy to deploy product production environments. So on several s platforms, because when you you're going to you, want to deploy an application, you'll have to go through multiple steps. That can be a bit tedious, so this plug-in makes it very easy to deploy your applications, we're going to use it in demonstration later on in term.

C

In the talk, so we can use a simple tool: that's rum! We can use the command server. Let's deploy, and it's going to go through this. These steps, the first one, is obviously to create namespaces on both products. The the first part of next place that we're going to create is a project inside the square wave service platform namespace inside square. We service platform is basically just a projects in which we are going to deploy our functions and containers, and then we are going to link this novelist in space with a container registry namespace.

C

So as develops inside the server system, we are also managing the container registry products. So this is basically a product in which we can store docker images or OCI compliance images. So we are going to build a docker image and push it into this private space. That your organization owns once we created both of these projects. We can create it functions that we are going to define with the service framework so, for example, my basic hello world function and then we are going to package the code upload.

C

It's on the private space inside the descale way, object, storage and then our IP eyes are going to be able to download the code, transform it into a docker image and deploy it to scale well capture the kubernetes platform.

C

So here, if we want to use this, the service framework will have to provide a server less manifest with the Yamal format to configure our application. So there are two kind of configuration. The first one is related to the provider, so we can use the server list framework to with different plugins to deploy a Savalas application to multiple platforms like as your functions, good cloud functions, Amazon, lambda or obviously, several scare away.

C

So here we are defining that we are using the plug-in to deploy service applications on skyway by using the runtime version, Python 3, and we can use our credentials from this query. Console to authenticate G is every API on the bottom.

C

We can see the configuration related to the function that we want to deploy, so we have a function ski inside which we can define multiple functions, so here I only have one which is called my function, I'm going to define my handler, so my handler is perhaps to the file in which I'm going to write my function code.

C

So here I have a handle function inside the handle and blood, but Python file, and here I just have to write a path to be spine, so handler rods handle for a handle method inside the handle the Python file, and then I'll have multiple configurations that I can use like scalability in scale and max scale and resources, a location I located which are going to be allocated to my function at runtime and then our environment variables, which are really useful because they are safely encrypted within our system and they are injected into your function at runtime.

C

So you can use environment variables to configure, for example, and the database connections and alter. So once we run that commands the server. Let's deploy comments. We are going to go through multiple steps, as I said before. The first one is to upload and build the code, so the server list framework is going to create a dot zip archive containing the rest of our files used for our functions so, for example, Java, scripts and Python files. This is going to upload it's on a private space on the stairway object storage.

C

Then we're going to build your code into a docker image that we are going to save to store inside a private space inside your container registry. So before going forward with the build step, we're going to talk about what is actually inside this docker image, because we only provided a simple function inside our code. So how do we get to run your code in production inside this docker image? We have obviously your function code and we're going to wrap this function code with the thing that we call the runtime code.

C

So this runtime is basically handles logic, for example, receiving HTTP requests, transforming that data into events and into context dynamically in imports and invoke your function code. So this is the part where you select node.js, for example, or Python or calling, because we have specific logic for each of these languages, obviously, and as an end user will be able to trigger your function code by sending HTTP requests to this HTTP server provided by the runtime code.

C

Okay. So how do we transform the code that you provided on the object storage into a docker image that we are going to store inside the container registry? So for that we are using a basic build pipeline? We are going to describe multiple steps that we want to run. You transform your code into a docker image, so to do that we are using kubernetes to do it, so we are using technology that opens of technology. That's called Tecton pipelines. This technology is based, builds on top of command.

C

It provides us with the crystal resource definitions that we can use to deploy a build pipeline. We are going to describe multiple steps and we're going to deploy this resource, and these steps are going to be run inside a kubernetes cluster. So there are two major steps that are being run: to transform your code into an application. The first one is to configure the environments, so we have to download the code that you uploaded on the object storage.

C

So to do that, we are using an application that we develop inside the server this team we're using a application. That's called the code finder, so its purpose is to find your code. Basically, it's going to download the code from object, storage and the second step is going to configure the environment so configure the environment means that we need to find rights. Doka file, matching the runtime that you used and obviously the runtime code.

C

So as I showed you before here we're using some specific runtime code to wrap your function code to invoke it dynamically based on incoming events. So we need in this little pipeline to get the right docker file for the runtime that you chose, for example, node.js version 10 and we're going to download also the runtime code. That's going to start the HTTP server and invoke your function. The output of this birthday is the package code.

