Cloud Native Computing Foundation KubeCon + CloudNativeCon Europe 2017 (Berlin), 11 Apr 2017

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Serverless on Kubernetes with Kubeless [A] - Sebastien Goasguen, Bitnami & Nguyen Anh-Tu, Skippbox

Description

Serverless on Kubernetes with Kubeless [A] - Sebastien Goasguen, Bitnami & Nguyen Anh-Tu, Skippbox

Serverless is getting lots of attention lately. It is positioned as the next evolution for building distributed applications, going beyond container based systems and letting developers build applications workflows based on triggers and events. The three main public clouds (GCP, Azure and AWS) all have serverless offerings. In this talk we will introduce kubeless, a serverless framework built on top of Kubernetes. It allows Kubernetes users to define functions that are dynamically injected in container runtimes and exposed via HTTP or event triggers. Events are managed by Apache Kafka while HTTP triggers are exposed with Kubernetes services. It is an open source clone of Google cloud functions, and provide similar capabilities than AWS lambda. We will explain kubeless architecture, show how we leverage ThirdPartyResources and an in-cluster controller. Above the interest of serverless, this architecture shows the power of Kubernetes and how it can be used as a platform to build new systems quickly.

About Sebastien Goasguen
Sebastien Goasguen is a twenty year open source veteran. A member of the Apache SoftwareFoundation, he worked on Apache CloudStack and Libcloud for several years before diving into the container world. He was the founder of Skippbox, a Kubernetes startup that develops open source tools for Kubernetes users and is now the Senior Director of Cloud Technologies at Bitnami. An avid blogger he enjoys spreading the word about new cutting edge technologies and also trains developers and sysadmins on all things Docker and Kubernetes. Sebastien is the author of the O’Reilly Docker Cookbook and 60 Recipes for Apache CloudStack.

About Nguyen Anh Tu
Nguyen Anh Tu (Tuna) - Software engineer @ Bitnami. A Distributed System enthusiast. Current interests include containerization, microservices, PaaS, serverless. Favorite programming language: Go.
Join us for KubeCon + CloudNativeCon in Barcelona May 20 - 23, Shanghai June 24 - 26, and San Diego November 18 - 21! Learn more at https://kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy and all of the other CNCF-hosted projects.

A

So my name is Sebastian grass can said going on Twitter and today I'm going to talk about server less on Cuba natives with the software that we just wrote a couple months ago, called cubelets and I'm very excited about it because we're starting to be able to do really cool things with it and go beyond. You know, containers and and actually privileges itself, which is you know very exciting. So the software is on the Geeta bitNami cublas, so you know check it out. Definitely so I used to I founded a company called skip box.

A

We actually sponsored the first cube con in the US and we did things like compose with a kid that you find in the incubator. Things like cube, watch, K machine and cabin the the iPhone Android app for for communities. But three weeks ago we joined the business, which is a company that does app delivery application packaging.

A

For for the cloud, we also create containers and so on, and you know the the connection between the two companies you know makes total sense to us, because we can, you know, help people go from bare metal, VM, Cloud containers, helm, charts and then potentially serverless if it picks up.

A

So that's why it's it's quite exciting, so I wanna I want to spend a you know a little bit of time talking about server less the named the buzz and I'm going to try to go relatively quickly through this, but I think it's very important to to get the context and then I want to do a demo live demo. So I have a few demos and we'll see how that goes there.

A

You know we still have a few bugs, so I may have to kick a few things to get things working, but we'll we'll see how that goes. So you know the first message. You know why why server less on kubernetes and it's not because you know I want the room to be to be full because it's a it's a buzzy word and I could have added several IceCube analysis.

A

Machine learning, you know something like this to to make a big buzz, but you know forget about the bus and and the the message here: don't dismiss serverless, because it was very easy in 2006 when I was working on a big project in the US when, when I announced ec2- and you know a friend of mine called me up- and he said: hey did you see, Amazon just released ec2 and I was like I'm not going to run virtual machines on on Amazon.

