Cloud Native Computing Foundation Kubernetes Community Days Bangalore 2022, 11 May 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Workshop1 : Kubernetes 101

Description

In this workshop we'll be covering basics for Kubernetes. We'll start by learning its architecture and explore different building blocks like Pods, ReplicaSets, Deployment etc. We would then explore the Service object and see how we can expose our applications with different configurations.

A

So good morning, folks, uh thanks for joining in this is nipped, and I welcome you for the kcd bangalore's first workshop, which we are starting today and after this uh we have many other workshops lined up just kind of give you a kind of give you a complete flavor of uh the uh what you call say, the clown native stuff right. So we start with qualities 101.

A

Then we have life cycle management for cloud native apps, then argo rollouts will be ci cd thing then for kiosk engineering, then we'll have. How do you deploy sto just service mesh? For you, then we'll have a security workshops and then uh multi-cloud communities, lifestyle management, that's kind of a bit on advanced side, so female workshop may come in in between so that will update.

A

But the idea for these workshops in this order is that you'll have some time till the main event, and if you want to attend all of them, you can do it. So you can take this workshop practice come to the next one and so on. So that is the whole intention of doing this workshops over over the week over these four to eight six weeks, whatever we have uh remaining there so um for for this particular commodities, 101 workshop, uh as we know that right we are going to cover um this agenda today.

A

This is going to be very beginner focused. uh I just assume that you know a bit of containers and docker nothing about kubernetes, so commodity is going to start from very basic. um Let me just show off hands uh how many of you have worked on kubernetes um there.

A

So I mean not for documents docker, how many of you have worked on docker earlier, like there's, no one who's completely new to containers as well here. Well, let me ask other way: anybody completely new to containers here.

A

No right, please feel open. Please helper display! That's perfectly fine! Yeah.

B

C

That you can only have the.

B

Basic uh set up a docker and running a container come again, so I worked on basic. uh You know: setup like okay, okay, setting up a docker uh container, creating a container and helping the team. Okay, I have a.

A

Basic uh knowledge, that's, okay! That's! Okay! Basic is only what we are looking here, not, but so basically, what container is? Basically, you basically package your application and put them in a box right and then that particular box uh call. We call it as a container image and from that image we can create containers wherever we wish right. That's the whole idea that you take your application, binary with dependency put in a box or a package called image, and this runs same everywhere.

A

This is going to save time for everybody uh as we use them in environments, yeah, okay, so I'll touch upon briefly, I think in our uh qualities, 101 overview like when we start this the course there. So I just want to make sure that you have access this particular page here. I am putting on the chat again so this for this particular link.

A

If you open up, you can sign up for our portal, and this will give you the access to the workshop, which will contain which will have the content and as well as the lab environment as well. So if you open this link, you can sign up for free here.

A

You can click on this access course, and this will kind of bring you not like this right now. This will give you a button here of lab setup, so I'll talk about that briefly, so just want to make sure that first of all, all of you are signed in to the portal here. Let me know if you're facing any issues in signing up for the portal.

A

Done: okay, great now, let's keep it very interactive. If you have any questions, please feel free to kind of just ask away. Don't wait for me to kind of uh finish a topic and ask so whenever you feel like. Please just ask the question yeah, so, okay, anybody! I assume that nobody is waiting now so once you're in here uh this particular platform has the content. As you can see here, we already have a pre-recorded content with the hands-on lab. What we'll be doing along with there is the lab environment.

A

So currently I see here open id, but, as you just joined in, you might have a button of lab setup and you can choose the number of hours you want to have so please choose for for four hours or eight hours. You can do that and then do that so you'll be given overall 40 hours. I believe, and you can choose for eight hours right now or four hours. That would be okay, um so once you have basically uh trigger the lab.

A

So car thing you can do for the mobile as well our platform. Should we support that? Let me know if you have any issues there I mean you should do that because what's happening in our I'll, show you actually so once you are once the lab has been triggered right.

A

What you'll be getting is you'll be getting a terminal like this, which would be having the terminal on which you can work on and, as you can basically look at our uh course curriculum here right uh wherever, when the commands right, you can basically because the lab is being triggered here, you can access or run the commands, of course, by terminal like the way I am running any commands here right or I can basically.

A

There's a button in front of every command, you can click here and this would also run the command for you right. So you can either do by the terminal by the just hitting the command here like this, or there is an id also on which you can write ymils and so on. So so there are different ways to interact uh with the setup and hope that will also sort the problem for karthik as well. If you want to just do it on the mobile as well. Okay,.

D

So we'll we'll do.

C

With a very basic.

A

Stuff and then we'll jump into the other topic there right so before I start, I just want to reduce myself quickly and then we kind of uh get going, so my name is nikhindra and I'm founder of cloud digger uh before starting cloud in 2015 2016. I worked before that. I worked for over a decade in different companies, primarily with red hat, have experienced a support, engineering file system, kernel development and perfect degree.

A

In 2015 I wrote a book on docker and post that I realized that is going to change the way we develop and deploy the applications and because I love teaching, that's why I thought of kind of move on and just try out with cloud and see how I can do it. But since then it's been going uh very good, um so I kind of went back to the corporate job uh as of now so um I was just joined in so that I'm giving a link on the chat.

A

Please follow that uh signing for the quotes and so on. Yeah.

C

A

Yeah um then, in 2017-18 time frame I also authored a course on introduce kubernetes uh with linux foundation. That course is now taken by more than uh 200 000 users across the world. So if you go to cncf dot io there, they have a kind of a say public training on mit cdx platform.

A

So if you look at this course, this is the I authored course for the first time, then chris is kind of working on that course to modify it and make it up to date um and I'm also a cncf ambassador uh and and running this uh kcd bangalore event along with others as well. Okay, so that's a quick intro about myself, so just a whole session. What I just shared a link uh on the chat, I'm going to put it again.

A

So if you are just joining, you can do that so just sign for the course there now in the course once you come in uh so in this course, we already have uh pre-embedded slides content and the hands-on labs.

A

So what we'll do is, if you just come in you, can trigger here the lab setup for eight hours or for four hours, and then it will kind of bring up the environment like this, on which we would be able to kind of run any commands, and of course, these commands can also be executed uh from the terminal itself or from the terminal as well as we can kind of do it from our uh platform itself. So, as you kind of look at there's, a part of execution is there.

A

So if you trigger the lab, then you should be able to run the commands just like this as well. Okay, so I think that that's good uh we'll now get going with our uh content. uh What we have come here for um so there is a quick video I have, which I would have you play a bit later other than write down. So this.

C

D

A

uh What is the why we need a current orchestrator? So if you have used docker earlier then with docker, you could run containers on a single machine, let's see if yeah. So let me just talk about the briefly: what containers are if that's? Okay, so traditionally uh we have been deploying uh our applications on physical hardware, so we used to have hardware os on that and then we deploy our applications directly on those on that os.

A

Now, if any of the application of fails right or have some kind of an issue, it can basically um make our entire os go, go bad right and my entire os can fall apart. So that's the problem. What we had traditional system, then of course, uh in the last two days I mean last 15 years, and so we kind of move toward categorize or virtualization there. So what we did basically on our os, we deployed a layer of hypervisor and with the hypervisor we could basically partition our underlying hardware into different chunks.

A

So, for example, I can say I want to have an instance of 2 gb of ram or 2 cpu system, and so on suppose you have a 64gb system overall hardware and then you can create these different vms with different sizes of cpu memory, and then you deploy your applications on top of that right, so guest os, it can be linux, windows, mac or whatever.

A

We deploy that any developer application. On top of that. But, as you look at here, our intention was to run the application. The intention was not to run the vm right now. What? If, if I take this application and run directly on the uh os, what I have in such a way, this app by itself behaves as an independent entity. For example, my app has its own ip address, its own host name, so it kind of behaves as an independent entity and that's what container is right.

A

So you take an application package and run on the system right now why we want to move containers or use containers see at a high level. There are only two two things which you want to do. You want to save time and money that is both our public or both our personal life.

A

And now you know what you can say office work as well right, because if we save time that's I mean all of those things matter right, so how we save time with containers, because um so let's say you have been in a traditional app deployment cycle. You've seen that right, dev q and operations right when developer develops. Something qa would need to kind of test it right.

A

But if the qa takes a lot of time, just set up it just waste of time for you, but when you go to containers the same app, what developed by developer would run on qa on on fraud environment, so your entire software life cycle is reduced and that's where we save time we save money because suppose you could run let's say 10 vms on hardware.

A

Now you can run 50 of containers, I'm just giving the number there 50 girls on the same machine, which is 50 apps. You can run that's going to save the money for you correct, so that's how the main reason why you move to containers. Of course, then you have immutability for you building image. All those things would come along with it now containers are the feature of the underlying os. It is not that docker has come somewhere on the fly and just built everything.

A

No, so our os have the feature of uh arc basically resource isolation, resource allocation, for example. I want to give every program it's an ip address, host name that we get from the feature of name spaces.

A

When I say I want to allocate a 2gb of ram to a program, the cpu to a program and so on right that covers the c group. So this is how the container features come into existence.

A

Now there are different ways by which you can connect to your os and request these correct, so docker you're, getting call is one of the way, but there are many other ways using which you can do that. That is. We call as a time so contain. Runtime is a way using which you basically connect with an underlying os and create the containers value right.

A

But then, when you create these containers, what we saw here this container is running only on a single host right, so this is running on a single host. Now, if one, if this host dies right for whatever reason, maybe kernel got corrected in power, go power went on and so on. Here all the applications are also gone correct.

A

That is the issue with a single load, multi container setup and so on, and that's where we need some kind of an orchestration where we can connect multiple nodes together and form a cluster so that we can deploy our desired scale make sense any.

D

Question so far.

A

I said that right, you don't need to kind of wait for the last moment whenever you have a question, please ask me.

B

What is uh hypervisor means.

A

Kvm virtualbox right, which kind of vmware right like vmware, acm site, that kind of emulates your hardware right, okay, but just hypervisor.

E

uh So one question here: can you hear me yeah very much indra nabindra, uh I mean I understand about the dockers, a donker rancher. They they have a facility that they can have multiple containers of multiple nodes. Am I right so why these are need of.

A

Just above it, so not they so by itself. If you just talk about docker right by itself, just a docker binary is not once you configure docker swarm, then they'll form a cluster. Similarly, with wrencher, a single node trencher is as what we're seeing a single node docker.

A

Only when you connect these multiple venture hosts, they form a cluster yeah.

E

Sir, if a docker spam was very enriched in their functionality, what was a need of bringing the kubernetes in the in the world.

A

Yeah sure I'll answer that question in a few minutes I'll just give you a few minutes I'll answer that yeah sure uh any other questions.

A

Okay now I said that right, if you have these containers on a single host and if the host dies, everything just go haywire right, that's where we need some kind of an orchestrator where I can connect these multiple nodes, so those nodes can be vms physical machines or any computer in general, and if we form a cluster to run our cutters at scale so basically think about. We take four vms.

A

We put them together with the help of software like kubernetes, and then we deploy our apps on that. By doing that, we are going to get the features which are listed here on the screen, so some of them not all of them, but there's all the some of these things. What we have uh listed here basically go over there, what I mean mean by each one of them right, first of all, container provisioning. So when you deploy a cluster, you don't talk to us. Individual nodes say that okay deploy my applications right.

A

uh So let me simply actually draw it that would kind of picturize it in a better way, so assume that I have these.

A

D

A

Right and I was basically going to connect these nodes together and form a cluster so now in this cluster, one of like and just give an example here.

A

One of the nodes is the control plane node, while other nodes are your worker machines right, so I'll talk about what what control load and so on in a few minutes, but for now let's say this is a control, machine, control, plane and others are your nodes like a typical worker nodes for deploying the applications right, so what we have taken is we have taken these multiple nodes together and form a cluster correct.