C

So it's basically a directory inside the file system in which we are going to find your JavaScript files or Python files, along with the the Doka file and the runtime code. The next major step is the transformation we get the packaged code in entry. So we get the word your code with the docker file and the runtime code, and we are basically going to transform to run the build of this docker file and push your the builds, builds image inside the stairway container registry namespace.

C

So, in order to do that, we're using the technology that's called Kaneko. This is running in through burn. It is so can eco is a open source project from Google, which is really useful because it makes it possible for us to run build inside the kubernetes without having to share a common docker socket between each container. So each build is independent and very secure, and then we basically build and push this image. Then the next step is just to take this image from your container registry and deploy it to our kubernetes cluster.

C

So once it is deployed, we are going to do to talk about how, as a user I can trigger my application. How can I acute the code running inside my function, so here I have a little diagram on the top. We can see the difference, even sources that are supported by our phone right now. So here there are two types of even sources reported. The first one is the HTTP request, so you can trigger your application by sending HTTP requests. So this is the little user on the top left.

C

You can use it, for example from front-end web sites. The second one is the cron job application. So here we can use time-based schedule to trigger our functions, for example, every day at 5:00 p.m. so both of these events, AUSA's I'm, going to send HTTP request with event data data into the app kubernetes cluster. So, in order to execute your code, we first have to forward the incoming traffic to the right services for your fiction instances, so this is done by a load balancer.

C

The only way to expose the traffic from inside your comments that you, outside of it, we are using a scale, a load balancer to do it as an ingress gateway, and then the traffic is forwarded to different applications running inside the cluster. So we can see here. I am I drawn three types of application, so the pink one is two instances of the same application just replicated the black one is another function from a user only one instance of this application and the little buffer application on the bottom right.

C

We're going to talk about it really soon. It's going to be really useful for the scaling up and down parts of this system.

C

Okay, so now I'm going to talk about the technologies that we use to to get to get it all working. So we are using both products from the scan way ecosystem like the scanner, a load balancer as an ingress gateway for the cluster, and we are also using open source technologies like humanities, to provide the cron job feature and also using envoi proxy and easier to use.

C

Networking features like a rate limiting, for example, and the authentication between the different services, so ok, now I can send an HTTP request that the skyways havelis platform provides an HTTP endpoint for and when I send a request to this HTTP endpoint. We are going to trigger your function. Now we are going to talk about how to be scale automatically your function workloads.

C

The first simplest simplest case is: we already have a function instance that is available inside the cluster and we receive an HTTP call with an event and we forwarded we forwarded to the function instance. We are going to execute your code and send back an HTTP response.

C

The main goal of server list is that you only pay for the resources that you're using. So when your functions are idle. We basically killed your functions. There are no resources allocated for your code, so when you send an event and no function instance is available, we are going to experience the call stats. The cool stuff is basically the phenomenon when we have.

C

We receive an event and we have to spawn an instance of your function and forward the traffic to it to execute your code, so it has a little little delay during which we have to wait from the function instance to be available, I'm going to show you that later on in the demo. So basically we can here see the buffer application.

C

That I showed you before in this diagram here on the bottom on the bottom, when there is no function instance available for an event, that's being processed, we are forwarding the this event to the buffer application and with buffer application, is exposing metrics about calls incoming events that are waiting for a function instance to be available. We have another component running inside the cluster alongside the buffer application, that's called the autoscaler, the autoscaler. His goal is to pull metrics and collected from the buffer, so the buffer receives an event. He says: I have an event.

C

That's waiting for a function instance to be, she become ready. The autoscaler will retrieve these metrics and he will ask the currency's API to create a new function. Instance to answer the the event and then the buffalo application is basically going to forward the event to the function instance and the first instance can send back finished in response to the client, so both the buffalo application and the autoscaler application are developed by a open source project wiki, which is called kinetic a native serving so cognitive is just like tectonic pipelines.

C

It's built on top of kubernetes, which means that we get to use system resources on top of kubernetes. We can deploy, specific kubernetes manifests and it will deploy and manage all the server is part of the system, for example, caving up and down your functions based on incoming workloads. So both of these of these applications are provided by the cognitive serving project.

