Cloud Native Computing Foundation Runtime Special Interest Group, 6 Aug 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: CNCF SIG Runtime 2020-08-06

Description

CNCF SIG Runtime 2020-08-06

A

A

A

Good morning, how's everyone.

B

Hello good, thank you. Yeah.

A

Let's give it a few more minutes, uh hopefully we got a little bit more.

A

A

A

A

A

So both of the I posted the meeting notes on the chat. So if, uh if you can just add yourself as an.

A

A

All right we we got like uh six people, so hopefully some other people will join laters us. So we got uh tom and jen luca. uh You guys are gonna present the packet, uh sorry you're from packet and you're gonna be presenting tinkerbell, so yeah just take it away.

C

I'm unmuted now hey thanks for having us. We are uh here to talk about uh tinkerbell, it's an open source project that we're building at packet and trying to engage the community in that takes care of all the automating and management of bare metal. Servers takes care of provisioning, and I think john luca is going to go a little bit more in depth and show off some some of the cool power that tinkerbell brings.

B

Yeah I mean it's yeah, it's it's mainly like uh you know. I I joined packets like a few months a few months ago, and it was mainly you know. My thinking was like when you get a server. What do you do with it when it comes?

B

You know you, like you, take it out from from the blogs, and how do you make it do something useful and yeah tinkerbell tries to to answer this question uh and uh giving to the user the flexibility to uh to do whatever they want with the in terms of like installing the operating system and running actions um from having like a server that doesn't do anything until it.

B

It runs like a cloud unit that we all know coming from, like you know, ubuntu or aws or any other cloud provider. So the idea is to is to close the gap between like a new, fresh server that you get from a shop and like the cloud, let's say um the way we do it like is via api, because that's what that's? What we learn like in the in our cloud native, like journey and um yeah.

C

B

My screen and see like if you have any, I know if you had a look at it and you have any questions you can just stop me. Otherwise, I I'm trying to. I will start from the documentation and share a little bit what we we did so.

B

Far so, as tom said, it's an open source project, so it it has its its home on github and it's made of like microservices or services called them like. If you want, and the pink one is the main one and it contains like it, provides the the cli and the thing server. The thing server is the equivalent of like a control plan um in in the kubernetes land. It's the it's the after that receive their casts, store them and you know, serve them to the worker and the worker are like your server.

B

So usually, when you have a worker it boots, when you have a server like you, do you power it? You power it on and it starts, and the boot starts. uh Nowadays all the servers supports like network booting, that is the technology, the is pixi and we use. We leverage that as uh to serve our base operating system. uh The base operating system we serve via netboot via ipxc uh is called aussie, and is uh it is like a variable open source as a project as well.

B

So the first stuff you you see when you start like a worker uh is like aussie itself. I don't find it here yeah uh and it is an in-memory operating system based on alpine, so you like the server goes on and it makes like a dhcp or guest asking for you know, appear and tinkerbell serves um the http servers. It is called booth and that is available, as well as a open source project and what booth does it? It responds with the aussie operating system.

B

So in this way the operating system starts on ram and it's available. So it gives you a shell that you can use uh inside oz. There is docker that we use docker has a runtime and, as you can see the documentation in practice, what you do is like you, specify uh a template and you transform the template into a workflow.

B

So the ten the plan the template is aml, is in the ml format here and, as you can see, it recalls a little bit like docker itself because it is used like the image. So what we are saying here is like worker, with the mac address, uh executes the the workflow from the template.

B

This template this template here and you know.

A

So yeah can you make the screen a little bit larger yeah, and then I have a question. So uh do you have a like an architecture diagram somewhere in the document? Okay, we have it. I just remember where it is. I didn't want to break your workflow.

B

But uh you know no.

A

B

Fine, thank you um yeah. So, as you can see like this.

C

Is the the architecture.

B

So there is the control server, that is the provisioner itself and it's the think server. So the thing servers provide grpc and http api and it contains, like the server, contains all the workflows. It stores all the hardware representation. So you can register your hardware inside the server and you start in place vlc li uh when, when the worker starts, as as I said like it takes, it, makes the dhcp request it, the dhcp responds and it starts uh via ipxc the aussie operating system.

B

It is an alpine inside alpine, there is docker and from there uh the worker is able to execute every task that you ask it for one of the one of the tasks we have is the hello world, one that we saw in the documentation, but it's not super like fun. The another one we have is called tinkerbell ubuntu and in practice, brings you to a fully persisted and working uh ubuntu operating system um and, as you can see like you, have the ubuntu template.

B

That is the same that we saw before it's a ml file that describes uh all the steps and the actions that has to be done. So the first stuff you have to do when you install a new operating system inside inside a server is to wipe the disk.

B

So there is an action that like uh wipe disk, there is another one that makes the partition because we have to partition disks and like set up the the swap or the home directory and where the operating system is so, um you have to install the root of file system and configure the grab and start the clouding.