A

Now, here um let me go back yeah, so I talked about container provisioning first of all right, so what I'm saying is I'm not going to communicate to the individual node, deploy the application node out? Basically, so if I am a end user who deploys the application correct, I would basically send my request to the control, plane, node and this control plane node would be deploying my applications on behalf of me on different nodes there. Okay, second thing is fault tolerance.

A

Now, if one of the node dies in that cluster automatically the application running on that node would move somewhere else. What I'm saying is if this node dies automatically the apps running on that move, node somewhere moves somewhere. That is a node level fault of the lens. Similarly, for single app has that, for whatever reason the a replacement copy of that would come up as well. That is the um the app level fault tolerance. What you have, then we can do scaling, which can be both automatic and manual.

A

So when you want to deploy applications at a scale right, your traffic is coming up more and you want to scale it up so now there either you can do manual scaling like manually, say: okay, now my load is coming up. I scale up or you can feel some kind of automation right when you say that okay, the request per second has gone beyond something. I want to scale it up right, so this scaling again, you can do both app level node level.

A

Then self discovery means that, because, when deploy these applications right, these applications will be deployed in a as a group or as a multiple replication right you don't deploy.

A

uh I mean you deploy the single replica as well, but in when you scale it up, you're, going to run multiple copies of the same application on different machines right or anywhere in the cluster. So how would you kind of refer them by a common name right? So, for example, if you open a website as a google.com right, google would be getting served by multiple vms or nodes behind the scene, but for you it's just a google.com now. Similarly, here we need somewhere to deploy multiple replicas of our applications.

A

I want somebody to refer them, refer them as a common name, so that is what's our discoveries again, we'll dig deep into it? Then we have a different strategy of deployment. So when you move your app from one to another version, right today are running version. V1 up you want to move to v2, then you want to say that okay, I would want to move to v2 only if p2 is 20 faster or I want to update um on a rolling basis.

A

I suppose I'm running hundred copies of my applications right and I can't just bring down all 100 at once and then create new ones right. I want to do a rolling update. I would bring down few of them up and then bring new one up and so on. So this is again a feature which you'll get with orchestration.

A

Then you have based on story. Orchestration means that when you, so when you create these applications or containers right, these are temporary right because they'll not have any volume or any content right which is um store for them permanently. So I need some kind of a volume management so that my data is stored outside and if my container moves from one root, another node, I would move my data as well.

A

So what I am trying to convey here suppose we have a storage which is a whatever some public storage you have and currently the storage attacks. This particular container, for example, right now, this container dies here and comes on other machines right, so I want to make sure that this storage is moving along with me. That is uh what I mean by the isla orchestration, so I need all of these features to be in production.

A

Correct, uh any comments here or any questions.

A

All okay, okay, so now there are different softwares, which are which can do uh the audition for you. These are your docker swarm, kubernetes, amazon, asia, ecs or hashicorp nomad. So these are different, tooling, build which have built around to kind of give you all those features.

A

But of course they have some distinguished feature, which is some of them are just for themselves itself or just for unique for themselves. Right, for example, amazon ecs would only work on amazon right. This is good for the amazon ecosystem, but it is not good for me and you who don't want to go on amazon ecosystem, for example.

A

Right so amazon is ecs is built for um amazon, specifically right, docker swarm was built on top of docker, but it lacks the feature which was needed to do a real broad environments, and then there were a lot of issues in between so there's the concept of commodities, part which was very fundamental, which I'm going to explain that changes the the way you deploy the applications on on the cluster right. So that is what kubernetes got it, but not darker.

A

So so those kind of things which we look at in detail right, you will feel that docker swamp lacked those kind of features and understanding which was needed for production environment and that's where docker form did not get enough attention while hashicob nomad is a really good product from the hashtag team, and this is going to do a very well for our regular app deployments and upscaling and so on. But where kubernetes wins on top of nomad is that I can extend nomad or sorry. I can extend kubernetes the way I would wish to so.

A

There are ways by which you can extend kubernetes for your own work for your customization right, so that extensibility is at a great great level in kubernetes and, of course, that kind of made it very much um and, of course, a popular and adoptable to different platforms.

A

So that's where kubernetes kind of won the war there, and then it is now the leading orchestration engine for you correct, also the the project by itself right, it has come from google, so google has been running containers from 2008 onwards and they have been using internally called borg. So borg is their project, which was they run internally and they had experience what it meant to run cadets at scale and in production, so when they saw that that uh docker is becoming popular for app deployment purpose.

A

The next thing is to deploy on production and docker was at that point of time, was not ready for or not was built for the orchestration engine, because docker sum has not come by then.

A

So that's when the google open source support called kubernetes and some of the google engineers a founder project and then com companies like red hat, ibm and so on start contributing the project and eventually what happened? Is this project, along with the trademark, a change to or ownership change, thus community?

A

uh It has come under the linux foundation. So, let's foundation, I believe all of you know whatever foundation is. It is like the foundation which kind of governs the roadmap of linux, down the line and so on right, so future present in what is comes under linux foundation umbrella under that the cncf foundation was found, and this kind of now owns the the officially the kubernetes project by itself.

A

So if you look at the cncf first of all, there are many members with these foundations, which kind of help to basically run this particular governments of this this foundation here. So there are different members like platinum, gold and silver members, and we are also the cylinder member on this foundation.

A

Now this foundation mix gives the common ground to the different projects which are in the ecosystem. So, for example, let's say you build something some project for yourself in the company for let's say, which kind of makes database running very easy on kubernetes, for example, and you want it to be adoptable by others as well, but as an individual as a company, you might find it difficult to make it uh marketable and adopt to basically find a valid option there. So what happens?

A

You can come to the cnc foundation and come to the sandbox stage, if you like they like it. If the community likes it, then you kind of go to the stage of incubation and graduated. So this is how the project kind of moves different categories so that they kind of get adopted in different levels right. So if something is graduated, which means it is like well adopted, and you can scale it up as as much as you can then csgo give different trainings and all that stuff as well, so that I'll just skip here.

A

um That's all I talked about yeah, so commodities can run everywhere, be it on your on a laptop, be it on a public cloud, private cloud, so commodities can be run everywhere because the weight has been designed. So it has something called as an api. So we have the kubernetes api server which I'll talk about later, so we communicate to this api server to perform any operations and that api is going to be common, irrespective of which environment you are in correct, which means you would not be bound to a given cloud to wider.

A

For example, let's say you are, you were on aws water right, you know aws and you built everything and around database, but for whatever you want to move to. Let's say azure right now, typically, because if you are bound by completely on aws, it's very difficult to move from one to another cloud. But if you are on top of um let's say, kubernetes right, it's going to be same everywhere, so migration and everything would become very easy.

A

So it kind of becomes a neutral thing where you can easily um migrate between the clouds or you can have set up with multi-cloud and all that stuff right. That's where kubernetes shines there, so it can run everywhere.

A

It is very stable at course, so communities is very stable, very minimal in the at the core, but you can also extend it for whatever your needs are so like different companies are publishing their projects for their uh software on top of communities uh for communities as a the cids, which you call them to extend commodities for the requirement of the specific requirement there.

A

It is very versatile for hybrid setup, so, for example, if you have on-prem set up and on cloud setup because again the apis are same, you can have the private setup with less number of nodes. Only on demand. You can go in the cloud. So that's where you can save very good amount of money. If you go to hybrid setup.

A

This is what a high level is. Why kubernetes is any questions here.

A

Okay, just want to check um anybody having an issue with the left setup, no right.

B

Like in a single mission, we are able to virtualize and like using um like, like hypervisor or like different tools to get connected about, like, uh like a device, has an ip right. Like I mean how does uh that networking happens like how does that we are able to provide public ips for different vms? Like can you just touch base on that public.

A

So see if again, they are different, so that different ways to give vms public ips or we can give by the node as well. So if you, if you go talk about vmware or any other thing right, there are tools available using which you can. First of all, you can attach a complete network to a vm instance right that is possibility. Is there correct and then, of course, just the network layer right? You can take it out. So there are different specialized. Hardwares are available.

A

That can uh basically give your vms a public ip as well. So those are, I think, pretty much well defined. Things are there. So those are I mean the solves all problem in the vm in the in the world of virtualization. There.

D

B

Okay, okay, good so another one like you were mentioning about the different orchestrators right like can. I also come up with this mesosphere so.

A

Measures was there now. If you look at, they have moved to kubernetes as well. So of course, they were very good at a time frame, but now they kind of they move to the different things. Yeah.

D

Okay, yeah: okay, thanks.

E

Stop in there sorry one question here.

C

E

It was a very great uh great question which karthik bring in so when he is mentioning about how the hypervisor vm gonna have an access to the node or the cluster. Are we referring here to the kubernetes cni, which is the network plugin, or are we referring to the default network which has been set up when we by default, have a docker been installed.

A

So there are two things he asked right. First of all, I believe we just asked about how would vms get the ip address right, so the so vms are independent of containers right. You install vmware, and you deploy vms on that right, yes or you deploy kvm and deploy vms on that. So for that purpose, as I said that either you say the nsx right kind of solution there right so that gives the ip address to your vms, which came publicly under seven correct.

A

Yes, sir, I'm gonna talk about commodities terms, we'll deploy these apps, we'll call them pods and for parts I'll touch upon that, how we can uh do the networking part uh for that. So that's a different story, so that's a cni way and so okay, let me go back here so now. Here we have different containers running on different machines.

A

Now these containers within the cluster they'll have a private ip and they are addressable within the network itself correct and that's where we will talk about cni the cni help in that scenario, where you are connecting the containers of different host correct. So so that is two different things. One is talking about vms talking about containers, so in containers uh in a very minimal setup or a very basic requirement setup.

A

You need to have some kind of a plug-in using which I can connect the containers of different hosts that we call it as cra, plug-ins and they're different drivers like calico, 11 and so on, which would be there. Okay, I think I'll come back to that questions or answer in a detail as after I talked about few more basic stuff there. Thank.

E

A

Okay, okay, um all good um okay, so we'll get going uh so now we'll touch upon the kuwait's architecture here, okay! So now um again I said that right we need an orchestrator which will need one or multiple nodes together to form a cluster.

A

So here we'll take an example of there are, let's say, four vms or four physical machines, which we basically consider as our nodes. Right now with those nodes, I am just going to create a cluster now in this particular environment, one of the node is a control plane node, while remaining three are your worker machines worker nodes? Are there.

A

On the control plane node, we have components like api server, controller scheduler and the college key value. So I'll talk about that. This is there to store your cluster state on every node. We have a cubelet q proxies and few more stuff in as we look them later, as a cluster administrator or as a app deployer, I'm going to communicate to my control plane to say that okay deploy the applications for me and so on.

A

So let's look at the workflow here so suppose I am a app deployer and I say that go ahead and deploy an application for me so deploy let's say three copies of an app for me. That's the request sent by my app um deployer, the application request, comes in to the control plane node by the api server. So, first of all, if you look at here in the api server, it is the main communication agent with everybody right as an end user. I am talking to the ap server with the control plane.

A

I talk to the scheduler controller with the ap server key value store api server, every node api server. Cp server is the main communication agent for you.

C

A

So we are sending a request to deploy three copies of an app for me. That request has come to the ap server. Now I need to somewhere store that information in persistent way, and for that purpose we have something called as key value store. So key value store is a place where you store your cluster state right or what is the desired state?

A

You want to be in that's getting stored in the key value store, which is like a brain of a system, and this key value store can be part of the cluster or can be out of the cluster, so it can be anywhere as long as my api server can communicate to it. Okay. So this is the place where I'm saying that okay, I got a request to deploy three copies of a given app.

A

Now that request would be set to a scheduler where it would tell you where those apps need to be deployed, because scheduler knows how many nodes are in the cluster. What are the configuration? What apps deployed earlier? So it has a control of what kind of knowledge about the resource, consumption and utilization there. So, based on that again, you'll send the request to the ap server. Api server would then say, go and deploy that this app on a given node.

A

So assuming that we said that three copies of my app, then we got a request to the ap server scheduler and we said now deploy one copy on, let's say: node one and two copies of node two, for example, so ap server would communicate the qubit of individual nodes and say deploy one copy here and two copy of the apps. Here now, once my app get deployed, I would need to make sure that they are running all the time correct.