C

So now that I have a function that in created you and I handle my event. We are going to see how do we manage the scaling of one instance to multiple instances of function available to answer the incoming workloads? So if I send multiple requests, each function instance gets can only handle one request at a time which means that if to read, HTTP requests are coming in. At the same time, the first request is going to block the second one until this handled properly and each function instance is also exposing different metrics.

C

So it's going to tell through the system, I have two concurrent calls at the same time, which should not happen. So the autoscaler is going to pull these metrics from the each function. Instance in the system he's going to realize that there's too many calls running on the same function, instances that are blocking each other and then he's going to ask the kubernetes api to scale up the number of available function instance. So these two calls are still blocked on the first function instance, but all following calls will be balanced between all available function instances.

C

So right now there is no construct, as there are enough instances available to answer the incoming traffic. So now we need to scale down its hope here is 4 a.m. in the morning. There is no call anymore for a function instances and how do we do it? So autoscaler components is going to pool metrics from each of our function instances and when a function instance is idle for more than 30 seconds, which means that it's not receiving any events to process then do to scanner will realize it and he will ask to kubernetes api.

C

She was came down the number of available function, instances until we reach stable state once again.

C

So during the service framework presentation, I talked about configuration that users can do about min scale and Max can so we make it possible for users to define a minimum number of function instances available at all time. The point of service is obviously to use a min scale of 0, which means that you save money when your function is not executed.

C

But if you don't want to use 2 to go through cost up to many times, it might be good for you to set the min scale of 1, which means that you get a container all always running and you can use the max scale property to prevent over bidding. For example, okay, so now that we know how to scale our function, instances in the cloud we'll need to know how to to execute the code inside it.

C

So we started to talk about the runtime code, with the HTTP server on the top container and the function code. So, basically inside the function instances there's going to be two containers. The first one we talked about earlier with the runtime, the runtime code, which ensure in the interns, invokes your function code with the incoming event to process, and the second container is called the Q proxy. So this is another components developed by the connective serving project. So this container has two purposes.

C

The first one is to manage the concurrent requests that are waiting to be handled by over your application code. So we, if we have two incoming calls, they are going to be stored inside memory inside the Q proxy. The Q proxy will then forward the traffic to the runtime code, which is going to invoke your function and once the runtime finish is handling the code, the cool, then the Q proxy will forward the next event and go and so on.

C

The other purpose of the Q proxy I showed before in the inside the scale part is function. Instance is exposing metrics about CPU and memory utilization and about we want requests going on inside the application. So we know that to do it. The Q proxy is going to do it. You're going to expose these metrics that autoscaler can scrape and as an end user you'll send your request to the Q proxy who is going to pass it to the runtime code.

C

So now we we went around every steps that are going through on the platform to take your function. Your function, codes, your applications and deploy it into production, scale it and make it always available, so the technologies that we use in order to fulfill this need multiple one, multiple projects from the skyway ecosystem. So the first one is that we are using kubernetes to run our users workloads to make it flexible.

C

So we are using the schedule, a capsule projects, the fully managed kubernetes infrastructure and in order to deploy the the docker images inside the steerer capsule, we also need to store this matches somewhere. So we are using the scary container registry in order to store your docker images when a user was to deploy an application, he needs to provide you upload the code of his application. So we are going to store this code inside its Carraway object.

C

Storage that we can then transform into a docker image that we can store inside the container registry and deploy kubernetes capture. So we are using cognitive, which is all the the brain of serve various parts of the system, and we are using invoice proxy and eto you for the networking part of the system. So now you might wonder, I want I need to do something more than just write function code.

C

My development in a team is already used to different technologies and we want to keep developing with, for example, Ruby on Rails, PHP or Python jungle, for example, and I don't want to be constrained by the runtimes all the functional service, so we are basically providing another service. Another feature inside the square wave servers platform which is container service, so everything that I described until this point is a is also the same. Four containers we only own.

C

The only difference is that we don't get to build your coat to wrap it with the runtime, because we only get to deploy your container that you're going to provide us and we are going to put the Q proxy in front of it.

C

That is the only difference so for this feature, you'll need to create a docker docker file, build a docker image with your image, with your application, store it inside those scary container registry, and then you can deploy it as is so now we're going to go through a little demonstration, so I can show you the different features of the platform, so both the functional service pots and the container service pot. So for this application here there is a little table- poetry that I posted here.