B

That's almost what happens when you install like every operating system, and if you want to know what, like each of those action, does you can follow the foldering and as you as I will show you like? Every every uh action is a docker container and the locker container can be as complicated or simple as you want. So the wipe action is uh executing a batch scripts and the bash script is way more complicated, but what it does like it's it wiped the disks.

B

So it erased all the data and it prefers um that they fit bootloader for the operating system itself.

B

And I mean this is like in the high level like very high level, how um how tinkerbell has all project works? And if you have a look at the repositories we have, you can figure out how we kind of like split the responsibility of the project itself, because if the ide, you can use a different like in-memory operating system. If we have your own one, so you don't need to be. You need to to use oz you you can use your own one. Definitely.

C

B

It has its own repository with its own we're, building, release life cycle for every project, and so on. The the higgle uh repository is it serves metadata so when, if you're used to, we are all used to cloud computing, so we know that we can call like an ip from inside the server and it responds without the metadata of the machine itself, and this is what we do uh with that microservice um and it is available for every like machine.

B

You start with think about boots as a is the is the first uh interaction like a server has, and it is a dhtp server. We leverage like, as I said, like pc fixable, so we're doing that booting and that's how it works. You name, the machine starts, it asks for an address and it gets the address and an operating system a temporary operating system.

B

We also have like microservice a microservice that help us to uh manage bmc to interact with with bmc, uh so we can like switch on and off like machine programmatically.

B

We are, we started to build a ui, a graphical ui, um so there is a portal and yeah. We are collecting workflows now and we are trying to to figure out how to make them reusable uh in a good way. Luckily, for us, we decided to use like oci and docker images. So technically we can use like.

A

B

A

Question so the oc is just an in-memory operating system, but that that's uh not the operating system that is running on the machine right. So that's no.

B

Yeah that you're right, we use that operating system only like as a first way to run actions on the server.

B

So as soon as you uh as soon as the as soon as tinkerbell workflow inks wipe the disk install the ubuntu operating system, uh we configure the grab, so the the grab will uh switch the booth from the networking one to the disk one. So from there. uh You start from your disk that has ubuntu or like debian or uh um whatever and like oz is not used anymore because it was running on ram, so it just doesn't exist anymore.

A

Got it, I are you integrated with kubernetes in any way, so, okay.

B

We got like a lot of requests from like the cluster api team to get an implementation uh with it, but right now we are working on like the tinkerbell core itself, so I think, for the next couple of months we will uh keep working on like a release, life cycle and stability work that we have to do, but the cluster api implementation is for sure uh a priority for.

A

B

But we hope that with the community to build one um very soon.

B

And yeah, I think I don't have a lot to say more time. I don't know if you have any partners for us.

C

Just that we're really excited about this project and we're looking for more people to get involved. If you all have any questions, uh feel free to ask now or you can reach out to us, both of our emails are in the agenda.

D

I have a question a couple: um the first question: what what what database do you support.

B

B

I think it's written somewhere, but we have we use postgres right now. So any sequel, like supporter, like yeah, any sequel, is the one we are using at the moment.

D

And you also mentioned mention docker docker reserve runtime. uh So after after the maina creating system is installed, uh does the docker remain on the host, or is it also temporary just for delivery grids.

B

Yeah, it's it's temporary! Only for the for delivering and installing like to making.

B

All the action on the server, but after that.

D

Okay, thank you.

B

And you have your operating system of choice without any. Thank you.

A

I think you broke up, but I'm not maybe it's my connection, but I don't know.

E

So I have a question: this is diane fatima.

E

How do you discover what hardware you're provisioning on I mean if there are accelerators on it or if there's something unusual about it, like a high-speed networking or something like that, do you have some sort of discovery of what exactly you're provisioning on.

B

Yeah, there are two different ways we currently support. One is that every hardware will register itself when we get the first dhcp request and obviously with that, we don't really get a lot of information about the host. We just know that there is one host that has a mac address. That's it uh another way is to register in the tinkerbell server, all all the hardware, and when you register them you can it's a json that you send and you can. You can save metadata as a like json, so you can mark and label your hardware.

E

Okay, so you've created this list of things that might exist in this metadata format. Is that also in that in the um the github repository just curious? Yes,.

B

Yeah, it is part of the of the think, repository.

B

Right the server right and show you.

C

Hey john luca you're breaking up really badly. Do you want to try killing video, maybe.

B

Yeah, let me let me change my connection for one second.

B

I'm back and just let me know if you if it works better now. Okay, thank you. So this is an example of the hardware data that you can register to tinkerbell itself and, as you can see, the the idea is mandarias, as well as the um as the mac address, because we use that as identifier to point workflow to, but you can store like way more stuff like we store the facility and we store like the delay out of the storage um and yeah. How do we want the partitioning to be uh yeah?

