Kubernetes Meetup Videos, 23 Oct 2018

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Creating Kubernetes multi clusters with ClusterAPI by Tobias Schneck Loodse GmbH

Description

No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).

A

Yes, thank you for your introduction.

A

Thank you for warm-up, commander Craig good science was really delicious, I must say, yeah I think nobody knows me and because upset ever tell something about me and Lutz say: what's our plan to tell you today, first I want to show you the basic setup how, by these classes, it will be pulled up if you put it by your own or you run it in the cloud like a class or something like that and then what's the new idea, what's chromatic, so what's we are doing there and why we sing if NASA API is a better approach to deploy notes than the old way to deploying notes, and there I bring some example.

A

How easy it is to integrate a new cloud provided to our our stick. So we integrate it heads now, it's not as pretty good internet provider for fast ends and unfortunately they say currently: I have no storage door load balancer as a service, but say working on that we are in touch with them and they say it will bring up storage as a service this year.

A

Maybe I need to plan for next year will be also, let's say, can offer like load, balancing and all stuff that you need like for real production, already kubernetes infrastructure which I'll provide them yeah, and then I will show you how we can use cluster API to create notes just the definition and yeah what the state was wrong cons and where is it? Where is it? Go me good, then.

A

First of all to me, I'm, not a ops guy, so I came from the Java development and there I got to test setting up test infrastructure for test automation. Basically, basically, we did like containerized through your eye testing, so you did like running kind of selenium tests inside of proper containers and do it like in a real UI, not just a headless process, and there are over I came to Lutz it, of course again and now, because of this experience to have like a common areas and containers for test environment setup was so helpful.

A

Now I came to this point and try to also involve setting up kubernetes cluster and to get deeper into topic. So a few words about loser, who have you know, would say already: oh cool so but few people, so we have fronted 2016. The headquarters is number we currently 25 employees plus worldwide. So we are growing crudely, quite hard and the basic idea of our company we want to write.

A

People were, are keeping the topic and we were are selected so I'm creating unique in that butter is in tempered, so it plays like don't care all. The setup is mostly remote and yeah, but we useful yeah made up like 101 time in the month or two turns in a quarter to do something at once and so on.

A

So and our mission is to work on cloud native technology made mostly on kubernetes, but we also support like cloud basically out consulting something like for the TCP cloud, because we are also Google Cloud partner and we do some trainings there, and also your member of the cloud native foundation was pretty good because there are not so much members in Germany I think it was two or three and for us it's a good way to get the first inside informations directly from them and also being such decisions would be their roadmap and gets his mom Asian like in the first level yeah and we put up kubernetes cornetist in a sentence is like emoji.

A

Hyper clouds contain an engine which tries to deploys moodle cluster in different cloud provider, so you can deploy it on premise and you can deploy it on a cloud and every time over the same. A and saying is that we can operate it from one single instance a weekend for an operation. Tasks from 4000 won our meta cluster and there we can control all the stuff like this kubernetes base metal.

A

We can control communities installations, so it makes pretties for us from like operating stuff, because if a classica have a problem, I just use Q control and look into the objects. What's the problem and that's the basic idea also for glossary. So that's the reason why I'm here today so but to start kubernetes.

A

What's the basic component is deployment so mostly the basic abilities to go and have master so first, you have to start so you have to pull up an API server and a controller manager and the scapula and for your data storage at CD. So it's almost two components you have to build up.

A

There are a good column for that, like you bury em or yeah like talk a machine for for local stuff or something just place it a basically, the problem is solved, but if you sing around, if you have to put this like for 100 clusters, you will come to a point of organization and there you have to maintain all the stuff and have to ensure that you have backups that you have to XD you running in like a setup and there you have like operation for your community's class of command lettuce clusters, and this should also solve your operation of the kroner Strasse.

A

So that's the basic idea behind it, because then, if you have to masculist, you have to add notes and then nodes, they have to also be in to be installed like cute. Let see advise that you proxy, and maybe some network plug-in or storage black and you're already here.

A

So you have to do a lot of configuration there to bring up new nodes and if you are thinking of bigger environments, you cannot maintain south nodes by your head, so you would have to automate it and therefore you can start again writing scripts and what's the pain about scripts, they have no stage so screwed just execute it on one time and they don't have like yeah. They don't care about. What's the current state, so this is kind of yeah engineering processed and you can write smart scripts, but there are still script.

A

So it is mostly a problem. Also, if you use Sarah form I think it will come to a point where give like parameters and then you try again to do a lot of stuff and we try to symbolize that and not. We only saw also the community and that's what's about the class API and last April is a official component is sick project. So it's a sub-project of c-class lifecycle and there are projects underneath like cops, q, barium mini cube. You already know I think so there are some reports.