C

This application I wanted to show you how easy it is to take libraries that I've been developed for lambda in the Amazon lambdas platform, and then you can deploy it, as is on the several schedule platform, and also we're going to be using technologies like typescript.

C

Just to show you that you can use your favorite technologies to deploy your applications inside the severest platform, so this application is going to be about graph QL endpoints, so we are going to deploy a graphical API that I'm going to show you right now. This application is a really simple API.

C

We are going to list a list of dogs and the list of owners, so each owner has a first name or last name and the list of docs that he owns and each dog has a name and an owner that he's being owned by so here, I already deployed the application I'm also going to show you how to deploy it afterwards. It just takes between 15 seconds and 2 minutes. So, instead of waiting for it to happen, I'm going to show you straight and I'll show you at the end, how to deploy the application.

C

So here I have I'm going to deploy the same application two times. The first one is inside square wave server dysfunction. So with the functional service platform and the second one is going to be inside the container. With the specific doctor file that we are going to build and deploy it to the production, so here inside the scariest at console, I can go on the service tab. Here you can see the different projects that I created and I'm going to diplomatic patience into these different projects.

C

So each project has a dedicated private space inside the skyway container registry. Here so I'm going to use the meetup project like deployed for this presentation. So here I have the Apple lambda I deployed for the presentation. I can click on it and I have different informations about it.

C

For example, the memory utilization that I can set and the environment variables that I set to deploy my application and connect to the my sequel database so to store the different dogs and owners were using a my sequel database, I'm going to go through the code right next, so here I have one site diplomatic ation. This is care. Waste Savalas platform is going to provide me with an HTTP input that I can trigger to execute the code on my application.

C

So if I click on the test function is going to redirect me to my function, so if I click it hey, we can experience the core stats, which is when our system receives an HTTP request and it's going to allocates some space on the equipment across that you start to function, infants and for what the traffic to it. So it was about four to five seconds. I believe and here I have my application running, so my application, when I send the get request to it.

C

It sends me back an HTML file with the goth girl playground and then I can execute my graphical queries for my applications. Maybe I'm going to zoom a bit like this and I can modify the query. A graph QL query: query that I want to use so, for example, here I want to retrieve a list of terms with their owner with their first thing.

C

So if I can click on 'pray going to execute inside the clouds, so here with my HTTP trigger and it's going to execute and send me back the response of the API okay, so that's good I also have the the container that I deployed, which is exactly the same application I'm going to show you here. I have HTTP endpoint the application and, if I click on it, we are going to experience the core start once again, because the container needs to be allocated somewhere and then our graph QL playground is going to happen.

C

So here I have exactly the same application if I click on play, it's going to execute my code, connect to the my sequel database and display the data here.

C

Okay, so now that we we know what this application is all about, we're going to talk a bit more about the code and I'm going to show you how to deploy it. So the quote from this application is available publicly on github. So here we are using typescript two with Apollo Apollo is the graph QL framework for Nijs that allows us to do to develop graph trailer API.

C

We are using NCC, which is a nice CLI tool that allows us to transform typescript files into JavaScript files built on top of web pack. So it's really inspired from the golang CLI, so you can run NCC build and your project and it's going to build a javascript file from your project. We're going to use a my sequel database to store data about the dog and about the owners and we'll need the server less CLI installed to deploy this application with the server this framework.

C

Okay, so here I'm going to go through the code of this application, I have SRC folders I have two folders the demo, one which is a basic hello world application and the dogs application. So here it is I have two and three points so one for my functional service code and one for my container as a service application. So here I have container load type scripts. If I go into this application into this file, you will see that we are starting Apollo server on the port port environment variable.

C

We are using the type definitions and resolvers from the same module. I'm going to show you right after and here I'm, using the Appaloosa of our module, so open open source project, which is a policy about to start my container on the other side from the function point of view, I'm going to do exactly the same code exact, except that this time we are going to export a function. This is how the functional service works. You hex port a function handler that's going to be invoked by our our runtimes and this time.

C

Instead of using the Apollo server open source module, I'm going to use the Appaloosa lambda, which is an integration for Amazon lambda, so I did not modify anything I download the Opera server integration for Amazon, lambda and I'm going to deploy my application on scale way, as is without any modification.