A

So, for example, if one of the app dies for whatever reason, then I need to kind of bring it one more up right, because my desired state was three, I said, deployed three copies for me in the beginning that has been saved now now the controller is a one which is responsible to make sure that at any point of time, my desired and the current states are matching okay.

A

So if I said that okay deploy two copies for me correct um or deploy three copies of the app for me right, I want to make sure that at any point of time, all three copies are running.

A

That is what controller make sure controller basically make sure that at any point of time, your entire clusters desired state and current states are matching. Otherwise it will try to bring into the state which is desired state there. So controller is make sure that your all the apps are up and running correct.

A

um Then, once my app get deployed, then we have these q proxies. This is going to allow the excellent traffic to come inside the cluster. That is what q proxy is going to help me do from where I get the traffic from the excel one to the internal, an entire cluster. Here it does more job which I'll explain it later, but for now q proxy is there to get the excel traffic come inside the cluster.

A

Any questions so far.

F

um I have a question on the um so.

A

F

You like, for example, you mentioned that I need to deploy three apps and uh one app has failed or died uh that particular information will be communicated to the api server and then that will be communicated to the controller because controller doesn't know anything what is happening with the app so how and once yeah so that part of communication, if you can just explain a little bit.

A

Yeah I just kind of simplified a bit so first of all, there would be some kind of a health check that would be happening in between so if, for example, if a container fails just for the health reason right like um so within the app there is a health check. That is what cubelet will take care of it, that okay, um that uh one of the app is failing. Let me try to restart it here, so that is going to happen within the node.

C

A

Right but of course when the given application has died completely, for example, if this node has gone complete out right suppose this node has a power.

B

A

Correct so on this node, you would be running a few applications which I know from the ap so from the key values to that. What apps are running on that particular node right and those kind of information, along with the controller, um doing the check whether the right number of apps are running or not again, this check again going to happen by the ap server.

A

Only whether the right number of apps are running at any point of time or not, and if they're not matching correct, then again they present talk to the scheduler again saying that deploy the applications again to make sure that they are coming in the right state there. So this is basically in a way that will just check whether the states are matching or not. If they cannot do more than that for the remaining stuff, we'll agree to kind of go to the same state of ab7 scheduler. There make sense.

F

Yeah yeah and one more thing that the key value store will all will only be updated once when you deploy the app and if the request doesn't come like okay, fine. Now I want to deploy two more apps, then the key value store will always remain three and and at any point in time. If any application fails, then the api server will keep deploying the the.

D

F

To make sure that the match the the numbers match correct.

A

Correct so key value store, just stores the cluster state, what desired set you want to be, and among with many other cluster state like what are the app configuration there? What are the secrets I have and so on? That is also go in the key value store, so this key value store is part of the cluster or it can be outside so like here we are talking about the software called etcd at cd, which we are, which we typically deploy with kubernetes to enable the key value store.

A

But can we end the software as well as long as my ap server communicate to uh over http and do get put kind of those kind of operations there so here in this key value, store uh we'll be having one leader and multiple followers, so, of course I can have a single node as well of the key values tool, but there is not for product environment in prod environment.

A

I would have multiple replicas of my key value store running, so we have uh like in this case we see, there's a five node uh setup of the key value store. There is a one leader using which I would do all the writes. The read can happen from anywhere.

A

So if I have multiple nodes in my key value store like in this case, so in this case I can lose up to two nodes and I still have the quorum right so, depending on how much high availability you would need, you would have three or five nodes kind of thing in your cluster correct.

A

E

uh It's gonna be very lame question, sir, when I'm talking about key value, stroke, I'm being referring to my atc. So can I have adcd on the on on on a physical machine rather than.

A

Having yeah yeah, so I don't have diagram here, but I'll show you from one of the hands-on lab what we have.

A

Yeah, so you basically deploy the the key value store, which can be part of the node itself like there. We have a third cluster and my key value store can be part of my nodes right or these can be outside. I said that right, my key value store can be inside or also of my nodes, so I can choose to deploy the way I would wish and in prod environment we have seen that we choose to deploy it outside of the cluster nodes.

E

But it's just. I need to make sure that apis have been happening.

A

Yes, you need to sure that the ap need to communicate with the key values. That's the main part.

E

Answer same for the worker nodes, can I have my worker nodes uh rather than having in a cloud on my physical machine, but actually on the same subnet yeah you can you? Can you.

A

Yeah, so let's go back here. So let's talk about few more things here now here I'm talking about. There is a a single control, plane node, but as I showed in the other diagram right, even in that case, I can have my multiple control plane nodes right so rather than just having a single node, where only my one of my node um is there to control the cluster, I have highly availability of my multi, my control player node as well. That would be much better for me for prod environments.

A

Similarly, um now let's say even a single node cluster, let's say if one of my node goes down or not not the worker node, let's say even in this case, my master node goes down. What's going to happen, okay, even even in this case, if my master goes down, my customers are not going to have a downtime, so customer directly do the my application access via the node itself, so customers are not accessing my website why the control plane don't know my apps are deployed on my nodes and q.

A

Proxies are basically allowing me to get the traffic from the access to internal world and there are many other ways as well, so when my customer, as my website, they access via the node itself. Okay, so if master orders goes down for a while, I'm perfectly fine, I don't have a downtime.

A

I just need to make sure that I bring down bring up my masternode again what I cannot do when my master is not available. I cannot change my cluster state. I cannot deploy more apps and so on, but whatever is running, I'm still, okay for a while, okay.

E

uh One question, sir sorry: it came to me for interview, so if my master has been down, definitely the provisioning will not happen. Correct. Even the fault tolerance will not have been not been happening. Sir fault.

A

Tolerance for the nodes right- yes, sir yeah that's going to be affected. Yes,.

E

But my existing uh virtual machine still be connected. I've already have that, but orchestration will be impacted for the for the, for the mean time so correct yeah. Thank you, sir.

C

F

uh I have a quick uh just quick question like so: you mentioned about the high availability right, so if your master node is down, I mean we still will have a uh the replicas of that node right. So that will take that will take the controls.

A

If you have, if you have a control, plane, high level mode configuration yes, then there's the of course. If.

F

You, which, which will be the case in any scenario right I mean even in a production environment, there won't be one master node, so I'm assuming until all the master nodes are down. You will still have. That is the worst case scenario where you will not have any provisioning done, because all all your master nodes are down, but.

C

F

Cases there will be a way to provision through the the other master node. No.

A

You so you need a quorum there right so so, for example, see um depending on how you have configured your key value store. Sorry, let me just go back here see if your key value store is part of the nodes right, then you would need at least two of them to be up to make sure your std is in a high level state. Are you getting a point correct because you cannot have the std running in silos?

A

They have to at least two of them have to be in quorum to make sure that I can write something correct this in this case, I would need minimum two of them to be a balance, while in this case where I have the cats xcd outfit, my cluster right, even if one of the node is there, I'm perfectly fine, correct hope that makes sense. So it depends on how do you configure there? Of course, many things will go down the line, as you kind of start, configuring, okay,.

E

Then the benders are uh as in uh one understanding, so when we create a gk cluster or when we create a uh aksui case cluster, we only get an access to the worker nodes. Okay, we never know where the master has been there correct. So how? What? What uh platform.

A

They'll, basically, they basically take the responsibility of that right. They would have configured highly available uh control, play notes behind the scene right and they will also take the backup of xc to the regular interval. So even if node goes down altogether, I have actually become. I can replay it correct.

A

So there are, of course, different things to do in the back end, but at the same time, because they are charging, you they'll take care of the things by even having multiple nodes, or do they take a backup of xcd which they can replay it on the fly and so on because see my state of the customer. The key values too right, if I have the backup of that, I'm perfectly fine.

E

I mean um that's what the reason is that taking the back of that cd is really difficult. I tried two times it really failed for me.

A

No, it's not difficult just that. You just need to follow the procedure right yeah, that's it! Thank you, sir okay. I think we'll move on now, um so so yeah so discussing about so, let's kind of dig deep further into each of these. So talk about ap, surviving to cover most of the stuff here, apa server also handles who can login and what they can do in the cluster. That is also part of the api server here.

A

Key value store stores the cluster state for me and primarily will be using http for that, but it can be any other tool as long as ap software communicate to them controller. Make sure that at any point of time, states are matching and there are built-in controllers into it. So it's like single binary, but it has multiple controllers built into it, which are responsible for doing the node management. Like I have the right number of nodes or not, I have the right number of apps running or not right.

A

I have the right end points for my services, so these things are part of model to single binary, but at a high level it's making sure that at any point of time your states of the clusters are matching.

A

Then your scheduler, which schedules our applications uh based on some resource constraints, whatever you may have given. So, for example, um um I want to deploy my application with minimum 2 gb of ram or deploy my node, where I have a hardware of gpu. So all those things are can be configured as a as a requirement and my schedule will try to fulfill them to deploy my applications on the respective nodes.

A

So when I do the scheduling I do two things there, I first find the feasible nodes and do a scoring of that. So what I mean by feasible nodes, suppose you have a cluster of 20 nodes. 10 of them are windows. 10 of them are linux.

A

So when you deploy your linux app, you want to deploy on linux machines right. So you would say I would find first feasible, node, so I'll say out of 20. 10 are feasible nodes, then I'll. Do a scoring of individual nodes or behind the scene for me, but best score node would be used for my scheduling purpose. This is how the we'll schedule our application on the cluster, any questions.

A

Okay, next is. We would then talk about a bit on the nodes here, so here um on every node we have a component called cubelet, uh which is talking to my apa server. We have the queue proxy which allows me the traffic to come inside the cluster and one more java will do I'll talk about later.

A

Then we have a runtime on individual nodes, so we would run containers on each node right. So for that purpose you would require a runtime which can be like container d, shear hyphen kata containers docker.

A

There are many more so you deploy a container runtime on every node and then we deploy our applications on that and in kubernetes the apps get deployed in terms of pod, so the minimum deployable unit com component is called pod and pod is the collection of one or more containers so like, as you can see here in this, we have a pod a single container or we have a pod with multiple containers. Okay, so when pod get deployed, they get deployed or it get deployed completely on a given node.

A

So you cannot say: half of the part goes on node. One half port goes another node, a part, get deployed completely on a given node. Okay, let's understand bit more on this odd terminology here, because very important to understand so we'll deploy the part I mentioned. That board is a collection of one or more containers right. So, if I talk about the part here, so let's talk about this is the this is my part here. In this part, I have a container one and container two.

A

So let me just do that container, c1 container c2 right now.

A

In a part, these containers shares the same network namespace, which means they are going to be a single ip per pod right, so suppose I give an ip here, 192 168, whatever this is my ip to my pod here now, depending on the port number I am listening or my cutter is listening or suppose container c1 is serving on port 80.

A

While c2 server port 8080, so if I hit the pods ip on port 80, the c1 responds back and if I hit the pods ip1 port 8080 c2 response time. Okay, so this is how it is. So you have a pod in which you can have one or more containers. If you have multiple containers. They'll share the same network name, space between.

C

A

They'll communicate over localhost, so, for example, if c1 want to talk to c2, it says c1, localhost, 8080 right and vice versa right. This is how they are going to communicate between uh between the containers of the same host.

A

Right hope. This makes sense so assuming that, if I ask you to deploy a container of wordpress right, if I ask you deploy mysql in this and wordpress here, assume that right now this is definitely possible right. I can deploy.

A

I have a part in this part. I can deploy a container of mysql and a container of wordpress right. My question to you is: would this be a good approach or not answer on the chat? Yes or no on the chat? If I deploy, if I have a part which can have multiple containers, I would deploy mysql in wordpress together. My question to you is: would it be a good approach or not?

A

Yes, when pete says yes what about others? Okay, assuming that remaining all of saying? Yes, anybody for no okay, so this is possible, but not a good approach. It is not a good approach. Why see, for example, here when I am deploying these applications right, I have picked up my wordpress and deployed like this now suppose this. If I just want to scale my front end right by virtue of having in the pod right, I cannot independently scale the front end itself.