A

You can take a look at that. The old one is Cuban a DC proxy here new one is kubernetes, six plus API. So if you want to see what there are is going, you can get in touch with them. They are for select channel if I think weekly or bi-weekly meetings, where you can discuss with them, they're pretty open to it, and you get like first site information about that, and there are Menken systems like google Ratatat. You try.

A

A primarily and sis are like our part, is to like implement it already in a similar way. We have now ready to cast the api, because the concept is the same is just like naming fields differently, but this is not getting standard. Why we want to have this as a standard? Is that you can build tooling around it, because if you have like class object the same way you have like a hot object. You can do more things about it.

A

You can maintain your classes over API, you can say: okay faster, please upgrade so, and then you have me and it's norm, controller who controls operate process, and so this is basically here that you get more features around this space class. The API which all people speak I, think the same, like storage as writer will provide the whole you can create by them machines and then the tooling around that can then work with them and you can more torque it like in future plans. You should billing information.

A

So then you can like compare what's the cost of different much machine provider and then choose what you want. So that's like plans. Okay, so of you with chromatic chromatic is like I. Think first production launch was a year ago now we have implemented according pro providers when publicly Outsiders heads not to be the ocean and web service mixture of a show and telecom cloud. But I hear to telecom cloud is like kind of cancers.

A

B

A

It can you want, and also like on-premise it can, is mostly opus, take in VMware and there's also like for Co for sexual URI. We also can speak with them and also be a metal, so terror is to use his that in like, if you have a potato Senora, did you get away from the virtualization layer and use it's a bare metal system and terrorist? The idea behind it to provision eyes like over pics food, the new, the new nodes and say, okay, picks food.

A

Please install Cupid, please install that and wet and join the cluster there. You have like to pre-configure some machines, because you cannot like find out what is reachable public spook, so you have like a database or some some kind of source for you for your to know. All other stuff is like provision eyes to cluster ad hoc.

A

So we don't add some resources, so we just created when we needed for new cluster or if you want to scale your cluster good, but going back to the note diploma in so that's basic idea of cluster I API to improve that. So, like it already said, there are much rules out there. How you can create your clusters, you can read it over the public clouds, so each cloud life is own specific way of doing that and more or less, then you are more or less stick on them, not only on machines.

A

You will stick on them with their solutions. So, like counting like house, a clean up builds how hit some more monitoring and also like tech, upping management work there and it's like yeah, you get more in it dependent on the cloud and there are some other way to provision eyes like Enzo terraform. They work mostly well. If you look in smaller or we say you want to manage three to four clusters, but if you want to match 100 clusters, there will be get a problem.

A

That's the use case. What we want to solve is last April is not to automate one single to medicine solution. It's already solved, so it's more like operating much because it's the target or over would without power one interface and okay, I already told corn brauch is you need Ops guys to scale a cluster? So mostly, if you, if you have classes in see, ok the fastest fool, you add, you would add nodes, and because this is like tasks, you have to give an opt guy or to organization to give someone a new work.

A

Mostly the Communists gauss's wait the classes get sized too big like in opposition you're, mostly if, like won't last for the whole company, and they have just very namespaces and if you build this up, say put it up mostly like this space.

A

So but this space you have to pay. You have to pay in with machines on private cloud as well as a publisher, so, but you may use it in a year or something, and because of that, you don't want to. You cannot do it like on click and that's not the approach. You want to be to save costs. So in the next step of class, API will be there.

A

You implement, like an horizontal note auto-scaling, who will already have it for a Google container engine, so Derek say like find out what capacity to cluster nice and then scale to cluster up and down up scaling is not a problem, because upscaling is easy. You can just add for something. Robin is more down scaling because down scaling. Is there harder use case?

A

You have to decide which note is the cheapest note to destroy, because if you like have stateful sets or something it's could be a problem, so you have to choose a smart and also may your eye. Tech chure have to know about it that the notes will gone, but you have a good across fixed cost.

A

You can save costs there like we already use for our master cluster. Sometimes we used kke is basic cluster because they are, you can use infernal machines. Note machines will just look for one day and then say a coolant and reschedule it, and suspicions are like I, think, one of one search officer she put in the normal machines. Why? Who will do who's? This say one to fill out two caps. They want to be able to reschedule the machine, were gaps are in your quick data center and singing in no small way you can.

A

Even you can also do it in the data center or if you have like a data center with a limited capacity, we can add notes, maybe from the cloud and or shift classes to the cloud and then that you can manage multiple environments and think about like comparing two different cloud provider. If this is extremely hard that you meet migrate from one flower to another cloud, yeah we'll still, some project costs or and then you cannot like compare it. Miss riser and that's. The problem is see this current way of setting deployments and scripting.