C

The type definitions of this application, so we have an owner represented by an ID first name last name and the list of talk that he hones and a dog type represented by ID, a name and the owner. We have different queries available here, so we have dogs query to list the dogs get to dog bites, ID gets a list, owners get to know, know by its ID, and we can mutate the state of the application by either creating a dog and providing an owner, ID or DB to talk from the catalog okay.

C

So now we are going to deploy our function and our container so to deploy my container. My container I need to provide a Duke, a fight to build this container this code, this application into docker image that I want to deploy. So here inside the docker file, I'm using a two stage, build to optimize and make it more lightweight at the end. So the first stage is about building the application, so I'm using typescript. So I need to install the CLI tools like NCC to transform the die scripts into JavaScript files.

C

Then I will need to copy the package dot.

C

Json file install the dependencies copy, the code of the application inside the SRC folder here I'm, going to build so just like with the Guerlain CLI I'm, going to use the NCC build with an output of build directory and I'm going to build the container dot type script file inside inside the index, dot J's file inside build directory and inside my second stage, I'm going to install PM shoe, which is a process manager for no changes I'm going to copy from the build stage the build directory with the index dot JSP.

C

So it helps me getting a more lightweight at the application in the end and I can just stop the application like this and the same for functions in order to upload them to my platform, Savalas platform. Only to transform the code from 10 scripts into JavaScript, so I just wrote a little shell script that uses MCC to create a dog's directory and building the handler every file from the dogs territory inside the SRC folder. So I'm going to run the script once it's done. I can use the server list framework to deploy my applications.

C

So here I provided the server list of GML file, but I can use you describe the state of my application, so here I'm deploying a container that is called Apollo container, which is the one that you see inside my scary console here: I'm going to use the docker file provided inside the same directory as server so channel file and I'm going to provide different environment, I apples that are going to be stored encrypted inside our database.

C

We are using the scare way server. Let's begin to deploy our application to sevales care way and we are going to define two of other functions.

C

So the first one is the demo one which is a basic hello world application, and then there is the Apollo lambda that I showed you before, which starts the Apple log of current server, with the application and I'm using environment variables to define the credentials for database, and here you can see that I'm, not using the SRC, slash dog, slash handler, got a treat file because I built it I built the data file into word index to JavaScript inside the docs directory.

C

So I'm going to upload this file instead and I'm using the handler function, exposed exported from the entity ojf file so once I have this so I have it already on my local computer. You can see that I run the server little comment here. So it's going to go through multiple steps. The first one is to use credentials that you're going to provide from and for either environment variables or static configuration inside the solvolysis ml file, which is which you should not commit to your github repository.

C

So then, is going to package the service, which means that it's going to create a dot zip archive with your different files, your JavaScript files- in this case here it's going to create our updates, because I already created deployed the function before it's going to update my function here and then it's going to upload the zip archive into the private space inside scary, object, storage and it's going to trigger the deployments.

C

So the deployment means that we are going to ask the API to deploy a function, which means it's going to take the function, build it into a docker image, push it and deploy it for the the container part. We are going to build the docker image locally locally, so you need to have docker installed. I have a problem with my doctor installation. He does a lot of noise on my computer, so I'm not going to deploy a container right now.

C

I commented it from the service to channel so I can just use the command server list. Deploy is going to look into my environment variables to retain the credentials to my stereo accounts and it's going to take each my files and upload it to scare away.

C

So here you can see that uploading, the source code and it's deep when I. So if I go back to my scale back on so now, this is inside another nice place that I call meetup deployment. So it's here and you can see that I have my two applications that are being deployed right now, so it shall take between fifteen seconds and two minutes. We are going to see about it and once the deployments finishes it's going to display the HTTP endpoints that I can use to trigger my application.

C

So we are good going to wait a little bit for it to finish.

C

And in the mean, be able to figure out code on the on the platform.

C

Okay, so my Apple oolong definition is become as become ready, so I can click on it and test the function from here. So it's going to trigger the co start of the application and we are going to access the graphical playground and it's also going to display it in my terminal here. So I can also execute my demo application, which is the basic hello world and you see I, don't have any cost anymore once the function instance is available, I can just restart. I can send multiple HTTP requests and there will be a handled three-straight.

C

Okay, so we come to the end of this talk. Okay, thank you. So thank you for listening to us. So we are looking for new to new people to join the service team. We are looking for DevOps engineer and the DevOps engineer too, and if you only want to use the products you can go and register for the for the pizza for the beta for this project.