A

I have to scale both back and forth and here, which is a challenging for me right, because I don't want I what my need was to scale only front end, because now, because we have club them together, they have become a unit right. So, if I deploy, if I scale I scale in terms of modern kubernetes right, which means unnecessarily, I have scaled my database, which I didn't intend to do it right.

A

So this is not a good approach to deploy to these kind of apps together in the same part, okay, it's not a good approach right.

E

uh I mean very good example, but I am not little clear about it when you said we can proceed with the solution, but it's not a best practice. Yes,.

A

E

I right- and you said something after that- that we go uh go ahead and do a scaling on the pod level.

C

E

That's something which I'm missing, that's something I need to do.

A

Is a basic deployment unit correct? So when you work on kubernetes, the pod is a unit that you work in right. Yes, sir, so you have to scale by the pod basis right so because here your pod contains two containers right both have to be scaled together or both you get a copy of the pod itself. Right. Yes,.

E

Sir got it high horizontal scaling and very uh vertical scaling concept.

A

Whatever it is right, because power is a unit right.

C

A

Scale then you get both things right. It's a unit for you.

E

So you can't go ahead and uh scale the container individually you're going to go ahead and scale. The all.

A

We can do do that if there are different parts right. That's why, when you deploy these micro services right, that's why we're going to partition them in a way that they are complete by themselves correct. So here they are to independent components, right, better, deploy them into different uh two different ways: okay,.

E

So in case, let me put an example in case we have radius, which has been here.

A

Okay, I got a question. Let me just answer that. I know that what you're trying to ask so uh sorry to kind of cut you short here. I just want to kind of make sure that we have a good momentum here to finish on time. Don't.

E

A

Now this is not a good approach right then. What is a good approach so currently what we saw here? It is not a good approach of deploying a multi containment pod. Then what is the good approach, because a part can have one or more containers. Now. Let's say you have a use case here so again, let me deploy a pod app here. Suppose again, you are deploying the pod here now in this part, let's say you have an app, which is, let's say, a web server called nginx correct.

A

This is currently serving some html pages, which are part of your. This is some folder here right.

A

This is serving your html content and it is getting served to your audience whatever right, it's perfectly fine, but now I tell you that I want you to sync up this html content in every half an hour right now. How would you achieve this right now? One way to achieve this is you modify the docker file of nginx and build a image again, which would hard code, this mapping for you that every half an hour's income, but that's not going to be very good solution, because what?

A

If I change you a different location or different time frame, and so on? So what I'll do here is. I would basically bring up one more container here, correct and that container is responsible for doing the sinking for me correct this sinks, every half an hour for me correct. This is what we refer as a something called a side car container which are doing some complimentary job for you, okay, this is one primary app. There's a sidecar, controversial, complimentary traffic.

A

Other thing can be you kind of write logs here right. These logs would be sent by the logs of block by the side of the log server correct. So this is a use case of a multi-container pod, where you are doing a complimentary job may be the caching you are talking about right. You want to put a redis cache that can be a sidecar correct, so those kind of things are based, so they are not an independent app. They are somehow complementing my work or job. For me, correct.

A

That is the right use case for a multi-container part here.

E

But an ideal form in a production environment rather than this sidecar uh requirement. We don't use a multi multi containers yeah, but we don't have to unless.

A

Needed right I mean.

E

That's comes in an idea that there's no point of having micro services. Am I right, sir.

A

E

If that comes in an idea, I mean except side cut if we are putting all uh if you're putting the previous example. There's no point right here. Yes, sir.

A

Okay, so that's that's there! So now this parts configuration file, I can give it to cubelet in a json or a vml format, so I can give that uh so that way, tubelet would run them. So I can give that just a ml file by an api call via file location or an shdp endpoint, or why an xcd box the different ways I can give this yml file for um to my ap, to my queue to deploy the application. For me.

A

Okay and of course I can have a run time of my own choice. All I talked about this part. What we covered just now, so q proxies does two job for me. One for getting the cell traffic come inside the cluster and there's one more job where it would talk from it will configure my part communication. So I'll talk about that uh in some time. So, just for now, q proxy is for accelerated communication.

A

Then, once you have the base commands installed, you deploy some additional apps on top of it does not come on its own. Like dashboard, login server metrics, it does not come on its own. It would be coming. um uh You have to install it separately, okay, okay, so with that we'll move on to the.

A

How do we access the kubernetes cluster? So if you have triggered the lab already, uh what happened is you would have already got a single load cluster and if you do a queue shell get nodes here, you will see that it is configured a single node cluster for you, which is both behaving as a worker and master free. So this would be sufficient for doing our hands-on labs um yeah.

A

So now what's happening here is once the class has been configured because I'm not going how to be configured and so on, because that's a one-to-one thing assuming that customer has been configured for you right now. What you have here is when you configure the cluster, you have a file called cubeconfix. If I have to show you so once the customer's been configured by default at the location of dot, cube conf in your home directory dot, cube config, you will find the configuration file.

A

This configuration this config file says how we are going to connect to your cluster. So, assuming that you have configured cluster on cloud, aws, azure or wherever it is right from a laptop, I want to connect to it. You would require this cube config file on your laptop now by default, I'm looking at this location, but this can be any other location as well. So, for example, if I let's say let's say if I move it from here, so dot q configure 7 tnp now now my default file is not there right.

A

So if I do cube, ctl get nodes right, it is not going to work correct because the config file is not there. So cubectl is your cli program using which you would connect to your kubernetes cluster correct, or we can talk to the aps, we'll talk about how we can talk to the api as well, but this is a cli program. This requires this cube config file for you to connect to a given cluster.

A

Now because I have moved the configuration file somewhere else, I cannot connect to it, but at the same time, if I do export cube, config is an environment variable and point it to the location where the file is. It will again start working because I am now telling specifically where my q config file is correct, so either you put on the default location or you copy it anywhere else. As long as you have had the environment variable, we.

D

A

Make sense: okay, now in this cube, config file. If you look at, there are few different sections. What we have the first is called clusters. Then we have users, then we have context now in all these sections. You look at this is kind of an array or a map right, it's kind of a list here, so you can see here I can have. I could have listed here, multiple clusters as well so cluster. Let's say I have a cluster of dev cluster of qa and so on right, so I can list every cluster location right.

A

This is an ap server location, so every server has an ips server and this my ap server location is there, so I would mention like dev cluster queue, customer and so on. Similarly, I can have multiple users.

A

Then, once I have multiple cluster multiple users, I connect them with the help of context. Okay, so context is a way using which you connect a user to a cluster and then again you can have multiple context which one to use now that depends on current context,.

C

A

Is that make sense or confused of it? So I can have multiple clusters I'm talking about. I can have multiple clusters, then I can have multiple users.

D

Correct sorry,.

A

D

Have them multiple users.

A

Correct I can connect these user to a cluster of the context and if I have multiple context, I'll choose the current context to connect. It makes sense any questions here.

A

Okay hope all if you're clear, okay, uh then next thing you want to talk about here is something called as um cubes little proxy. So so I said that right, the ap server. So when we connect to our kubernetes cluster, we connect to the api server correct.

A

What I'll do is that is going to click on open id here? This will go open an id for me so where I can have partitioned terminal here. Just for it is here.

A

So what I'm doing now is, I am running a command called cubectl proxy. Now now this command. I am running on my client side, where I have the config file. So if I run this command, what happens now is um this cube. Ctl creates a secure or a proxy between my workstation and my api server. Okay, it is creating a proxy between my local environment and the cluster, which means, if I access this url correct.

A

This is actually giving me the axis so not cat, sorry curl here.

A

This is actually giving me the access to the uh what is happening.

A

Okay, sorry, okay. My config file is not found in s5, okay, because I have uh moved it right, yeah, that's what I copied it, so I just run this proxy command here or again. So this is a tunnel between my version, my cluster and now, if I do a curl on this right, you can see.

C

A

This is giving me the entire api server endpoints right so because I'm going to make a request to my api server right. These are my different end points to which I can connect to correct. So I have like slash api, then slash apis then come down. I have slash health and many others as well. Right now excited one more time here, I'll go to the next section here.

A

So what view you see there is like a like list view. There's a tree view here what I have so here we have this slash api, which we saw there slash apis. Then we have metrics health and so on this we refer as a core api group. So slash api is being referred as a core api group. Then we have slash even other api groups right now. Under these api groups, we have versioning here, so we can have, let's say: uh ap core api group. V1 v1 means stable version.

A

Then in some cases we could have a beta as well or alpha as well. So because we would be building these features, they would be available in stable. Alpha beta end points. So if you want to connect offer topics of a source beta feature will hit the beta endpoint correct.

A

This is how they are being designed right that you have one api group there you have versioning and you can choose which one you want to connect with. So let's say: if I want to work on a pod object correct, I would hit the core api group version v1 and the pod object.

A

If I want to work on a deployment, then slash apis, then again I have a subgroup called apps. Then I have versioning here, v1 only v1. I have now demonstrate deployment and so on correct. So so there are predefined location or predefined endpoints to which you can connect to work on a given object. That is how they have been working. Okay.

A

So if I look at the, if you just search for kubernetes api reference, documentation right.

A

You will find the you find the entire api documentation of of kubernetes there, and you can see here, like part, comes in the core api group. Job comes with the bash api group, so this is a predefined location where a respective object is there.

A

So now, if I have to create a pod object right, if I have to create a pod object, then I would hit the core api group. So when you say that I want to work on a pod object, I need to first come here then here and then here correct is my tree view right. It's my complete view. So when I say api version, this is my I'm writing a vml file. Here it can json file as well to create my pod object right now to create a pod object.

A

uh I need to hit my core api group so when I say api version v1 only right, which means I am mentioning here. This is like a shortcut or whatever you call it so apa version. V1 means you are already here. Okay, now, if you are here, you can work on parts nodes and so on. Then we will say something called is kind.

A

I want to work on a spot object, correct so I'll, say, kind part, then, in the field of metadata I'll give a name to my part, called mypod I'll talk about nexus in a bit.

A

So just talk about for a name for my ipod here, then we have a spec, so spec is means you are define a desired state of an object. So you are saying that I want to have a part, and in this part I want to create this containers or these containers here again against an array here. I can have multiple containers here correct, so I want to hit the object of. I want to create a pod object with this configuration here.

C

A

C

Questions here.

A

So, let's create a part if there are no questions here, okay, fine! So let's try to uh now. If you kind of go to this part section here so we'll find this uh pod dot yml, which I was kind of referring here I'll, be collect one more time here. So we are saying that go and hit core api group. Work on a part object give a name to it. We are giving a label to it and talk about what labels are in.

A

Some time is like a tagging to your objects, then we are saying I want to define desired sheet of my object. So in this port I want to have a container and container name is going to be this one. This is the image I want to use, and this is my port number okay. So now, if I run this particular command here, you can type it of course, but I'm just writing the command by this. So, okay, it's already exists because I think I triggered this earlier.

A

So let me just remove the pod for now, because I think I was showing you earlier so I'll just run this command here and you can see here. This pod is created now correct and if I do get pods here, you can see this. My pod is in the running state and if I do one more option here with some more option- minus 4 wide, it is telling you that pod has an ip address, which is this ip address.

A

That's running on the given node, if I have multiple node plus, will give me the different node, but I am on the master node right now and if I describe the part I can find additional info about the part.

A

Okay, I'm just going to wait here just want to make sure that you guys have done till here just say thumbs up. If you are done till here that you can list the parts get part minus or white.

A

So next is, if I kind of now find out more info about the part so I'll run, a command of cubectl describe the object type and the object name. So your object type is a part and the object name here.

A

So, as you can see, I am getting other information here, like my part here, the the name space I'll talk about that in few minutes. What's my pods ip here in this pod, what katniss I am running here! I have the information here then there's other events right. So what happened as I deployed this part right, so this got assigned to a node image already there. Otherwise image would have got pulled in and then my curtain is created right.