A

So so how class API works in technical? Who is so? You have like an infrastructure provider, and this could be like a table is Asia or on-premise whatever, and that's all my providers and providers has provide a specific configuration so like machine-type. This is some called different in AWS then leave em there, yeah I, don't have to machine table. I can specify CPU and RAM so, but they all have like common things around so I want to install my basic operating system.

A

I want to install cue plate, I want to install doc engine, and this are the same informations and best that the IDE is here to use it. But first it you can install a class API, because last API is a custom resource definition. You need a bootstrapping cluster, so he's serve up. Myths. Existing tools would strike a machine cluster and they're there.

A

You have like your normal kubernetes set up those SD and there you have in class API server what just have stored the customary source definition. Then, what bring the custom usin resource definition into the cluster? It's a controller manager like custom controller, which can control machines and the machines have like the object. Then machine and machine set and cluster. So you've like get clusters, you get a list of class, so you created over there and in the user class. So we see it later, then you can keep control get machine.

A

You can see what missions feel you have there. Then inside of the portion control, you need to provide a specific card, but if you want to have like five VMs or five machines of Asia or we ambarish appeal from perspective from specification should be the same so there it plays in like automatic machine controller. This is like, oh in the play, implementation, how we use it and how we provision a table is Asia and VM. So currently the provider are not really ready. We see our implementation of kpi.

A

It's like rely on the class API because just API is in alpha state. So what we do we just make four throws it and bring it production ready and then after we see it, there are some new features. We make the sing again, and this works bright, good and the good sing. If the glass API will be released as sub stable API, we will be mostly the first who can use it and are, like speaking just API in official way. So what Ben is happening.

A

So the controller takes care, I, say: ok, I want to cluster new cluster I want to have master components, and then we want to have note components and then define it in them. Gamma and then the controller takes care to create it. Like you said communities it serves, it's like you define a state in kinetise should be able to create the same. Let's see here same you define the state of the cluster, and the controller would take care about that. The class get that the casket iterated in this kind of state.

A

If you change something the controller, get the notification and can also like to eat machines or something like that and we compare it and this you have to think about you. You maintain your clusters, like you, maintain your pots and pots set or replica sets and about all the stuff and yeah.

A

Then you can boost up the clusters, and so therefore, first the API selves and the control components will be installed. So you have like then user class, the API server, and this awesome LCD running- and this is like installed by QA DM in technical detail, and then they will connect it to the master server so get the controller get informations. Ok, how can I create new machines or something like that? And we see ok, a specific idea like 5 node machines? Oh there's nothing!

A

So this controller gets all red and trigger their machine code or in and new note machines will be added in the same way you can do is with this class upgrades, so you can like define, make not alive update so just adjust at new note. So not maintain the notes, more notes like likes, kettles, you handle notes and messages like petals and it's a new way. You want to maintain it because I see it in the past and I did some overshift consulting and there was like. Oh the note running, fuller I.

A

Do it as a chair and clean up the talk of cash and oh maybe we should write a chop for that and then you're building up your own maintenance cycle and okay works. For that version, then you have a new upgrade and all things has changed, but why I say maintain the notes because it was too hard to created and that's what's the reason if you have like heavy to it like Google on Google, Chrome click add new node MS there.

A

We have no reason to maintain it or just at any node of stair some Brooklyn and it's like a note, is not ready, recognized.

A

A

A

Yeah, this picture is for the glass API say on specify it in our chromatic installation, the HD here and the API server run inside of that cluster. So we match this cluster. In the error. We have a containerized, API server control Avenger in HD, and we match that sinks in the master class or, and you can have multiple classes- okay, mass cluster C cluster and then cosmic Lustig, mascarpone's, neutral data's in them, and we just the notes, are outside of the control. So yeah.

C

Because they essentially, but it's writing in the on.

A

The chromatic I can maybe just open my slide.

A

So that's like the standard which are maintained by the.

A

I get the right presentation, but so.

A

Kinetic we do it like. We have the Masterclass already set up, get scripted so and there are mascarpone's and we have some workers for masculine, so conditioned welcome next class. Then you install cube medic, it's just me and deployment at least so there it will be included or caster resources and all stuff in the controller, and it's like our API controller and machine controller in the UI to create custom.

A

What you can make you make you use it namespace, this kubernetes master components in there would like to use a class, the API server, control of my schedule or again CD. So the next step yeah it's on, and so you do it for one class and can do it for more cluster. And if you have like a dispute datacenter, you not want to have like mass components, Germany and the workers in China.

A

So we want to bring their masks for sauce to China, but not, but also want to maintain it not in traditional the ends like ratcheted it to IBM's for for the mass components. So it's this way we have like C classes and that's are also like chromatic known cluster and because we go over the see cluster deployed there, the mass component and all that C cluster, the neuronal yeah. So what a good thing is, you can run for all side offices. You can run like our pickup shops.