B

So, are you ready near? Do we and aunt went for the question? Yes, yes, we are. The first question was by Greg and you already answered it. We so what are capex and OPEX needs? Maybe you can you want to precise your answer together.

A

Yes sure well, Quebec stands for a capital expenditure and affects the operation it to meet you. So basically, the main difference is when you need to choose a server either you need to buy it or your. You need two rentals, that's the main difference.

B

How long does it take.

C

To see and to spawn the container and application.

C

Where is the question? It's.

B

On the the question tab, yes,.

C

B

And it's from Frederick and it's the question that had to a boat, okay,.

C

I can see, I can see it. I can.

B

C

From women attic sound and.

B

It's very creative. You.

C

Can you can you repeat this because.

B

I can respond.

C

A container so the the coast at is about between it depends on obviously the resources advocated through the application. So when you deployed the application, you can choose how much memory and CPUs you want to allocate to functions or containers, and if you, the more, you give obviously the the faster it's going to start, but usually it takes between three seconds and and eight seconds or a classic application.

B

Can I spoon several container instead of one with signal events.

C

No, you can, when you send an event, it's going to be routed to one instance of container, so.

D

C

Only use one container for an event if you want to either scale up the number of containers available, you have to send multiple events at the same time or you can use the min scale, obviously to set min scale of containers available for two, for example. So if you want to long running tasks in the background, then you can use this set this property.

C

B

There is another question about money in the case of container of the service, where I got a release database for my best persistence, how can I know the IP of the egress that will sent as a request to my database? Is it always the same? Do you have typical Latin C matrix for this kind of architecture.

C

Okay, so when you send requests from your application, I think I believe it's going to be the same IP, because I did not check it. We'll have to take a look at it, but I believe yes, because it's going to go through the the egress gateway for the cluster, so you'll have you'll be able to check the IP coming in and what was the last part of the question?

C

B

Was about the same, do you have any matrix? It's.

C

Gonna market, it you, okay, yes, there is going to be matrix, we already own them on our platform, but they are not exposed right now to our users. But yes, we have. We have this kinds of metrics, okay,.

B

And would you like to come with us on the scene, because you have several questions and I know that you want precise answers. So maybe if you want I invite you on the scene, if you can just let cask in the mic, and we will answer your question directly on the scene, maybe.

D

Not enable in my webcam for a reason, but I can always ask a question so I was I was mostly asking the previous question about the like starting different containers with a single event with different kind of containers. So, let's suppose that I have two tasks that are coming from two different containers: do I need to have them bundle inside a single container or do I have a way to manage, which kind of container will start up and down.

C

Yes, you will have to put the two applications inside the same container or you will have to send two triggers if you want to.

D

C

D

And it will make my question was mostly about security. So, let's suppose that I got my containers of service as a stateless like webserver and I got my persistence layer that is persisted on database-as-a-service. One sense by scale way in case I want to put firewalls to protect my database and be only accessed by myself, Alice application. Will the egress always stays the same for given name space of functions? Yes,.

C

D

It will stay the same to stay, persistent.

C

Sss, okay,.

B

Okay- and you do, would you like to go on with me? You have another question about my spirit. Oh.

D

Yes, so could you explain maybe for people that don't know exactly what it is, what is the M CPU and how do you merge your like a CPU time? Okay,.

C

So basically, inside K we are using kubernetes exposed resource allocation API. So when we deploy were clothes inside humanities, for example, users, containers or users functions, we can precisely say how much how we want to choose to allocate resources to this container. So we can say, for example, we can split CPU inside the in 1,000 parts and we can share this 1,000 parts per CPU core, obviously, between all the functions running on the same node.

C

So when you say, for example, hundred Melissa Peas, it means that you're getting one one tenth of a CPU for your applications is.

D

It always the same CPU, no matter which kind of instances are morning on the.

C

D

C

I don't think so, but for this platform we are using only the GP one L instances right now, so you're always going to get the same city. Okay. So we stay.

D

B

D

Indeed, because when the containers are coming up and down, there is like different kind of lakes of lots of time before the time where, as they go up and they go down in Cuba I test. Do you allow like and of customization of auto-scaling behavior right.

C

Now it is fixed at 30 seconds. So if your functional container is not receiving any events for more than 30 seconds, then we kill with scale it down, but it is configurable not exposed by our API, but it is very easy to reforest you to write the culture to allow users to customize it. So it's maybe an upcoming feature, so it might be an upcoming feature.