A

That's you know, that's crazy, so it was very easy to dismiss the the beginning of beginning of the cloud, and now we you know we see how in ten years you know things things have changed and everything we can do in terms of Ikea IT in the cloud. So you know, let's not dismiss a new tech just because it's a it's a buzzword and people are going a little bit crazy on the internet.

A

Marketing is going a little bit crazy and so on and and also you know, I see a lot of discussion about several apps. You know what does it mean and so on, and we see this all the time when there is a new tech when there is a new word, you know people like well, you know it means this. It means that well, let's, let's not get hung up on the name. Okay and that's that's from Twitter. You know couple months ago at server less Austin.

A

What is server less, it's just somebody else, fully managed execution environment and I only pay a fraction of a cent when my function is run. Okay, so of course there is a server there's a process running at the end of the day on the server and the idea with server less is that you only paying you know a small fraction when the function is being called.

A

So, let's not get hung up on, let's not dismiss server less, because it's buzzy and let's not get hung up on the name and I'm, not even going to mention fast. You know function as a as a service and also, let's not mind the fact that you know it's a hype and we know this hype cycle and then server less is on the on the trigger. So everybody is going a little bit crazy. But, let's you know: let's keep the hype.

A

Let's, let's be open because you know even when docker started, it was very easy for us. You know IT professionals with many years. You know in the in the industry to say what containers have existed for a long time. You know we've had that in Solaris we have alexey, you know what is this docker thing? So if you, if you do this, you know you're going to miss on a potential adoption curve. Okay and that's from Simon worldly I totally stole that from him, and he said: that's the danger where you know in in red.

A

You have the enterprise curve where the enterprise guys say: I know where nowhere. You know dis missing the new tech and then you know you see that the rest of the world is adopting it and I like I said no. No, ah maybe ah you know it's too late, okay, so the risk of this missing. All of this is just that. You know this.

A

This new tech is going to be adopted and then we're going to be left left in the dust now you know for me, so I started looking at server less because of some talks at cube con. You know from Kelsey and from Brandon who are trying to we're talking about compiling, to cube writing an application locally on your laptop and then you, when you to run your executable. You know it automatically distributes the the various functions in incubators.

A

So I was interested by as I said that we could do a framework on top of Cuban ages. You know quite at least get a proof-of-concept quite quickly, but basically to me you know server less. It's about missing services together, so we're going to deploy services in our communities clusters, maybe via helm, so we're going to deploy an object store. Maybe we're going to deploy a MongoDB.

A

You know you name it we're going to have all those services deployed in Cuba natives and now you need to stitch them together and you need to glue them together. So for me it's just you know, knitting and deploying those little logical. You know units in in your communities cluster. You know the other. The other analogy is like gluing Legos, so Cuban it with Cuban Isis. We can. We can deploy apps micro services and then you know we talk about all those Legos being assembled to maximize utilization in the cluster.

A

So you know with functions with the server less or function as a service. You, you know, you choose your poison. You know we're talking about finding a way to glue all those services together. Okay, at least that's you know, that's that's my view of it now, of course, server less you know imagine lambda is is leading already there they've made a def made a big push, so in some sense, Google cloud function is already behind. You know AWS is you know it's quite far ahead there?

A

But when you look at the examples that you know you you can do with lambda, you got relatively simple things like a thumbnail creation pipeline, so you drop files in s3, automatically create a thumbnail and puts it back in in another bucket. All you get data streams pipeline. You know you listen to a stream on on kinases and then, when there are events in that stream, you trigger functions and then maybe you insert some something in a database.

A

So you know those are the types of things that we can do with with lambda, and you see that you have an s3 service or you are the kenosis service at DynamoDB service and the function deployment is about deploying this logical unit to stitch everything together and create a more interesting application. So what happens? If you know now we're deploying those services. We have communities to be able to do this, but now we want to deploy those functions to create much more interesting pipelines and complete applications.