A

You can then motor it all of different lofts sizes overall and you could assist class. Their problem in this mass components is don't affect others. It's also like if you can begin bigger. So if you like a scalable way yeah, and so it could be like a setup- could be that around like mass components VM there and you have this user managed your own internal em clusters and down units. Like a sure yours running over some.

C

A

We can do that and I don't detail.

A

So yeah I sort of picture.

A

A

So, in the next step you have like outside events, and you have like upgrade triggers or or later on, maybe auto-scaling events. You already can speak with the API, so we can like create scaling kind of cron jobs or kind of a promise us matrix. Then you trigger the API to stay down stay up, but it's not like an auto scaling. It's like more Diagnostics like scaling, you can already implement and then but the same.

A

What changes is the castle Yama in the machines Yama, where you have the governance configuration our class of, looks like look, I changed and you apply it more or less to the cluster in the controller. Okay, take care that you classic it to the state and if you delete it and the castle will take care to delete the resources, and so that's the cool thing that you just can say like an upgrade I changed version of kubernetes and the controller knows how to operate a custard and it's different.

A

So you don't run its group again, so just define another site all right, so how's. This integration looks like more detail. I think we already have talked a lot about it.

A

You make like you can do like you clear you control, create missions, buy your own or do it order plus API server, who mentioned that, but you can then have like machine objects. You can handle that you can maintain that you can look into that like normal coordinates.

A

Okay, what's a machine machine is at a crafts spec for nodes? It's not a node itself, so it's specifies like the virtual machine, where the run notice, running and inside of system machine there's like a property like the couplet version, which yeah is the node and then if the VM starts and it's tautoko installed, couplet joins the cluster and yeah so how the code looked like. So this is like spec how you can create a node on headstone. That's all! That's like define a type, it's machine. You can already read it.

A

Then you have like SH H key, which we edit automatically. If you really want to look at node, I really didn't who it's quite a while, but you can and then you speak like well provider, and this part here this part you damn pointer this part here is the cloud irrelevant difference, so only you specify what's the server type like utopia and that's very cool and have something it guys were small, and then you have operating system, that's not a cloud provider spec. So it's is just a general vacation.

A

What you want to run there, so you see it's like learning. New kubernetes object or operator objects like the same units, but you find maybe your PostgreSQL database all right and you see the nexus is still I thought we already don't know when they will get better, but ok and the miche's machine could roll out then uses provide them specific information and no like singing like a driver and use this like implementation and knows: oh ok, I'm now on hats, nah I have to cause this API to create a machine.

A

Missus sizes I have to create a network here. I have two great, maybe some security crews for virtual networking and something like that. So that's that specified there, how you can add your machines to your cluster and, like algorithm more, it's like you, what for a machine or a node change, so you take a local machine. The specification has changed or just created and then or note is not huzi. Then you say: ok, accurate, good machine and stand check.

A

Ok have a little need to be, sometimes them why you just thing if I delete up object, your scone, it's this case. You cannot because if you delete job check, no no controller can know what to meet. So if you delete their objects, this is let the controller keep it and we start with a leash process and after this process is gone.

A

Mr. object, so we have like activity minus 10. So as long there we have like a state. Treating if you take a look, then you see Keating, it's sometimes you clock it. For example, a if you have like regular service with stabbed open and so on this bound to a load balancer. We want to eat custard. So we stop the deletion because you have to. You cannot delete that not Peninsula instance, so if it's just reverse or PE independent observers yeah.

A

So if then, or if you're, not in the use case, you could say: okay, if I have to just pacification, don't fit to instance or in fact instead, if I provide instance, I want to create one and after I created one all right, how the time first of all, I get information for the club provider, okay to VM started after this I wait unto the cubelet connect to make cluster and then I say: okay, the machine is ready and the notice ready can get some working load of that all right, yeah, and in this great example, we have like a render use data like correspond to Center it yeah.

A

Each collaborator, you may have a different way how you can achieve to deploy into or and chorus so like, and where you have different templates, you specify in the installation and say: okay, if I use like horror as I use that one and you can even specify more than one but that's like standard and in in AWS, used like cloud image, and then we use mostly cloud. What's that name cloud cloud in it 210 provision eyes the Commission's, okay, good, then test to provider interface.

A

If you go to the kuma net gonna this class API repository, yes, the providing of her face, each provider have to specify. So if you want to bring your own but for maybe arm devices or something you can just have to implement this message like first validated to check what's the state of a machine, then you have to get so. How can I identify a machine again that then also like a cloud config. So that's like transfer between respect to the directly to the specific law provider config. You.

A

B

A

Yeah, so we speak with them in API. You can take a look at that. It's open source or you can at your own yeah.

C

A

Yeah you can, you can so the whole chromatic setup is not open source but the machine controllers, but you can install the machine controller to your cluster and then that your own UI, if you want yeah and then we have specified- to create how to create a machine and to the ticket machine so and for heads now it was release, is like a 300 line of code code for the file.