B

Okay, you asked about the price also, so maybe you want to talk about it. The network network goes between the objects originally.

A

Well, as long as your data transit between between surf skate away, different products- and they are fresh and it would cost- is free. Okay,.

D

That's a very good news, so, no matter how much pull and push I do from skin way to skin way. It's free! It's not it's not built! Yes,.

A

D

Long as you're.

A

As long as your new said regime, okay.

D

Does it account also didi bugs and also a product like if I got the debugs traffic's coming to the server or less? Is it visitor.

A

You asked me a tricky question: I'm, not a completely sure I need to try that okay.

B

And would you like to go on about it? Yeah.

D

So, do you confirm that, because server s is completely compatible with AWS lambdas, that the event and trigger objects that I will get in Python will be exactly the same with this exact same information that what I would get on AWS lambda? So.

C

This is specific to http events, so right.

D

C

Reporting HTTP events and cron based events are sending HTTP requests, so it means that you get the same structure defined inside the AWS documentation for HTTP events in the future. So we already implemented the MQ style of events, which means that you will be able to push messages inside the IOT products SKU, and it's going to trigger your application. So where we did, we did not follow the AWS lambdas standards for this one because it is related to the event broker.

C

So we went with the cloud events instead, which is a big open source project to define events, data and properties for the context, paths which is Amazon lambda is a really rich on the context, because you can trace the different where, from where your request, your event, is being sent. There are a lot more informations about identity management, for example, that is specific to the Amazon Amazon products. So we only focus on the function, run, sign name and the resources allocated to the function for for now, on. Okay,.

D

Make sense and finally, one question that is more related to my team and why having a different repository like a wave function, go instead of using scale where SDK go so, do you plan to having several less go function available from the SDK.

C

It's a bit specific: the square wave functions go repository that public on github is the runtime for Co, so the runtime is a bit tricky because we need to wrap the users code with the custom project you handle the events at you manage authentication if you decide to deploy a private function, for example, and we do not want to and we need to wrap this code with it so with scripting languages like nodejs like python like ruby, it is easy for us to do it because we only need to dynamically import your function from the handler that you configured and execute it, but with golang language with the compiled languages such as :, it's a bit different because we can not dynamically import the code and execute it.

C

We need to come to build this. This runtime wrapper. This is the runtime code, the HTTP server authentication. All of that should be a built inside the binary that you compile. So we have two open source libraries for compiled languages, so you can you use them inside your severest functions for cooling and it's going to rub your code into it. Okay,.

D

Now I was wondering how to tell to our user and to our customers. What is the difference between SDK into a scared wave function and whether or not they could use the same library for both from maybe we can add no distinction on well, thanks a lot, it was a very good way to cut presentation. So thanks a lot to both of you.

B

Do you have any question more, as you know,.

D

It's okay! Thank you very much. Okay,.

B

Thanks for that Amy- and we have just one last question- so maybe if you on the chat, you have another question or something that's not clear. Please do not hesitate it's a time. So one question from Alex: do you plan to add more triggers with object, storage, for example, so.

C

Yes, sir, right now we only have HTTP and Connor time-based events, and we are we added the the IOT team. I did MQTT part, which means that you can buy and qtt broker from the IOT products and when you push topics, messages into a topic from this queue is going to trigger your function, which is a great way to handle asynchronous events, and then we are planning on on building other event sources, for example, object storage, although we don't have a ETA for this feature.

B

Okay, I hope everything is clear and thanks a lot Louie and thanks a lot at 1:00 I. Think it's I, don't know what do you think in the chat? It was enough. Complete I think it's really interesting to have a demo part, a technical part and the marketing part. But if you don't think it's a good idea, please do not hesitate to say yes and thanks a lot.

B

We will be back next week on next Thursday, exactly the same hours to make this webinar in French as we are French and so do not hesitate if you want to to to go to another another with not the bit late, okay, okay, okay, interesting and and thanks a lot in the see you see you soon and have a nice weekend long weekend enjoy your a lot and.

A

Yes could like: we provide a recording, yeah.

B

Yeah yeah on YouTube, exactly thanks that we will provide the record just maybe in few days just on YouTube, as we hope you will have a long weekend that, yes, it will be on on YouTube, and yet thanks thanks.

A

A lot for listening give.

B

A bite well I. Thank you.

A