A

We need a way to be able to do this if we don't have server less. That means that you know we need to write code and then go through. You know we go through CI, we generate another container, we launch a new deployment and you know up with everything that gets connected, but with lambda we can do this, you know relatively easily, so it's basically you know deploying functions if this than that type thing, there are lots of solutions out there already.

A

So you got lambda cloud functions on Google as your functions open with function, you know initiated some guys from JBoss web tasks. I Renee, oh so you got. You know you got already some some framework, some solutions, but none of them are truly kubernetes native and what I mean by that is a you know: a function framework function as a service framework. That's going to reuse all the primitives from cumulative, so deployment services, third-party resources. You know a you know, all those all those great building blocks that we have in the communities API.

A

So with cubelets. What we wanted to do is reuse all the base, primitives incubi natives and allow us to build a framework so that we deploy functions. Okay, make sense.

A

You know and I cannot do this talk without you know pointing to Simon Wadley again that he wrote a very interesting blog. Why the fuss about server less I'm not going to go deep into this, because I want to concentrate on the demo and try to you know, show it in action, but Simon writes very interesting things about you, know, strategy and I key and then evolution of IP and how it affects the way we do things. So you know, look it up on Twitter and then Chinese blog.

A

Why the first about server less it's a very interesting read and yes, I totally drank the kool-aid I like I like what he says about server less it makes you know it makes it makes total sense to me the way the way Simon looks at this is that it builds map about value in your company and then how the technology evolves, and it makes the point that you know new new concepts, new ideas in technology.

A

They go from descriptive, you know concept and then they become a utility and when something becomes a utility, you know especially like platform infra and so on. It changes the way we we develop and we write applications and we the way we practice software and the way we do our business okay and if, if a practice changes, then it may move the where you have value in your application or where you have value in your company.

A

So you need to be very aware of this and very aware of you know, adoption of new tech and how it's going to impact you. So here we have, you know, we've had the end the cloud. Now we have something with communities to deploy services. You know, containers what, if server less, really becomes adopted in the industry.

A

How is that going to impact the way that we manage infrastructure and the way that we develop applications so that you know that's the real question to me and, and you know, Simon talks about the coevolution, so coevolution of you know the basically the platform, the infrastructure and the practice of writing apps. So if, if functionalists takes over, you know, I mean a server less, not function based server that takes over. That is going to change our practice of software okay and it talks about workplace development.

A

Where now you have an application where all the logic is just a you know function, you can monitor all those functions, so you know where the cost is, and you also know where you're making money. So now you can start developing trying to optimize, for you know, lower cost and higher value.

A

So that's a little bit fast there, but the takeaway is, you know, check out check out what he's writing about IT strategy and- and you know it informs a lot about why server less may be impacting the way we develop. Software, okay and I and I three I truly believe believe this so now surveillance on communities and why? Why cube less so I mentioned last Keep Calm, so there was I, mean Craig McKee from estée. Oh, he mentioned something about server less, that he was looking forward to to see a sterilised framework abilities.

A

They are existing frameworks, but I believe that none of them are truly kubernetes native, and then there was also this talk by by Brendan burns who its own is github I mean he's developed a system where you can basically some type of language bindings where you can write an app and say hey. This is a service you know distributed on communities and so on. So this talk reminded me a lot about what I used to do, which is parallel computing, and there was a time in parallel computing where we had Auto paralyzers.

A

So ever, if you guys remember some of those tools, but you had serial code and then you would fit it to an auto paralyzer and then it will run it in parallel in a cluster okay. So you know what, if we were writing an app which would be you know, most a monolithic app and then we would automatically distribute it in a creative cluster and I. Think that would be a that would be kind of a kind of kind of interesting but cumulative.

A

You know it's rewrite of Borg Google cloud functions runs on board, so we should have a sort of less framework on top of communities. The number two is that so I've been I've been using cue, bility snuff, for quite a quite some time. You know maybe two years or something but grant is not in production. Okay, so I'll raise my rent, I don't run cumulative in production, but the thing you know, even though I develop software for communities and I.