A

So I didn't write a code, the don't blame of me if I say something wrong in detail, but most of that there is pretty easy to implement there because they have like a really clean API, because it's pre new also and it's a good example to show and they re. If you want to look there's the link on the repository.

A

B

A

So here you see like create machine.

A

First of all, you get to get conflict and if you go there we make like this is the spec value and Marshall it and to transfer it into a cloud specific code and yeah yeah. You have some meta motive like for a head, start, open and so on and that's all then you have liked configured, let's go tick, and if you have like conflict, you can and then execute it and you use flying just talking, and then you cars, like the operating system and mr. claw provided details like your provider.

A

Teachers are mostly new to the API of God provider. So because of that is not much code, but because it's part, you can just add to the payload from the from the shuttle ride really easy and then you apply it to get a risk operating system and well. Yes, then it's great the Moschino object and you have the Machine object and it's will edit will be okay, edit to a data center. So no data center.

A

You have like multiple data centers, then you know in which data center run your machine and if all is done, yeah you may SH key.

A

Someone return, you can return the object and look at so I want to show.

A

Yeah, so that's the conflict. You met.

A

You get the location.

C

A

Can give the data center as well as location.

A

C

Yeah I can will choose like I, want to have this in our x19 and the next hour, 16 yeah, to be like normal.

A

Locations like the country and the rate is a novena to like frankfurt and so on, and that's kind of to spec just want to show you the class object, but I don't find it um we're on the machine. So I think, of course, so.

B

That's yeah. You.

A

See here that's the like machine object and you have like metadata provider, configuration versions and so on and so on. So this mostly specified machine is not hard. So the not much, of course, and also it's this class of it's known as 300 line of code. It's the religious trees, it's not so much! So if you have time you can take a look at it and you'll I'm, sending a virtual good, then yeah.

A

The machine control is on the pipe which to license, and we contain currently the Machine controls for a Tobias asiatique. No, she gets no sickness fear and they are currently included. There wasn't option like bring your own that you can, if you want, if you like, heck some other nodes into it, can like specify, bring your own config and edit. The note and joined in cluster.

A

All right, so, if you want to install it, I didn't try it but should work like that, maybe not so easy, but I think it's more or less. You can shut trust, all machine control, oh and then you can act over the API, at least to create machines.

A

Alright, so what's new now we also implemented the Machine set now so they are. You can then say: ok I want to have replicas of machines, and you can then say: ok, that's my template and there is like normal machine configuration like you already know from replica set, and the next step would be create like a deployment where you can then create go in the governments of cluster versions so automatically and I'm.

A

Currently, no, you add new nodes with an uber or you can upgrade master version and then you add new nodes and delete the old notes, and it's also works fine, but this is maybe also not so hard to automate it. So we can automate it and the next step later on, this would be used for the Machine set will be used for the autoscaler. The old skeleton can kept information from the Machine set and see. Ok. Maybe there are some information which kind of machine I want to.

A

Like the hassle of there's, a pre, pre, II I, would say pre-installed images so like they're mostly way to configured, and you can then say like upgrade on class start or not yeah, so so what we tried now is to create your cosmetic and Chester and it uh possess organically ocean. You can also try the premise of I, don't currently anything, but how it is you can by a way of our own.

A

We are not a service provider, so it's just our demo, given some guys use it like for production and say mean it's a free service. So there's no more charge. It's a monitoring for testing our monitoring, bigot race. There's! No, like I, don't know. Currently the detector will work. So we have no operation team, that's like 24/7 or reachable debt. It's just them cluster. How you can use it, but there you can care to provide us. Aws, Asia and the ketone are are available because they are you. Can you bring your own credentials?

A

You can use your AWS account. You can use Eurasia account or o'clock tip the ocean account you create faster and can play around in scale. If you want so, but the idea of karateka step to install it in your kind of infrastructure and then you can also add different configurations and product accounts and services which sell a product, so it's one clear, try so and brand-new development branch, and they are.

A

We also have to approach it that we can then come together like a user management, more clusters and then add different people to allowed access to cluster. So that's still develop land, but I wanted to show it I mean like here: I trust. It's me project and I have like SSH key here, and there will be also possible that you have like a shared SSH key or with user accounts. You just don't need as HP.

A

You can just add users and all the users I'll be able them to log into the clusters and to maintenance all right, then our great. So what's great, a third one.

A

Yeah, you can specify your version.

A

You can already install one cut, 11 and certified so then, okay, what's the newest and then you can choose like your provider want to use. You have perforation with low providers.

A

It's the bring your own can like Hector. You get a cube, ABM collaboration. Then you can join your.

A

Justice itself.

B

A

You know we do it with a cloud in it. The project and I'm, not sure if they maintain different image, I, don't think so. I think they have a basic image and then install the Cuban I just motion yeah.