A

So I can describe the part to find more information and then, if I do get part minus or yml, this is giving me. The output in a vml format- or I can have minus or json, will give me a json format. So I'm saying get parts minus or voil.

A

So, as you can see, respect right, this spec field is of course bigger than what we have mentioned there right. This has kind of got some default values as well, so spec is the desired, which I want right. If I come down, there is a field call status. That is the um current state of the object. So when you look at any objects, spec is your desired state. It still is the current state, so then control what kind of match that states are matching there. Okay, to remove a part, I can delete a pod here.

A

My pod is gone and if I do get parts here, nothing would be there right now.

A

So far so good.

A

Okay, now suppose I want to deploy the pod with multiple controllers.

E

Sir one question uh I mean putting up a command wide and describing a pod. What is the major difference.

A

No just a more option than nothing else like miners. So what will give you more parameters there right.

E

I mean the command, which is cubesatl pod, my bot hyphen, o yaml. That will give you all the yaml information correct.

C

E

Cry will give you the exclusive information, sir. Is that.

A

E

uh The describe pod.

A

Yeah, it's kind of a you know, one not on the same thing, it's kind of just a different way of different view of the same thing with additional information, but information is common. There yeah, I could think of that describe and by ml may be same output more or less yeah. It's just a viable form so that you can format it better way. See. If you look at him describe it, don't you have the events right.

C

A

The yaml, you don't have any events, so there are some minor changes, would be there. Yeah.

E

Yes, sir you're right. Thank you, sir.

C

A

More questions.

E

uh I mean one question here, sir: how will we we'll go and determine the name space here.

A

Okay, I'll answer that so basically, if you describe the part, you'll get it but I'll talk about this. Just give me a few minutes I'll talk about the responses after this. So next is a if I have a pod with multiple containers here right, so a part can have multiple containers in the spec section. I would define this right container, one two and three right so whatever is I want was if I deploy this now. If I do a get parts here, you will see it saying radius 0 of 3 okay.

A

So now this part has 3 containers. That's why I'm getting 0 of 3.. In the previous case, I got only one of one, because there's only one container also look at here. I have used the apply command instead of a create command in the last session. Right so create can only create the object for the first time and make change on the fly as well. So if I have made some changes on my application, I want to reapply them on the fly right. I can do that.

A

So in general cases we are going to use the apply command because can can create and make change on the fly as well right. So that's what we would be doing in most scenarios.

A

So now my partner would be in the running state because all would have been pulled up and if I want to remove the part, I can again do a q field delete part on the pod name. But here I am saying: delete minus f in the yml file. Okay, now remember this particular command does not remove vmware file. It just removes the content or remove the object created from that particular yml file so like here we are saying that in this vml file I want you to create a part called multi-container.

A

So this would remove that multi-container part for me and it won't be there anymore.

E

In the multi-connect container part, we just uh eliminating the changes in the in the specs form. Am I right, sir?.

A

You're just changing here because containers we want to have more containers here, correct yeah,.

E

And these images are pulling from a docker hub repository yeah.

A

By default, yes by default, they are, but if you want to perform private, they cannot prefix that you have to give a location for the pvc right, not the piece, the not the pvc, the private registry of yours right suppose you have a registry at some location right like register.guru, so I'll prefix. That here, that's all.

A

Thank you, sir.

B

And uh like uh the difference between the create and apply like, is there anything which gets recorded to playback means in.

A

It's not like that. It's not like nothing to do. Information is under the key value store anyway, when you do apply or create.

D

Right, yeah that.

A

Is any way to run the keyword list because correct okay, what you can change? For example, let's say I am adding a label on the fly, cutting.

C

Okay, so that would be.

A

An on-the-fly change.

C

A

D

A

Okay, so we look at the the name spaces now and then post. It will take a five minutes break after that. So when you deploy the cluster uh or when you kind of have a cluster which is shared by multiple teams, uh you want to make sure that not one team is able to or kind of just take the resources and then consume everything on that cluster or there right so or or just want to kind of enable the multi-density environments.

A

For example, you have a dev team, prod team and so on, and you want to have different users go on a specific team and work there. So commodities name is commodities array using which you can basically partition this cluster virtually. It is not the kind of a hard partitioning per se, but you are creating multiple virtual cluster on the fly like here. For example, I have a dev team and a pro team correct, and so I can- and I was saying that on those teams, dev team can consume at max 4 gb of ram overall.

A

Similarly, pro team can go up to 8 gb of overall ram, so I'm limiting the users of the dev name space to go beyond 4g so that they cannot consume the entire resource of the system and so on. Correct also, the objects deploying are unique on a given pod are unique on a given pod. Only so, for example, my part, the dev name space is different than the myopod in the prod name, space correct.

A

There are four default name space. What we get as we install the cluster default cube system, cube node, is in queue public in that queue, public and cube. Node lease are primarily used for classification boost traffic, so we kind of ignore them.

A

Cube system is a name space in which companies run community, specific applications and so on, while for users like us by default, we go into the default namespace, so by default we all end up in the default namespace.

A

So now, if I kind of look at get namespace command here, you can see here. I can see six of them, two of them created by us for our lab purpose, but here these fours are there, which you can see now as an admin. Currently we are admins, we have the full access and we can create new name space as well. So I am deploying a new name space called test here now, while I deploy the applications right, I can.

A

I can hard code, my um what you call so I can hard code my pod.yml file like this, so suppose I had this dot voil file in the earlier section right.

A

So I can say that run this particular part on the test name space. I can hard code like this, but rather than hard coding. What we are saying is, I would pass that cube, ctl, create part command and say minus n test. So I deployed the name space here which current doesn't have any parts running anymore right.

A

There's no pods in that name. Space now correct, but then, if I do get parts uh sorry create part and minus and test right, I'm saying that run that part on the test name: space, whether there's the pod dot yms. Let me copy that actually.

A

From the other location, so I'll basically copy this.

D

Into the namespace check section.

A

Yeah just have same iml file. You can apply it manually. I can do it manually for you as well.

A

So just need a. We just need a file there. It can be anywhere. So it's not at the part of this, but I'll have it here.

A

So, as you can see here, I have the spot dot yml file same file, which we had in the last section, and now I can do cube ctl apply or create or dot ymin minus n test.

A

So now this would deploy the pod in the test name space and if I do getport you can see here I'm saying mypod in the test namespace here and now. If you discard this pod here right so cubectl uh describe for my part, minus n test you'll, see here because of course, you're saying the minus n test here.

A

But if you look at even in the describe the quad, you see here where this part got deployed and all those will come here correct and if I remove the namespace automatically, all the apps of that place will also go away because I am removing my part name space. All the apps of that dashboard will be removed as well. So now, even my fastness is also gone here, so I'm coming back to the normal other things.

B

The best practice uh with regards to like when we are actually working with uh namespaces, like what you suggest some tips or like some uh gotchas, which needs to be careful.

A

B

Means, for example, I'm working on a real prod or a real kubernetes environment. So what should one consider or keep in mind when we are working with uh namespaces? Basically.

A

Of course, you set the resource product correctly. That's one of the biggest things correct, because perfect, perfect.

C

You don't want to do.

A

That, then, somebody can take out kind of consume that, like infinite source of the name, that can be an issue right, yeah.

C

A

Of course, then, when we we're not covering in this topic, but when we create our users on the cluster right, we want to make sure that our users are part of the given project itself right other than a dev user. Get access to plot cluster right, name space. It can be a difficult thing right.

C

A

Will take care as a part of an admin like that? You don't want people to go to a random name and deploy any abstract. You have.

C

A

Them the permissions in a way so currently.

F

We are not, as we.

A

Can do anything, but in the real world the users are not admin in all the namespaces. So just say that correct of those things yeah no.

B

Perfect yeah make sense, so one is for the users and also for the resources correctly to connect. Yes,.

A

Okay, fine, um so with.

C

A

Information we are now going to next section here, which is application, lifecycle management, in which you want to look at how we would deploy the applications uh in the way that it can scale up scale down on demand. So we just saw right now the part suppose.

A

If I would deploy three copies of my application, I would deploy three parts separately now right, so they'll not be like a group together and if one app one part dies, orderly new part comes as a replacement of that right, so we don't we or we want to kind of deploy these parts in a declarative way, and that's where we have this object. Called replica set and replica set help me deploy these pods um uh declaratively so now.

A

Currently, here is a example of a replica set in which we are saying that this is my replica set with the name, whatever the name I would give. I want to have three replicas of a given part, so I want three applicants desired. Count is three current three of them are running.

A

We are all good, but now, let's see if one of the part dies automatically a replacement part comes in place of it right, as you can see, another part has come now, and this is my replica set yml file in which now we are not going to only v1. Now we are going to slash api's apps v1, so when you say apps v1, it means it's under slash. Apis, slash, apps, slash, v1 kind of a thing right, so that's how it is kind replica set, define the roughly account and define your spot template here.

A

So I'll talk about this labels, letters in few minutes, but for now replica set, would deploy x, number of copies or replicas. We have mentioned about this part template the part which you have mentioned. Okay, so let's go ahead and deploy this replica set again. The same example what we have seen kind replica set replicas- and I want to have three applications of this given part right. So now, if I apply this replica set here, this is going to create a replica set.

A

For me, short form is rs for it you can say, get rs as well, and now you can see here it would create three parts for me right. So, as you can see here, three parts have come here, so this is like replica set name and some hash value we have now. If I want to. If I want to look at if I delete a pod here right.

A

You will see that automatically a replacement pod would come.

A

This is getting removed. You can see here right, as 2 is too fast. Another part has come in place of a replacement of this so because we killed one part. My desired count was still three current count was two: a replacement part has come and we got a pod with this new uh id here.

A

Any questions here and if I remove my replica set, every part would go away as well correct.

F

Because so yeah um so like, if your pod has not died, but for some reason I mean it is not working that application is not working. So that's something.

A

F

Okay, so in your your replicas I mean so in this case it has to completely go off. Only then.

A

It will create a new one, so what you're referring is, uh let's say, failure of the health correct? Okay. So what.

C

You're, referring.

A

Is maybe the the pod is running, but inside the pod the app has died correct. So for that purpose we have a different thing called you do liveness probe in the lab health checks. You can do there, so those are a different part which will configure and though, in that case, if the health check fails, then I would try to do the container restarts on its own to bring that up correct, but.

D

F

A

It's kind of completely died for whatever reason.

F

B

And uh the um if you see the yaml basically means the amal. We see this selector.

A

I'm going to explain that I'm going to explain that.

B

Yeah, okay, yeah.

C

A

Let's, let's try to explain that in few minutes. So let's do that so here I am going to again install it. So again I have three parts created.

A

Correct now let me deploy one more part here, uh so let me get that part created, so it would be. In my other section, I'm.

C

A

Going to copy one more part definition here, it's not from! Let me just copy it here, actually copy.

A

And I'll copy this slash tnp here now here I am going to actually remove this label for now here I'll copy, this yml file for you to consume once if you want to so this is my current vmware file.

A

For some difficulty copying here I'll copy from here.

A

As you just want to copy it, you can do that as well. This is my pod dot vmware here and now. Let's see if I apply it here, so cube, ctl apply, minus f, slash, tnt part.

A

So now what we have is we have. uh We.

C

Have four parts running.

A

Here right- and we saw that for uh for those three parts right when we deployed with the replica set here. This is my port template. This is my port template. So this is my replica information. This is my pot template in this spot template I have defined. The labels called app is equal to nginx, so three of these pods have a label app engine x, okay. So now, if I look at the command cube, ctl get parts, and I put an option here: hyphen iphone show hyphen labels.

A

So, as you can see here now in this case, when I run this command right, you can see these three parts have a label. App is equal to nginx, while the first part doesn't have any label here, because we are not.