A

Do you know a lot of testing lots of you know developing new ideas and so on. The one thing that is changing in my mind is that I don't care about the infrastructure? Okay, I'm, just talking to an API I'm, just talking to the cube, API and I just want to use all the API primitives of communities, okay and then the other thing that's happening lately is that I actually don't care about the containers?

A

I, don't care about the runtime, you know container the rocket whatever I, actually don't care about the runtime, and actually you know the actual containers that are running I. Don't care much about the containers themselves. Okay, what I'm interested in is deploying services and then building some interesting pipelines. So what this is bringing to me is that I'm actually going back to focusing on application. It's applications instead of infrastructure, and you know I mean networking and you know the the actual system underneath. So it brings back the focus on on applications.

A

So that's that you know, that's that's what what what's happening and why you know Kuban Isis, makes a lot of sense for me to use as a as a platform, to de put you to create a framework for for service. So the let's talk about the cublas architecture before going into the demo, so it's on github, so you know have a look at it. Definitely it's still rough okay and it's definitely usable. We use it everyday now, but it's still a bit rough.

A

So if you try it, you know ping us on github file an issue and then well, you know we'll help you out, you know actively so I mentioned Cuba native native, so a tube con I was like hey, you know we should be able to do this. You know quite easily. We have third-party resources to extend communities. It's a great mechanism, let's just create a third-party resource that defines the function now we're going to have a cumulative endpoint, which is functions and we can start create. You know, functions in that custom, endpoint, okay, so cube.

A

Let's use this third-party resource, it gives us an endpoint with you know, rest rest endpoint for functions, but you know it. The third-party resource in itself doesn't give you anything. Now. You need to write a controller, so we have a cublas controller running in the cluster that watches this custom endpoint and when there is a function being created or deployed, you know, then the controller does something. What does the controller? Do? The controller creates a deployment and a service on top of it. So it's all you know Cuba native primitives, now the actual function.

A

How does the actual bit of code? How does the bit of code get inside the container? We actually stick the the code in a config map. Okay, so you create the function with the search party resource. It gets a config map gets created and that allows us to inject the function with a volume mount inside the container. Okay, we're not worrying yet about the the start time. Okay, we're going to that's going to be the next phase to optimize the the startup time to get to get it going.

A

We did a little bit of a hack, okay, the big problem when you start going with server less is what about the dependencies of your function, the basic ello world. That's extremely easy! Okay, it's just you know, there's no dependencies, but what happens when you have? You know loads of dependencies, so we didn't have to confess a little bit of a hack, so the deployment that the controller create uses, an image container and the image container just loads, all the dependencies yeah.

A

You know you know in a volume, that's shared with the actual runtime a little bit of a hack, but it works great. Okay.

A

Now you know a lot of the lot of the frameworks that that you did have that that you find out there. They are mostly looking at HTTP triggers so deploying functions that you can trigger with just an HTTP call. It's pretty much like you know, an easy way to deploy web hooks. What we wanted to do here is go straight at the events mechanism, because when you deploy all those services, what you want to do is deploy those logic that are, you know, triggered on events.

A

So how can we manage events in the system like this? Well, when you install cublas, we also run Casca, okay, so it's deployed in communities. You have the controller and your Casca, and now we can start feeding all our events in our systems through you know various topics in our you know: Katka Casca broker and the functions will be triggered on events, and it's all you know, bundled in in the in the system. Why Casca?

A

Why? Kefka? Honestly? Because I was you know, pushing cublas with a an example with the s3, the mini o s3 object store and it you know it publishes events quite easily to to Casca. We could you know we could use something else if you guys are totally allergic to Casca, which we could use a different type of message. Bus now, you know what happens when you do cube, let's install it creates.

A

The third-party resource is, you know by default, creates a cube last name space, it launches the controller and then it also launch it launches Casca. So it's very much like helm, you do helm in it here. You do cube, let's install and you know everything gets gets created now you end up with a third party resource for functions, and then you have the share like the cube left CLI. So the the main the main function is cube. Let's deploy that allows you to create functions.