A

So the session.

A

Okay, I would choose to charge because the first one.

A

And you only have one talk so like that: I have an AP item or escalation.

A

Because it was the reasons of so small.

A

Then you get, you can specify the number of notes and you can specify what kind of machines I did it to all shouldn't, say droplet and they are you get information from the cloud provider. So we ask them: oh hey, what's the size or you wanna you offer, and then we can choose it so, maybe not to bother my boss. I just used a cheap one and they are.

A

We have two different container in the images and this is specified in the chromatic installation which maps to what image so cocaine Allah knows because of the fastest and yeah it's just a QE would also use cry or is runtime if you want to use it it's possible, but it's not enabled by default.

A

Yeah and I think you have some additional option. This is also provide a specific like in a couple years. You can specify your own network and some I'm just networking stuff, and here dick tours you can like say: ok, it should be due to mention Becca piety go to ocean or you want the monitoring of to those particular ocean and it's like a cloud provider features. You can. Then choose them and you can add text.

A

All right and then next and then created so.

C

This case you have C cluster.

A

Now, in this case, not sister, it's this environments, so a play environment or for the developer, so make no sense to pay their money further attention to see Justin, but like this cluster Todd Collins live, it's also not huge. It's like I, think nine notes or something like that surrounds on dks was just nine node cluster and alright. What then happening first, we took master will be created, so the Masters like a new name space under master class or in there the new single will be started and yep so I sing.

A

This takes on a while yeah, so I think we said: I show you the another cluster when it's ready then fly for our ewic of like a API, so control st is the Machine Jerrod. It's like our at all and for a double SP also apply in the cluster, like drop configuration that you can use log balancer in the storage class, and then you have the overview, work machines you have and because PR more in the development part we take it. You can take look to sis cluster, so I can download to cube own fake.

A

A

And if I go to my obvious account, I already see sis machines running and there is like Tech 4 cluster.

A

Or is it yeah, that's the new one. So that's spinning up here, I, stick.

C

A

Behind sana, provide observers.

A

Let me show it so.

A

Here in in AWS, we can either use your Chi group, and there you have, like your own MVP, see.

C

A

You create it straight, it automatically doesn't it from so if they have fun like it's, not didn't, have a little private network or something like that. So if there is one we created and then like the mask components and the notes, we talk. Oh well, if the end connection so forces network to come to the mask cluster, this will be talk between ppm yeah.

A

A

Yes, alright, so I downloaded it and then try to.

A

That's the EWS class I just wanted to show there how to it not now, and then we can go back to do another one I created so sis classes already running and there I now see that I have. If you is a new API resource, that's here the machine and then I can take a look at that. So I can say he get missions and I, see. Okay, epic right to our before three missions and I can also say.

A

So, okay, informations about how it is created.

B

A

Step is that we want to apply option. The only thing I did to create this option is that I get some like IDs in status from the AMA and I can maybe compare it live.

A

A

So this would be the one I want to add and I can compare it there with them.

A

Right well over here.

A

Shouldn't do such things.

A

Okay and that's the you want to apply and that's like the cool running, so we only only needed such kind of resource version and the status which is running and then credit like the new machine, were specified. I. Can here's better specified like this size or something like that?

A

I can also choose they're, not a value so that I create because this and apply some labels and some stuff, and there is see the VPC in this case you have to edit because there's already created custom, so there will be civil to create before the machines all right, then I say keep control.

A

Yes, you can choose it like the dialogue, accept it and then you can specify it. If you want to override it. Okay, then, let's see access watch. So now we already have a machine and if I connect with that I see.

A

Yeah, so in this case currently you see what's happening, so the validation successful successfully, create the machine so and in AWS now we should see that there is a new machine coming up. Let's go back you see here is one machine initializing. This was the named Manoah to be, um maybe you don't see it, but here's the name or decorated, and this is now being created from from the cloud provider and after that happen, and then you of that you get a new node.

A

So you see like here, the Machine object is here, but to notice not here so two stakes minute or two: it's depends how fast the Machine creation is. So it's already created initializing. So we take this long and then attach to.

A

Although there you can also like to eat notes from the user classes, because and since you don't have the information, Texas kill, but promotion go to life.

C

If you just take any.

A

Of the existence.

A

Vpn is just so we have come over here and that's for the internal and civilian Klein has cash for the management that we can communicate from all Master components through to the custom.

B

A

So this is running in Frankfurt and.

A

I'm not sure what was the decision, so maybe just ask it: we have a selection or you can ask it there but yeah. That's. The connection is between.

A

This components between is VPN connected.

C

During the communication.

A

C

Yeah, it's like.

C

A

You spin up multiple clusters, so if you have like.

A

Okay, maybe just give it to after the talk.