C

A

Right, there's no label has been given here. That's why there is no pod, in this case correct no label in this case now now there's something called a selector using which I can select my objects. So, first of all, these labels can be given to any objects, not just pods. I could label my different objects as well and once I have labeled them, I could select on them. So now, if I do cube, ctl get pods and minus l for selection here.

A

Okay, so if I say give me all the parts where the label have is set to nginx right now, you can see here it is not going to list all four parts. It's going to list only three parts for me, because only three parts have a label. App is equal to engine x.

A

That is a point right and now, if I look at some more way, if I say give me all the parts where app is not angelix, I get my my part correct.

A

So this way I can basically make some selection criteria, so I deploy different objects and if I have label them, I can do a selection on them.

A

One more thing, if I say give me all the parts where there is a label called app, I don't care about the value right, just give me if they have a label collapse. I get these again three parts there, so I can deploy these applications, give them a label and do a selection on them.

A

That's what happening in this case as well, so what's happening as the replica set gets created for the first time. It is saying that I want to have three replicas three applicants of what I want to have three parts under me. If they have a label app is equal to nginx.

A

Let me repeat again: I want to have three parts under me: if they have a label of nginx and if they don't have it then create them. With this configuration.

A

D

Want to create three replicas, I want to have three applicants.

A

For the parts, if they have a label app engine x and if you don't have them create this one, that's when, when we have, when we don't have any parts created, three parts come up so being said that let me give an example to you. So what I'll do now? Is I'm going to label my pod, which I have here right? I am going to put a label on this part here.

A

This is my new configuration of my pod.vimen. I just added a label there.

A

Yeah so now this is the label which I have right now. If I deploy the spot.yml correct.

A

You will see that I'll have this part here, and this also have a label called lab is equal to nginx so far so good. Now, what we'll do.

D

A

Going to apply that.

D

Replica set what we have.

A

From the last step,.

A

C

Here, what we are saying is, I want to create.

A

A replica set- and I want to have uh three parts which have a label app, is equal to nginx, correct.

A

Right and if they don't have it create of this particular category, okay, now before actually do I just paste one point: I'm going to come back to it and then I'll come back with this example again, so I'm just going to remove this particular uh dot ymil for now, but I'm going to come back to it just after this, but I want to show you one thing so currently I have this replace the dot wireman. That's perfect! I'm just going to reply the file here. Currently I don't have any part running here.

A

Just to make sure that yeah I apply this replica set dot by mill here.

A

Correct and now, if I do pods, are there perfect now, if I do a cube, ctl get part the pod name, and I do minus oyml here. So I'm looking at the part and saying give me the part in a minus or via ml format. Here right now, if you look at when I do this thing- and I do a less of this, so you can see here it has something called as a field called owner reference.

A

So I have a part and as I deploy this part which came under the by the help of a replica set it you can see here there is a field called owner reference, and it says that this particular part is child of a replica set called nginx. Any questions here. Are you able to follow.

F

I have a question um when I'm I'm I'm trying to follow, but um you mentioned something like in the replica set when you go from the upright you have. The kind is replica set. The name of the replica set is engines and within ngcs you're saying I want to create three replicas in that replica set, and then uh the label of the those three parts will be uh uh engines right. Yeah.

A

Whatever app is going to enjoy yeah.

F

Whatever in this particular case yeah, uh but you you mentioned something before that, if the app, if the label is not ending, then you carry.

A

It I'll I'm going to that point I'll, come to the point. Okay,.

A

So what I said right now is I'm just saying what I'm showing trying to show it to you earlier. I had no pods there. I deployed this replica set, which gave me three parts correct now I am referring one of the part here by the get part and minus low by ml format. Here I have. I had three parts now because of a replica.

A

I am looking at one of the part here in the minus of varml format, and here you see that there is a field called owner references being set. So now this pod is because it got created by the given replica set the owner. Reference is being set like that. Okay, it is saying that, for this particular part, the owner is replica set name, nginx correct. This is clear to everyone.

A

D

A

Assume that there's no question here, I'm just going to do one more change here, I'm just going to open this part remove the label from here. This is my part here and I'm going to apply this part again.

A

Correct now I am going to look at the my part in the minus o yml format. Now, if you look at for this particular part, there is no such thing called owner reference, which means for this particular part. There is no one that has been set. It is an independent part which is like. If you kill it, nothing happens. This part just goes away because it is an independent part for you. Okay. Now, why I'm talking about this particular point here?

A

There's a reason for that for sure I'm going to remove the replica set for now here remove replicas at engine x, which kills my first quads here, which I have so I just have only one part running here and currently I don't have any label tube correct. It doesn't have any labels being set here. Okay, now I modify my pod.yml again and here I basically add a label.

A

Labels app is equal to nginx, okay. So now I am adding a label in my app engine x. I'm sorry in this particular part here. My part I'm going to reapply my part here and, as I reapply you can see here, there is a label has come to my part. So currently, I'm going to I just have a pod, which has a label app, is equal to nginx done now, I'm going to apply my replica set again here right now, my replica set.

A

If I am creating again, I am saying that under this replica set, I want to have parts which have a label app engine x and, as you can see here for this particular bipod, I have the label app is equal to engine x, which I'm looking here correct.

A

What's going to happen now is because there is a part called myopod which has a label app is equal to nginx. Also, it does not have any parent being set right now there is no parent is being said, which means it's like an orphan pod.

A

Correct then, what's going to happen, is my replica set engine x is going to take it as one of the child correct, and it is going to create only two new pods rather than three it's going to create only two more parts. If we deploy our replicases, let me show it to you.

A

If I apply my replica set now right, you now see that it basically creates only two new parts. Not all three are you following and now, if I describe my part, you will see that owner field has been set, because when my replica set came, it found the part with app engine x. It has. No parent has taken one of the child here and then moved on and just deployed two more new parts.

F

The question here uh I mean- I don't know whether you'll discuss about this later but.

C

F

What if, if there are two different users, and by chance they gave the same name? And but they don't want to be part of the same replica set.

A

The same label right once.

F

Yeah yeah, it's like they're two different uses and they don't know each other and they they create the same label, app engines.

C

F

What will happen how to avoid that situation? Where why do then they will be part of the same replicas.

A

Right, of course, there's a problem of course, so what happens is first of all, that's why so, when you are working in a given team right, you will be part of a given name space consuming that right. So of course it it. We will assume that, okay, because we are part of our same team, we may know each other in a way. But that's not that's! Let's keep this aside, but as a guideline, when you work in a team in a given name space, you would basically choose the labels very carefully right.

A

So at least you'll have one of the unique label and selected here. So, for example, in this particular case I can have not just one label. I can have multiple labels here, not just when I can have multiple labels. Similarly, I can have multiple selectors, but at least one has to match here. So when I'm working on these level selectors, I work on that that I have some kind of a unique key which I am giving it for my application.

A

Only that's a good practice to do it, but being said that, let's say if I now create one more replica set here, so let me just actually show it to you.

A

So if I um yes, I can.

A

Now assume that if I change the replica set name here same with shadowpoint w, I just give a name new here right now. Let's see if I reapply my replica set now, but do you think that I'm going to have a three part or a six part now answer in the chat? Please answer the chat because I just changed my replica set name here applied it again. You don't want to create a replica set for me. My question to you is now we would be having three parts or six parts there in the environment.

A

Answer on the chat, please: three: okay, three! Okay! So overall we'll be having six right. Why? Because for the new replica set right, the new replica set when it tried to come up right, it would see that okay, there are three parts already, but they are having the leverage I'm looking for, but these already have a parent right. I cannot just fetch from somebody's else right so because these new, these parts are part of the some other replica set. I cannot take them, which will create three new.

A

So even in this case, labels are same and perfectly fine correct, but it's a good practice. You basically want to put some kind of a unique labeling between your applications that there's a confusion like. As you can see here. It looks perfectly fine to work as it is. There's nothing happened here, correct.

C

A

This can cause confusions and some kind of on and off situation. So, rather than doing that, you choose the label selectors wisely.

F

Is there, is there a no? Is there no um internal mechanism in kubernetes which can indicate the user that we there are some pods which doesn't have a ownership, but they have the same app or the labels, and it's going to be included in the replicas that you want to continue or something like that. You.

A

Know there is nothing like that, but you can build intelligence on their own. It's like this is being said: that's not part of the base cluster, but as a uh as a cluster, as in cluster manager to a given environment right you can deploy some kind of a logic. I said that right it is extensible right, so if you can build, if you want to build that, you can do that kind of a logic and built in okay, build on europe, but does not come by default.

A

Anything like that, because this rule is very clear here, like that, um what let's go to as if they say, is working here, but of course, as every cases are there, which can become a problem. But if you go look at the next example called object of a deployment. This problem is also not there there.

A

Okay, I'll talk about that. Okay,.

B

So for one question I had sorry for interrupting. Why did not discarded the other two which were created because the replica it will try to match the current state right.

A

Which we're talking about the new one or the older, this replica.

B

The older ones that engine x dash 9 r and that engine x, dot jb this.

A

These are pod names, so these parts are only part of some of the replicas right. Why would I want to be the owner necessary.

B

This is a different file. You have created.

A

Yeah, you can just see that this is the body just say the different replicas, that name it right. Just everything is same, but this name is different. Correct.

B

A

It should have been started. No, no, no.

B

A

A new object right.

B

Okay, so older one, it does not discard that.

A

Is still there, it's still running yeah.

D

F

I mean if I understand the question, I think uh so, if I have a replica dot yml and I create a replica set and after that, if I just update the same file and then apply it again and if I.

A

Change the the name is different right. There's a different object altogether.

F

No, no, I I understand what I'm trying to say is. If I have a replica set.yml file and I I create a replica set and it gives a name say abc now I apply, I apply it, so it will create a uh like a replica set name abc and there will be three parts.

A

Within it, correct.

F

Now I then I go back. I update the same replica set yml and then now I change the replica set name to xyz and I apply it. So that means that the file can be the same. But as long as I keep applying it with new changes, it will keep uh adding a.

A

New form, what happens what's happening at the high level right, this ml file has no role to play. What actually happens is this particular request. Whatever you run, the command right, these command goes or sent over the rest. Api called to your api server correct, so this file actually have no role to play at all.

A

Correct what's happening is, as you apply, it basically kind of makes a json request like a payload to a thing and then send a cross correct so because, anyway, because you can change the name when you reapply you're, sending a new json data to the server right to apply it.

F

Yeah, I got it: okay, okay,.

A

Okay, so I'm just going to delete all the replica set for now so I'll just say: delete short from this replica set I'll, say hyphen iphone.

A

This removes all of my replica sensors.

A

Okay, so with that, I'm going to move to the next object of uh something called as uh deployment object. Okay, now deployment object uh sits on top of replica assets so a way when we do our app deployments in our cluster or wherever it is. Typically, we use the deployment object. We don't use the replica set as its as I mean to deploy the applications. We use the obvious deployment creation for deployment. Object, creates replica set for us. This would create parts for us.

A

So I'll tell you why we need separation between deployment replicas in few minutes, but for now deployment object. Create replica asset will create parts for you. Okay, so let's understand that part for now, so let's go ahead and deploy a deployment object. So if you look at the yml file, the yml file is similar to the replica set here. It's just that we have changed the kind here.

A

So kind does not become deployment object and, of course, some as you applied the default values which get applied as a this object is different than the replica set. Of course.

A

So you apply this vml file, but, as you apply it, some default values would automatically get attached to it, which we'll see so now, if I do a deployment creation here, this would do a deployment creation which would create replica set for us which would create parts so deployment replica, set parts and you'll look at the naming here right. This is my deployment name and the replacer name is deployment plus some hash value and part name is replica set name and then some hash value correct.

A

This is how they have been kind of connected deployment, the place and parts and, of course, they'll have a parent relationship all right. If you look at this particular part parent, it will be replica asset. If you look at the representative, the deployment is a parent. So if you just this, get the pawn in a vyaml format, the owner reference will be set accordingly.

E

uh Sir, in deployment sets, are we creating three replicas? At the same time,.

A

Then only one replicas replica set is only one. I mean.