B

So you can. This is the way we we teach like think about the how the layout of our hardware and, as I said, if it also can be done magically from the the first dhep request, but obviously you get way less flexibility, because there is not much to to get from the from the http request.

E

Okay, so they they specify, whoever is providing. Those servers specifies it in this way, it doesn't go out and investigate you'd have a script or something that investigates what hardware you're running on.

B

E

B

Part for now it's it's free, but you have to do it. Okay,.

E

So like the number of of cpus and things like that and if there's hyper threading and all that is that, can you do that or is that something that you're planning.

B

Now, for for at the moment there is, there is not that part is not made covered by tinkerbell. We are thinking about an inventory management like uh solution that would be like. I think it will come, because we had the same like uh question you raised, uh but we didn't get that far yet so it's it's under discussion. I think we will get at least a prototype that will. uh I I think there is a. There is a reconciliation phase that happens uh when aussie starts that uh sends a couple of information like this.

B

The the architecture and all those information are are coming from aussie, because aussie runs docker and our like worker uh that sense and reconciles those information. um So we have a bit. We have a something, but it's not a fully like an inventory as as we as we usually see it, but I think we will do it at some point.

E

Okay, just curious: it's not an easy problem to solve. There's so much so many different flavors of hardware out there and yeah and so to just do discovery automatically. It can be difficult because you don't know what you're looking for in every case right. So I think this is a nice approach. You have that yeah.

B

Yeah you're right, it's also. We also took the direction of um registering asking for the registration, because we don't want for tinkerbell to take over like all your data center. When you start it because it it. You know, if you have the the http every every everybody starts to get the response from php from tinkerbell.

B

uh If you do not segment your uh your network, so there is the possibility that if you don't configure it like, if you, if we do auto discovery in a very strong way, like think about, we start to provision stuff.

E

So you say what you want provisioned, maybe you have an accelerator out there that you don't want to include or something, if you put it here, then.

B

Yeah, so that's why we have those two different and very like uh far away strategies. You can decide to use thinker as a inaudible discovery mode or you can register your your, your your own, outdoor by yourself.

E

Okay, thank you. Thank.

A

A

So, are you thinking also about uh maybe uh making this a cscf project, or are you guys interested or.

C

So, um as I'm uh um we've definitely been in talks with that, mark coleman is kind of heading up. What that's going to look like in the long run, but um that's definitely something of interest.

A

Okay yeah now we have a new sandbox uh process. uh I don't I mean I mean we have the cncfs sandbox incubation and graduation, but uh typically the projects are submitted first through sandbox and then right. They stay there for a little while and then later they want to, you know, become more of a real. You know project, I guess the more people are using and uh but some of the toc members do do diligence and then they go into incubation and finally, they go into graduation. So.

C

uh Right, yeah we've got a we're working this quarter heavily on some stability issues. Making you tinker expanding. What tinkerbell tinkerbell does how it works, what it offers um as well as, like I said, getting some some more stability there, um getting a few things off the road map into play and into production, and then that's something we'll definitely be looking at a little bit more.

C

But I know, like I said, um mark, has done a lot of work with cncf uh and we have a great relationship with cncf. A lot of the cncf projects are built on packet, so um I know we've got, we've got a deep relationship with with the foundation and we'll be moving forward, uh getting tinkerbell over there as soon as it's.

F

Ready yeah and lena just posted uh the link to the sandbox applications. So thank you so much molina. That's awesome!.

B

A

So we have, we had another project that I present uh a couple of months ago. uh It's called metal cube. Are you guys familiar familiar with that project and without metal, wet metal cube uh like metal? Three.

B

um I am not yeah, I I'm personally familiar uh with we yeah.

A

How would you describe like, uh like tinkerbell as a I mean, I think, metalcube uses kubernetes, but how would you describe and some of the differences and.

B

Yeah, I think I'm familiar with for what concern our uh like cluster api implementation, because uh we, our the packet one is, is like new and we obviously had a look at the other implementation, but I don't have like. uh Like practical, uh I mean I didn't, try it, so I can really do a comparison.

A

Yeah, I mean yes, some of the questions that get asked in the toc, sometimes when the projects are going to incubation like how some of these projects are different, because I think a lot of times, they kind of just want to fill some gaps or for some of these projects that don't exist and and maybe they want to promote a certain project in a certain way like like okay. This is good for this type of things right, so so maybe those those are your questions to to keep in mind right.

C

B

Yeah, I'm also sure that somebody in the company has an answer. We developed the answer already, I'm just not right right.

A

Anybody has any other questions anything that they want to.

A

A

Well, thank you for the presentation. It was really helpful, uh yeah and I hope hope uh you know this can become a cncf project in the future.

C

Hey thanks for having us, we look forward to uh working with y'all a little more closely once we get this ready to uh to roll over into the cncf.

C

B

Thank you. Thank you. Thank you.

B