A

The CLI is currently is totally compliant with the Google Cloud function CLI. So if you use Google Cloud function, you know cube less gives you exactly the same type of interaction. The API is not compatible because the API is, you know, just a rest and point of the third party resource, but the CLI is actually compatible with Google class function, so you deploy the function. If it's just an HTTP trigger, it gets injected into a pod, and there is an HTTP wrapper so that you can make HTTP calls for events.

A

We have a little bit of. We have a convenience wrapper with cube blast where you can talk to Kafka and you can create topics automatically, so you can do cublas topic create so you can manage. You know, manage your topics and then you know same idea. You deploy your function, but you say that it's triggered on this particular topic and then you know it gets. It gets deployed so time for demo.

A

Our gauge should be function deployed. We change that so starting a function going to look like this cube last function deploy name of the function here. It was get I. Didn't put. You know, I just put the flag, so you here you specify that it's triggered by HTTP trigger HTTP. If it's supposed to be triggered by events, it's going to be triggered topic runtime. We support Python, nodejs and soon Ruby from file. That's where you have your code and dependencies.

A

You know for Python, it's like a requirements file or if it's node, it's going to be a package that that reason. Okay, so demo.

A

That big enough in the back, yeah okay. So if I first look at you know all my pods in my in my cluster you'll, see that I have a cube last name space and in it I have the controller that's running plus Casca. This is definitely not a production deployment for Casca. You know it's a single pod and then you know you have one controller for zookeeper. One control Gasca- and you know we with for production, we would need to you know, come up with a proper Casca cascade deployment.

A

But now you have you know if you look at the third-party resource.

A

You see that we have a third-party resource which is function. That Kay does that IO, which means that now we could potentially I mean we can do get functions, you know and, and you get it okay, so we have a function. End point in our enough cluster. So now you have the cube, let's see Li, so you can do you know, function LS, so right now, I have no functions and then, if you want to manage topics, you can do topic. Ls talks to kefka. Maybe are you yeah?

A

That's a pretty in a very basic wrapper here. So I have one one topic here: that's s3, okay, so I'm just going to do a first example, which is an HTTP trigger example, I'm going to deploy a function that sends a tweet okay, it's kind of funny, but I mean I. Think it's totally useless. It's funny and the example here is in in Python I'm, going to explain a little bit, because this this looks long. The actual function is this three lines at the end?

A

Okay, so it gets the payload that will come from the HTTP call, and then you know API, post and I'm just sending the tweet okay. Now, what's what's at the top? You see that there's a dependency on a Twitter client and then I'm also using the the Python client for communities, okay check it out in the incubator. It's quite it's quite good.

A

The reason I'm using this is because I wanted to be able to load secrets, because if I want to talk to the Twitter, API I need the keys to my my Twitter API, so I actually stored all the Twitter stuff keys into a community secret and then with this bit of you know, dirty Python I can actually start to communities API and retrieve the secret directly in my function. Now I think! That's that's quite you know, that's quite handy. So if I look at get secrets have a few secrets there.

A

I have secrets for me know: I, have secrets for slack, that's my flag, token and secrets for for Twitter. Okay. So now, let's I'm going to cheat a little bit here and look at yeah just copy paste. Yes, so that's the actual deployment of the function function deploy a name trigger runtime, Python handler, which is the name of this file and then the name of the function, it's just like Google, Cloud function and then the requirements. So let's copy this and now what what happens is that hopefully it works.

A

So we have function now we have a function which is HTT type HTTP and now, if I look at my pods, hopefully yeah. Thank you. I have a thought that was just created by my controller.

A

You see that currently it's running the in each container, because I need to be able to install the Twitter API and then once that that run I'll be able to be able to to curl my function so I'm just going to I'm just going to just do this, which is my curl and in another window, I'm going to okay I, have the proxy running I can do it on gke, but I thought it was a very bad idea, because then you know the func I can expose this function within ingress.

A

And if everybody could, you know, Carol that and then tweet on. My account Dyess does not. It was kind of funny, but.