E

When I'm looking to the output of kit deployment, I can see uh desired status and three, and uh there are three replicas here for you: yeah three replica assets.

A

Three replicas of a given application: three replicas.

E

Of given applications means.

A

E

Now, here uh the label, one minute here: lubricant ngnx, they.

A

Have to match they have to match. There's no other option.

E

Got it? Thank you.

A

Any other questions, anyone.

E

So in deployment sets it's just we creating a replica sets.

A

This is a deployment not a set just a deployment object yeah, so just it just sits on top of a deployment, create replica parts and I'll tell you why we need separation here in few minutes, but for now just look at this one deployment create replica set which will create parts for you. Okay, I'll tell you why we need a separate like a layer on top of the replica set. Just in few minutes is everybody will deploy. The deployment object, get the three parts running for that: okay.

A

Okay, now once we have the this is deployed now we can scale up and scale down our deployment as well right, so we can scale up and scale down our deployment as well. um So here.

C

A

So what couple of ways of scaling up and scale down the applications one is, I can always apply my vml file. I can change my wi-fi reapply. That's one approach I can have what I'm saying is. If I look at my deployment configuration file.

A

If I have to change right, I can change in this volume will find reapply, which I'll do it in few minutes, or I can do it from the command line here. So now, if I run this command, this is basically changing the uh deployment desired state from three to one. So, as you can see, currently, I'm in three parts, because my reply account has been set to three desired count is three, but now I am changing my replica count or my desired state from three to one, which means two of the part would get killed.

A

As you can see, it happened just on the real time right. I changed my I changed. My deployment count or the the replicas from three to one so desired count has changed three to one, which means, because I had three parts and two would be killed and I am left with the one part here only similarly, I can scale up as well, and you can see now two paws would be there.

F

So this this deletion of the other two replicas, they are dependent on some like there's a user traffic on one application and it will not kill it right or or how? How does it work.

A

Obviously, these are all similar applications first of all right, so there is a so we don't communicate to the parts directly. We have a layer called service on top of it right which I'll talk about the next section section here. So service is on top of it all the parts. So when you hit service that traffic go to your pods, okay, so from the end your perspective hitting the service only and this service kind of do a load balancing for me behind the scenes from multiple parts.

A

So if one of the part goes down, it would move the workload somewhere else correct, but being said that, if, if your pod get killed right now, those parts can be killed. Let's say, for example, uh because the power failure you don't have a control right. If the power goes down, everything goes away there and there itself correct. So, first of all, when you are going to microservices world, you should be writing an application that you can fail any point of time and you can know how to bring it back correct.

A

So that is not something companies kind of solve every problem. For you. No you as a developer. Also you take care of those things right, because you're building micro services, you are should be okay, that I can die any point of time.

A

Correct. That's one case where, where your note in the power failure everything is down, but, for example, if you kill the part like very specifically, I want to kill a pot right. In that case, what happens is no new traffic would be sent to the pod which is getting killed at the same time, the killing part would have some graceful period time to clear out whatever that cache and so on it may have so.

A

We can configure that the default value- something is there, but I can change that and say that, okay, if the part is getting killed, I would do some these things before it kind of get removed completely.

F

Okay, yeah but yeah. So when you are when you're changing your replica account from three to one, it just randomly, selects the two parts and deletes them right.

A

Right actually, basically, yeah yeah. You know it kind of deletes the last, uh but that's created one okay. Whatever the last question one there will be, it will be removed. Okay,.

A

More questions.

D

A

So now let's go ahead and scale up to 10, for example. So I'm just going to make the count of 10 here.

A

And scale my deployment to 10 right so now I'll have 10 parts running.

A

Because I have scaled to 10, okay, now all of these applications, which are running right, they are running the app version of with the image of nginx 1.9.1.

A

So if I describe my pod here or cubectl getpod or describe the part whatever.

A

You will see that currently I'm using that image of nginx 1.9.1. This is something think about as a version v1 of my software. Now, assuming that you are running 100 copies like that right, you have scaled your app from 100, 200, 100 reps are running uh continuously.

A

uh Now you want to move your app from version v1 to v2, for whatever reason right, you want to upgrade to the newer version correct. So now, what are your options? Your option?

A

One option is that whatever the replica set, you have right you remove that replica and bring a new one, which means you would be first deleting all the 100 parts and then treating 900 parts again, which is going to be kind of a say, a downtime case right, because you would first be treating all the pot including again and so on, correct, but to overcome that we have something called as rolling updates here, and that is where having the deployment with multiple replicas that help here or having the replacement of the deployment help.

A

So this is my current state here, in which I have a replica set, a which is running all of my parts here right now now I want to move to the I want to I'm upgrading my app from v1 to v2, for example right.

A

So in that case, what's going to happen, is the deployment creates one more replica in parallel, which currently doesn't which doesn't have any part as of now? Okay? Now, what's going to happen, is this the deployment part point to w two different replica sets? Then I'm going to create some parts in the new replicas and kill some old ones correct.

A

This is called the rolling updates, so here I am not actually getting a downtime, because my applications are all up all the time. But what may happen is some of my customer may see an old view. Some of them may see a new view, but there is not a down time in this application, because I am upgrading with the help of rolling, upright strategy here and eventually, all of the part would move to. The nearest state here make sense.

D

A

Okay, so let's go ahead and uh change our image version now, so I'm just going to run this flag here so cube, ctl get replica set I'll. Do a watch flag here, right watch, like means you just minus w, is watch. I am watching on this object right, so I am saying watch on all the replica sets here now. I am going to basically change that image version to lets say version of stable.

A

Okay, so I change the version of nginx from 1.91 to stable version, and now I am going to redeploy the deployment right. So as I redo this deployment or reapply it you'll see that there is going to be one more replica set would come in parallel and you will see the uh the spot migration which I mentioned earlier. Okay, so let me apply this and, as you can see, that a new replica set has come correct and in this we are deploying new new parts here, killing some old ones, and this is going to continue.

A

Eventually, all of my pods are now moved to the newer version. Correct and the older places kind of gone cleaned up. It's got cleaned up now, so every part what we had is now. Basically, all the new parts are the new replica set and older ones are gone.

A

This is how the flow has happened, which I was mentioning earlier with the help of replica set rolling update strategy, it's a default strategy as well. So if I do this here, so if I do a cube, ctl get deployed.

A

Deployment name.

A

And if I do minus oml here, what happened nginx deploy, I think not meant deploy.

A

So, as you can see here in this particular configuration I have specified. Where is it here this one? The strategy is the default type is called rolling. Update, of course, are some parameters I'll talk about in a minute, but, as you can see here, I had only one replica set here then, as I did apply, I got new replica set and now you can see that in the newer replica set, I am creating some new parts right desired. Count is three, so this is desired. Current and ready state so desired.

A

Count is three the new replica set okay. Now, how does this three number has come? So if you look at here, I currently have. Overall, I have 10 parts overall running and if you look at in this strategy here, we have two things here, which are the type and two parameters here, which is saying max surge and max unavailable.

A

Max search means how many you want to create in the new replica set, which says: 25 percent it can be actual number also, but here giving a percentage.

A

So 10 is my overall parts running 25 is 2.5, so then I'll take a ceiling value. That's why the three number has come, so it depends on whatever the max search you have given here. Those my new part will be created here, correct so in the new.

D

A

Set I'm getting three new parts created the same time saying max unavailable means 25, basically same time, saying that how many I want to get removed right, older version, so here again the 10 percent. There was a two point: five here is the floor value right. So here two, so I'm changing my desired account from ten to eight for the first one, because I'm trying to remove two parts there correct so ten has become eight now, so this ten would continue until all of my parts are now being deployed under the new replica section.

A

So we have done that so up now, all of my parts would be of a new version. If I look at my new parts now right.

A

They would be all running the stable version of the application.

A

A

Describe part or name, if I look at the images, the image name is changed as well correct.

A

This is what we have is a default strategy of the rolling update of our deployment.

A

One strategy here called recreate okay. Now the recreate basically simply means that you are going to first remove all the parts and then create a new ones. Okay, so recreate means that I'm going to first remove all the parts from the current replica set and then the kill will encrypt the new ones. So, in this strategy I'm going to have a downtime. This is required when I don't have the liberty of running two different version of the application.

A

At the same time, when I can't run this kind of thing that, where one version is new, one version is old. I can't do that situation there. That's where I'm going to use the strategy of recreate when you look at, I think I talked about a bit on the label and selectors uh earlier, um so let's just quickly uh kind of review it uh so yeah, so we can assign labels to any objects. So now labels can be given to pods nodes deployment services.

A

Whatever object is there we can assign labels to any one of them. So now, here we are giving a label like release, table or, and then we have a release, table or release beta. Then we have something called as my or I o c and v dev or q a so I can assign labels to my objects. Okay, then I can do a selection on them right, so I can do a selection on them here. I am saying that give me all the parts where uh release is set to stable right.

A

So, as you can see here, this one part and blue part is a green product. Whatever color it is whatever it is, but I am saying that I am loosely coupling my applications, even if they have not come from a same deployment or different same parts whatever is created so as long as, if I'm having a label which I'm looking for I'll select them, I'm not saying that they have to be coming from a same object right. No, I'm just loosely coupling them as per my requirements.

A

So here I am saying simply that release select me all the parts where release is set to stable. I don't care from where they have come from correct. So there are two kind of selectors we have. One is equality based other is set based selection, so equality based that we can say release is equal to stable and not stable and so on.

A

Then you get a set base release in dev or in queue stable, so logical r will not work with the equality based, but with saddles you can do it correct so there in equality based. You can't do a logical.

C

Order but such a typo here actually.

A

Should come under only for equality, but this.

C

A

Okay, so now, let's look at example of that, so I'll first remove my deployment whatever I have running so that I have a clean slate here.

D

A

Must be removed.

A

Okay, now for the example, first, I'm going to deploy three points here here. One part has no labels here. This is just my part with no labels. Then we have another part here with tier front end release prod. This is another label. Other part I have, which has a label tf back front and beta or release is equal to beta.

A

So I'm deploying three boards here and if I look at the show labels, you can see that the label, what I have got correct now, if I work on a equality based selection, uh so I would say now give me all the parts where tier is equal to front, as you can see here, tier is front end set to first and third one. So those will be listed here correct and I'm saying give me all the parts where tier is front-end and release this product. So I'm just saying this one. Third, one right.

A

Then I can say that give me all the parts where cure is not susceptible, so tear is not set to front end, which in this case the tear is not set to front end, which means it's only my body here. My body doesn't have that label yeah correct and same I'm saying give you another part where tears value is given here.

A

So you can change again to kind of choose from different kind of combinations. As per your requirements to get the right number of app all right write listing of the app okay, then we have this uh set base selection, in which I am doing like a set operation. You would know so release in front end right, so release in front end. We have somewhere release is either beta or prod, but not front end, which means for the first.

A

I want anything because there's no part in the default, which is a level release front end, then I'm saying give me all the parts where release is beta or broad so I'll get first and third one.

A

Then I'm saying give me all the parts where pier is front end and the release node will be top rod, so I'll get some more things here. Another thing is okay, so tear in front end tier in front end and and what I'm saying release not in b time process release, of course in beta. So I can't have anything: that's why I'm wearing nothing here correct and then I'm saying give me all the parts where here not in front and at least not in beta and plot, which means this is going to give you this.

A

B

uh I have one question for you: the commands that you are using all these commands. Do we have any link or any cheat sheets, because those are very informative commands. We are noting it down.

A

But ah these will learn the practice there, um so we actually published a cheat sheet earlier uh I'll see if I can find and send across to you so.

C

A

Village, which are there you can actually just search that queue. You may get it not from me, I mean, maybe you might get from ours we have published, or there may be somebody that's published to answer these things. Just practice, you'll get it it's just the practice thing.

D

A

Anyway, you have access, for, I think, around whatever four ones, I think getting access to you guys for the six steps. Yes, this is there, and also now you can also combine the equality base and set based as well. So here I'm saying that give me all the parts where uh there is a label called component to redis and it will be end here and tier is set to cache environment, not to dev.