C

A

The notice there finally and.

A

A

Okay, so there you see that sorry couplet and connected, and we have all notes ready and then you want to lead one, but first I want to show you that it's like also came up in the UI. So we now have here like much different object there and also can delete it over the API about we mount us. We want to it over it that well, okay, so.

A

Yes, there will be removed, so what one the manual one or animal.

C

Quite interesting to kill without the machine.

C

The problem with like double States right with the key key, where.

B

C

One state and the other wha if.

A

This get unredeemed or Jesus chair this note little bit, it's a machine control over wait, 15 minutes, not that we would need it and create a new.

A

So if the note is like a now see our amount of time, then the machine control say: okay, I want to try to repair it and delete the machine and also the note then, and spend up like a new machine.

A

This cluster them are you going to destroy the cluster. You cluster itself.

C

A

But this is another state, so listen.

C

You have the big problem. That is for some reason you can't access the master, then you can't get the state of this specific, that's dirty. So the question is what that would it's the cluster lot lately, therefore, the state has to be repaired. Therefore, a new cluster has to be smaller, which then maybe like tries to get the same, but.

A

Not if it has another life cycle, so the classic said.

C

A

So you talking here about machine notes or machines and the constant serve, has not a state so like you, don't need to chromatic components or class itself. So if you like, if you have a seat cluster and you killed your master cluster I did it to get all the instances. Trust me, the class was still reachable and work of them so just from scale or image something, but they store all information sets T, so they've tracked, and to reconnect it and as long the cluster is object is reachable.

A

So don't you get to class the classes? I mean it's not reachable. So, yes, another safe, there's another life cycle, I'm, not sure which, which different states they're possible, but there's an understatement. Last object.

A

Yet it's sometime of connections issue, so he was already, but you will see that later on.

A

A

Sometimes they're in the starting-point give more load on the note- and this was looked like a big mission, so no matter sometimes get unready and the connection is also not there good then, but it's you're not deleted like immediately. We just see it safely like stable, stable, not stable, okay, not need one.

A

A

A

So and Dan has already gone and four clusters already scroll away and we will see under here it's also God, but in like an AWS here, I have to state terminating I, think power. Shutting down so and like you made the is you set it down, set, delete, call and say will confirm it to you and then we say: okay to distill, it stinted and drop it.

C

Off somebody do this machine, might I, don't know misbehave and fail to stop.

A

Yes, there are phenomena so.

A

Yes, if you like.

A

It's just a creep on him, no yeah. No, this is the newest version. You can then upgrade so you will operate the massacre points. This is please for us, because just like rolling deployment for the mass components and art Edwards, you can add new nodes and can then specify like the new version of the master, will be edit there and the master version with employed. And then you can delete alt modes and that's just the next step. You want to automate just three upgrade to the new version and all guitars.

A

Have it upgraded or maybe multiple customer, you see alright I think I shot a lot yeah. Take it awesome, just I, created by the way.

A

It's like the same.

A

We have here components. Oh, we have no notes.

B

B

A

I, don't know, maybe resort things interesting.

A

Just in time now and sometime, it's like the crop writers, not reachable.

A

A

Yeah, okay, so what the thing is that we can like manage all the cloud providers about one state in a one objects and if, like error problem, would be I think it's feared on a secure API, just okay, but the thing is: if you like, Aris problem I can take a look at the masters cluster to the lock and see. Okay. Is there like most of your issue with to go to ocean mr.

A

de epic, I cause missing or whatever problem, and then I can then maintain the class to like a British object, so I can see to class. The state class is not ready because of that and I can see in the drawer creating the class stuff layers about the society ID. Okay, then I get near red brief past God.

A

It's all rigged Molly spoke too long, so this is to keep oh okay, so hopefully I only need someone. We all can already connect.

A

This must be the.

A

Pattern is this: is the class tidy yeah and that's no random name, and you see two notes are ready and.

A

Normally, it's like I think three to five minutes for me cluster. So mostly it's the start of the games. All right.

A

A

Then sync, your school, so more I expected and yes, what's the status of class API so plus API currently is just no alpha. It's not how our product offers like trusted protection, but I think for itself. So there are much discussed you can think about. If you have to like Google me, my redhead and all the big players aren't able to specify one API mmm could take a while and say quite doing change backwards in capital currently. So this also reason why we just do some force of them still haven't given time and yeah.

A

It's a don't solved like broke limbs pick up at your storage cluster. So that's not something cafe. Api class API is focusing on creating objects, so each other have to be care about it's just great cluster itself. You classify it not solve all the problems.

A

But what's cool thing is you have like a decoupling and you furnish provide a specific culture if you'd like building up any new application like we did give a reading around your hand, proof can use it for multiple clouds. There's like interface, a cloud cause. You can then compare home. How much you cost. Stick your clusters of different routes and can guess there are some informations all right, good and there's also good products like all kind of public cloud provider like Google and calculator might read, it are on the table and.