A

There you go so now that sab it slow, because it needs to install the dependency okay. Now, of course, if we want to update the function, we can do this just by editing the config map, that's kind of handy. We edit the config map. Sorry.

A

Can I add another function yeah, so you could edit you could edit the the config map and another function, but they would need it. We need to be called by the first one, but then there is probably in communities that when you update a config map, you know the part that none that config map don't get redeployed. So we need to. We need to work on that. So, let's, let's tweet this well okay, I'm just going to put this too to save time here, let's tweet this so I'm curling, my my function and where's.

A

My uh here, if I go to my Twitter, did I tweet this I.

A

Did yeah fifteen seconds ago this rocks from Cuba great No.

A

So that that's the basic, that's the basic HTTP trigger function, okay, deploy it on the fly of course. Sending tweet is not, you know, really easy, but you can imagine building lots of web hooks and I was talking to someone with like yeah. What I want to do is just be able to trigger a rolling update by calling a function, and that would be super easy to write a little function in Python that actually updates a deployment. And then you have your CI pipeline calling that function, and then you know triggering the rolling updates automatically.

A

So that's great! So now, let's do something a little bit more interesting. So what I've done here is that you know you mentioned. I mentioned stitching services together. So what have? What I've done before is that I've deployed mini ODS, three clone the object store of the project with helm, okay, so helm, release I have mean you're, running and I have mini running so I have mean you're running here.

A

So that's the web interface to to me know it's been configured so that when when something happens on a on a bucket, it sends an event to two Kafka, so I'm going to create a bucket called fubar yeah I'm, going to create a bucket called fubar and I'm going to deploy a function that you know, watches files being put in that bucket and every time there is a new file there.

A

It's going to send me a message on slack sounds sound good, so I go to my slack example, and this is on the all of this is on the the github repo. So the the function you know is same idea. You got like this. This secret loading in Python. You have access to the menial service.

A

Okay, with the you know, it's minyak is exposed by a service. So this is you know standard, and then you see the function here at the bottom, which is just saying. Okay, if I receive the put event from me know, you know, extract the bucket name, the file name and then send a message to slack. Okay, so I'm just going to send this okay, this.

A

So cube last function. I can delete functions with the the CLI in Amite and you know so on so the function is there: did it create a pod? Yes, and again you see that the pod is being is being created. It's a different runtime because now that runtime by default as a cask, a consumer okay, so the function is basically injected by a config map and we get it dynamically with like loading, the module and and so on.

A

Maybe maybe running you want to improve my Python, because it's it's pretty bad, so we just need we just. We need to to wait. You know a little bit so that it it installs the slack client.

A

So we've discussed internally a new new architecture to make this. You know a little bit faster so that we can create a you know, an image for the runtime with the the dependencies loaded so that when we, when we start function and when we scale them, you know they're going to start really quickly right now, you could scale, but every time you everybody in the in your scaling, we'd need to run the unit container so that you know that's not super optimal, but we we have. We have a solution for for this.

A

Also one thing: I'm not I'm, not going to have time to demo it, but the HTTP runtime is currently instrumented with primitives, so we actually serve primitives metrics for the runtime, so you go to primitives. You see the front. You can monitor the function, calls okay, so we're going to put also the the primitives monitoring into the event system, so it will be a it will be fully fully monitored and then the idea is that with primitives we can use custom metrics for the original product of scalar, and now we have our functions scaling.

A

You know if there are lots of calls. So, let's see if this works, so this is my object, store I'm going to drop a file in there, meaning okay, whatever that, and then that should send me a slackness and I think what's happening is that when I, when I mean view like this, when I mean view somehow I don't get the notification.

A

A

The other man, both Nick bakas name, is wrong.

A

Fubar, okay, fubar. Let's try again a different.

A

So that that's a little, you know little alerting mechanism here. Thank you for catching that it's that's handy. Okay, now, I want to do something a little bit more interesting, okay, which is the thumbnail creator. This is out of the box out of fresh. You know from from yesterday, so we're going to see if this works.

A

Let's do so its resize, it's one of the example for AWS lambda. So let's look at the resize and it's Python because you know I mostly know Python and I'm, actually more than she said min and a developer. So you know this is bad, but please, you know help me improve it. It's using a Python module which is a pillow, and here you see the the image thumbnail creation. It's very dirty. I'm, listening to me know I'm, creating a temporary file inside the container fetching the file.

A

That's the F, get object, creating the thumbnail and then putting the thumbnail into a different folder that doesn't make sense. So I'm going to go to my I'm going to go to my menu and what I need to do is create another bucket called Sun and I need to start. I need to create the function so where's. My function here so function: okay, I'm going to change its name because I, sometimes it's a little bit. We have. We have a tiny bug in there.

A

Okay, so I've created a new function, function LS. Let's try to watch the pods okay, it's creating, so we have to wait a little bit that it gets created, we'll put minyul here, I'm going to kill that kill that kill that keep the slack open so now notice that I still have the other functions running and those of you who know Casca a little bit is that each function gets in a different consumer group.

A

So I'm going to deploy this new function, it's going to be cold, but the slight one is also going to be cold. Okay, so messages get into different consumer group.

A

So we just need to.

A

Yeah the dependency handling, if once we once, we figure out the build system to actually preload the dependencies and then use those images. You know the this cold start up. Time will get much faster.

A

hmm Tell it again yeah.

A

Now so the images are already there, but what's what's happening is that you know it's it's it's it's trying to do a pip install of the the pillow inside that beneath container and right now we're not we're not reusing the the end that the image- that's the other DM but yeah that can be improved. Okay, I'm, going to I'm just going to finish the slides and then just before, I'll come back to it. So the the road map here is that we have to improve our run times.

A

I mean it right now, it's you know, it's still very rough, so we need to. We need to improve things. You know, in the run time, better dependency management, that's what I I'm mentioning with this this in each container. Currently it works well, but it needs to be a to be improved. We also need authorization for authorization for functions that definitely something like feedback that we've had from from several folks, because otherwise anybody can call your functions. So we're experimenting with things like 0 to be able to add to have us 0 authorization directly.

A

In the run times, we will see how that goes. We could automatically create ingress resources which would give us some type of API gateway, but it doesn't. You know it doesn't make sense all the time so it may be. It may be a flag, an additional flag here, to create the ingress instrument, the runtime for event, so that we have the proper primitives monitoring which gives us the auto scaling. I mentioned that and then look at a lot of services that we can deploy with helm.

A

For example, and bitna me has been, you know, doing lots of work on the chart. So that's where we're going to look at all those charts and figure out the ones that we can easily. You know either repackage or or use that are going to emit events. Okay, we need services that emits event and Casca is great for this, because there are lots of cascade plugins that can help us.

A

You know like, for example, you have my sequel and then every time you do an insert in the database sends a message which would trigger a function.

A

So they that's the the services by helm and the service broker is going to be interesting in all of this, because, with the service broker running now we can potentially bring in services from public clouds, so we're going to have on Prem services deployed, as you know, helm, charts, and then we can bring in external via the service catalog and have those lat that logical glue with you know cublas, so that that's you know, that's the main idea right now.

A

So, let's see, if you know, I got lucky and I know I shouldn't have pushed it. Didn't like my okay. Well too bad. So you have to believe me that the thumbnail creation works I'm not going to try to debug this. But you know, please give you know, give cubelets a try. You know give us give us some feedback.

A

Let's say you know, and let's see, if you know we can, we can push this further and and make it you know even more interesting, so don't dismiss server less just because it's buzzy, you know, few blacks is Cuban. It is native I think that's the real strength of the system. We're not writing our own API server. You know Reuters things like this. We're just reusing everything, starting with third-party resources to expand kubernetes and it looks just like Google Cloud functions.

A

We have lots of ideas to be able to to take existing lambda functions and make them run in cublas. So we'll see how that how that goes, but yeah. Definitely, you know give it a try and give us some feedback that you know. That would be perfect. Thank you.