B

A

I think, if all all wants that- and so this would be a good approach in the future to handling clusters and class Tom Skinner is already possible to go ahead. It so and say also community say the open sources.

A

I'm, not sure if me is already open up at Salem, yeah, all right, so what's different what we build around so you're at first of all, you have to connect dashboard so and we edit it like a backup, restore concept and there you can't in like match all your cluster over one point to backup restore the SD and if you want, you can also like at their control us this art that you may tend a copy.

A

All your persistent Mullins, the cool thing is, you can do it in more place so on our cluster, all clusters to justice and there you have more possibilities. Just if you have like in the organization, if you have ten classes, you have to duplicate your backup strategy, each time and yeah. We have to cluster a premium, listen that you can automatically change the rest before the mask master, no past components and then for roads. It's secured communication between the bootstrapping and cluster. This was all UPN, the bootstrapping class and the class. Our communication.

A

Yeah, the one master plane, muy, Grande, Tech and alerting exec all classic about it, so any more dreams. You can then see. Okay, if a class there's a problem, you can drill down to cluster to the HD and see when desti sorry to get in trouble and use management is pretty new. So also you can act in over text users and roles to clusters. So it's like our part and will be deployed in the user cluster and it's not really official release, but it's kind of 90 percent ready. Our alphabet are between.

C

A

Is you see it already? So it's just to me not say Protectorate e, so but yeah. This will be the next step that you can say: okay, whose last user, who is cluster up mean who can maybe not delete machines and who can do it and yeah so be pleased soon, yeah, and if you want to try it, maybe they are not all features, reachable like monitoring, logging and so on, but you can use it in May around I use it for my own.

A

If I have to do something with containers on my private I use, my private tutoring account and create classes there and yeah then delete it. Firstly, please ask them to get a little machine and.

A

A

A

A

A

So yeah already we are hiring.

A

If you are able to build your around.

C

Can I don't play with by myself and.

A

No kubernetes itself is a product, but you can do the machine control so as you can create machines and therefore you can just need to I'm not sure if to cluster definition as well as open source, but there are control plane and they're like pickup and all the stuff you have to fill by your own yeah.

B

Kamesh that you have existing clusters by 50.

B

When everybody.

C

Comes see you at the ocean.

A

So you mean like.

A

If you have already infrastructure of Phi of X clusters and want to add them to Kiba medic I.

A

Think we yeah I think we need I'm, not sure, but.

A

A

Think it's not a use case totally it our product, but it's just like a requirement from me. Writing it and I mean there would be available. Do it mostly. You start from crach and putting up new justice and maybe Federated trust us. A new author, Oh.

C

But you will have it, you have to do it one by one anyway, and it seems people have health infrastructure that usually indicates that the department are responsible for by death. Also, there is also foreign customs right now. The reason is because, if you're able important that existing clusters, you don't have to go to speak to so many people, you're, not.

B

Doing basically huge change for them like if you want to just use our information yeah, you.

B

A

Them diversification of image and you link them all together. You start singing.

A

Currently it's like the current state of, though at least class type, a itself not stable enough work, a state, but if you use our stuff, the use case mostly wasn't there already so, but there are all components, aren't just components like uh some resource definition. If you are able to to define your cluster in the meet or destroyed a level correctly and then connect it, it would be possible.

A

But if you have like, like the overlay network of we currently not supported, maybe then we have to may cannot deploy a new node or something that it so for this case, we have to ensure that your classes will be like kind of getting equal.

A

The control plane is in the user classes own so like here in the new create cluster. We are also deploy a control plane of command s itself, so I sing this.

A

Know so it's like ke to does it so there you also get electron and I see you can't go there over.

A

A

And twist your conflict.

A

Now, what's the right question.

C

A

This was the wrong idea: Benitez.

A

I'm not sure it's possible yeah, you might actually and I'm sure, cancel so issue.

A

A

Just can export to this one.

A

C

A

All right so yeah, the cluster we created, so it's the normal control plane which select greatest for chemical. When do we not it's all additional components? If you look at the components we install it here, you know took it here all in spaces,.

A

First column is that's our components for her chip system, your so having other clusters, and only thing like the cubed n has to be get information from. We can talk with them, but all the other components there's nothing like special chromatic friended. So you can also like choose what you are applying in the cluster. You have like an autonomous marionette, there's no resources like starch, so we know the source provider, so we can enable the default storage classes for a cloud provider and that we also implemented.

A

But if you want to use like something also, you can add like an add-on which will be applied after classes. Ready and install like the in the loop to castle and soul takes care. If you change the add-on, then the atoms will be applied.

A

A

More questions.

A

Okay, then I think we can stop for all. Thank you very much.