A

So this is how you can kind of define different uh expressions, different equality and sandwich together and find a grouping which you are looking for.

A

A

Then we have this something called as annotation objects, using which basically, we can put properties to your object. So again, annotations are your key value pairs, but they are primarily used to put properties, so we don't select based on uh these rotation values or key value pairs. This is primary set, some kind of properties, for example, you want to put a property of who is the shareholders?

A

What's the on-call people, id name email id whatever, or you want to say that? Okay, how would I collect my logs monetary data and so on? So you can put any kind of properties to your objects, which somebody else can kind of look at and perform something on top of it. So, for example, um the example which I gave pointed to logging and monitoring right, for example, you want to say that okay, do you want to enable monitoring for this app or not?

A

So that can be a property correct, so my lock server would collect the logs only from the applications which would have this logging start logging. Something enabled correct because, eventually, when you have to pay you to pay for per data injection and so on, on the server on those, so you kind of control, whichever you want to do, which you want to understand and so on.

A

So those kind of pointers can be useful for other purpose. Not just maybe you but annotations are there to see the properties of our objects, then what we can do is we also have something called as field selectors. They also help we kind of select. They also help us basically select our objects in a our battery.

A

So, for example, uh I want to find out all the parts which are in the running state right. So if you look at these parts right, so cube, ctl get parts, nothing is there. I removed, I believe everything. So let me remove a pod on the fly now.

A

I believe we have a demo here yeah. So let me just so. This command cube, ctl, run nginx and you give an image name and that image name, so this command just creates a part for me on the fly right. So I have a pod created now and if I look at the spots you see it will get pod nginx, minus oyml.

A

You'll see here in this particular part under the status right, I have something called as phase is the running state right suppose I want to select the parts based on this uh running right or phase running state right. So these are not the regular field which I am kind of working on, but if I want to select on those fields which I want to kind of query and so on right, so I can query based on this field selectors here. So I am saying give me all the parts where stress is the running state.

A

Now I am saying the next thing is give me the parts from all the name spaces if they are in the running state correct.

A

So these kind of uh filters kind of help me to do this better querying there. That's a whole idea about putting this thing further. Now, let's say I'm deploying a new name space here called v1.

A

I deploy a pod also with the name of finders in that namespace. I'm saying give me all the parts from all the namespaces if they have a name set to nginx correct. So this kind of a query which I want to make with the help of these field selectors, I can make it easily, rather than kind of doing some other trick. This command can help me do these things in a better way.

A

Okay. So next we look at uh the section of services right. So so far we have deployed these applications with hello deployment object and these apps were deployed on my environment perfectly, but now I want to expose these apps publicly um or within the cluster and so on. So let's see what kind of challenge we'll face if we just use with the pods ip right and so on,.

A

So if I look at the service object, why do I need? First of all, so suppose I have these pods running. Some of them are blue. Some of them are green, pods and so on. Right now I have this parts ipad which are available inside the cluster, and my user can communicate to uh these pods right, so these parts can communicate to that's perfect, but now what happens is if a pot dies right.

A

So previously all the parts have an ip address right and I can communicate to the port ip, but if the port dies, new part would come up, it would have a different ip address right. So unless I go and tell my new user saying that a replacement ip is this one correct or assume that if your app got scaled up or scaled down for whatever reason right, you would have, you would have got new ips for the parts or kill some more of them right.

A

The ips are gone so to keep telling my user in this scenario that the this is a new pods, iep and so on, which is a bit challenging right like of course, and not impossible, but it becomes things, make things very challenging there, so what we do is rather than uh kind of accessing by them with the pod ip.

A

What we do is we group the similar kind of apps together right, like, as you can see here, we have uh a group of similar apps together, so we have a group of app blue and a group of app green right now we would refer this group as a service so that, as a user, I communicate this group only and this groups somehow load balances between my backend part, which I may have so here now. I am going to basically send a request to my, of course, the service layer.

A

Now this is a service layer here which has a service name, so I can give a name to this particular layer and a respective virtual id. So this ip is not given to any interface per se. It's just like a firewall rule saying that if a traffic is coming on this particular service blue or this ip address, I am going to forward to my these parts behind the scene.

A

Okay, so my user is no longer hitting the pods, it's hasting a group together and when I hit the group, my traffic would go to the backend parts accordingly for the whatever the labels and status I have so here I have labels here. I have selected any questions here, so even.

B

If the pod dies, then this, uh when, when it is, the request, is coming, this labels will end.

A

Yes, service is still there right, yeah, that's what I'm saying.

B

Is still there.

D

C

D

A

Layer is not gone anywhere: okay,.

D

B

And uh I don't know how relevant is this question, but uh this is uh the concept will be same for the private cluster and public.

A

Right correct, it's the same thing. Yes within the classroom. It's happening inside that this is not.

D

All the process.

C

Now inside the cluster.

A

It's going to be same irrespective of on-prem or on cloud cluster is the part of ins within the commodities cluster.

A

Yeah, okay, great so now one more thing here uh we hit so when looking service layer right in this service, we have this incoming code and from this service the traffic would go to my pods.

A

So we refer this incoming port as a port here, but from the service when it goes to the pod v sets a target port. So we hit the target port at the port. So we hit service layer on the port and from service. We go to the power and target port by default.

A

They are same if you don't specify, but you can say that target mode is different, so, for example, you can hit poor date yet here, but when you go to the pod, it'll say 5000 there, so you can do that if you need to right that is your kubernetes uh like service object here now, let's look at a demo of this to understand in a better way, so there are different kind of service types we have so one of them type is called cluster ip, which is a default type as well.

A

So when you deploy the service object, you can say type. If you skip it, then by default cluster ip only so this cluster type is available only inside the cluster. The main thing to focus here is called selector here that this service is going to select, which kind of parts correct.

A

So it is going to be app, is equal to selectors equal to app is equal to nginx, so it is going to select all the parts which have the label. App is equal to nginx, okay, so here let's deploy an object and then we'll come back to this particular example.

A

So now I am I'm deploying a deployment here which, if you already have running you can have it, but here I'm doing a new deployment creation with the replica count only one. So I'm deploying one part here: okay, so again, for this particular part, I would have the part with the label of app is equal to nginx. Okay, now I'm deploying the service here in this service, I'm saying so kind service. Again core api group kind service.

A

D

Is a label given.

A

Here, but this label is no longer used for me, because I don't worry about this label somebody else. Let me select on them whatever. What I am looking here is a selector here, so this label is for the pod object, which I don't care, so you can ignore it. What what you can't ignore is a selector. So what we are saying is this service is going to select all the parts, whichever label happens in the default namespace and this is of service type called cluster ip. This is available only inside the cluster correct.

A

So if I apply the replica sorry, if I apply the.

C

A

You can see is created and if I do a get service here, you will see that there is a service name which I have referred correct, automay called a cluster ip here. This is my port number. So now, if within the cluster, if I hit this particular.

A

Name I go to this ip and so on.

A

Any questions here are you able to achieve this so typically, all the database kind of apps would be of type cluster ip.

B

So uh does that mean we can all always refer with the name name as well inside.

A

Correct, that's the whole point right so what's happening is inside the cluster. You also get a dns for free in a way- that's already already configured there. So when you deploy a dns server or deploy that cluster, a dns server automatically gets created. So let me show you so.

B

It will be created by default. Okay, yes,.

A

If I look at cube, ctl get names get service in the name, space of cube, hyphen system.

A

You will see here there is a dns server running in your cluster correct with this ip address right now. Let's see if I deploy a pod here, I'm going to deploy up order, cube ctl run nginx is my pod name.

A

Image is equal to nginx alpine version, so I'm just deploying a new part here correct and if I do an exact, so I know what uh so, if you have done docker exec earlier right. Similarly, we can do a uh q fill exec to go inside a pod. So let me give the command to you. If you want to try it out, so I'm going to just create one part now and in this part we'll do an example. I want to do inside the pot, so I'll say, exit minus.

A

I t for the interactive mode and give my pod name and say shell, or I can say a shell like this only or I can say it can be hyphen iphone. So whichever works for you so now, I am inside the part of engine x. Correct now, from this particular part, which I am in, I should be able to access my service, which is running in my name, space with a given name. As you can see it's my service name.

A

So if I do curl this particular thing, you can see it's going to serve my package too correct.

D

A

You can see here if I just do an exec to my engine export and if I do a curl on my just a service name correct, you can see that's going to give me the page now what's happening is, as we know that we have configured a dns server in the etc dot com file. Correct.

A

Now, when you give this particular name right what's happening here, this is getting converted into a complete dns name and what it is. It is the service name, dot the name, space name dot as your customer local. This is the complete dns or kind of say fqdn for your service in the default namespace.

A

When I run this particular part, you get the same page though, and if I do a ping, the ping command is going to fail, but you will see that the mapping is still there circles, so ping card has not been enabled, but, as you can see here, this particular itunes ib is nothing same right.

A

A

Are you able to follow guys?

A

Yes, so now, in a similar logic, let's say if I deploy an app right. So if I deploy an app.

A

A

So I'm deploying a service again, but in this case I am now creating a new name space. First of all so create name space dev, for example.

A

Then I deploy the similar service here, gives you apply minus f deployment.

A

And minus dev name space.

A

So, let's deploy a service also, there.

A

Correct so now I'll be having a deployment service like the dev name, space, so cube, ctrl get service, minus and dev and say I can do for parts as well and so on.

A

Okay, so what happened in the previous case when I did a curl this right?

A

This basically served me the pages from the default name, as you can see here. This is that was my default. Qualified domain name right, but now here how are the fully qualified domain name? If you look at how does the dns work right? Sorry.

A

If I look at the the dns part of it right, so what I want to continue yeah. So if I look at cat, slash, etc, resolve dot com file, so you can see here it basically. First have the service name, dot the name, space dot, hr customer locals. This is going to give you the service access to the default namespace correct. But now. Similarly, as I said that right, I deploy the app in the dev name space right.

A

So if I want to kind of go over the new series boundary, then I can say service name, dot, the name space name so I'll say that go and communicate to the test.

A

And that's my name space there, so definition, sorry dev name space and then what was the next is dot svc.

A

So it is not actually the hyphen. This is a dot here.

A

Yeah, as we see what did I do wrong. Sorry, it's engine x, hyphen, svc, sorry, yeah, correct, as you can see here now, it is searching on the part from the dev name space right, because I said now, you should go and list on the dev name space applications. Now, as you can see here now this ip and this ip.

A

What is no, I think I need to the svc right yeah this ip and this ips are 90 right, because I am able to connect to the dev name space service here.

A

Are you getting what I'm trying to convey?

A

So we have a common deer server running in the entire cluster right and as you deploy the applications, the respective dns entry get into the dns server. Every pod knows where dns server is and that's how I can resolve it.

A

A

Got it got it? Okay, great.

B

So internally, in the code should we refer with the ip or with the name.

A

B

Right name: we should refer okay, yes, correct and.

A

B

Space is always suggestable right. No.

A

No, not suggestible whatever you have in the cluster right. What are the brand applications yeah? So this is the plus side which available only inside the cluster. So what have typically going to happen is suppose in this case. Our service blue is a public eye. Public facing service right and service meter database right, so service blue can be a public service, while green can be a internal service.

A

So when I hit the what you call say when I hit the the service blue from the outside world, the blue candle to the green in the internal world right so all always my database application kind of things would always be in the uh cluster ip mode. Only.

A

Makes sense so.

B

Nepetra, um the thing is the um applications which are need to be accessible from outside you're gonna, create the cluster ap and, for instance,.

A

Then then, we're going to do some more things, so I'll talk about these dividend, two stuff right, okay, that will be for.

B

Internal communication, uh for example, between this uh node two and node three, you mean to say like um it automatically talks or we need to again create a service. So.

A

First of all, you are not about node. We talk about pods, so suppose this is a blue app right. This blue app! This is a blue color right if the blue part need to communicate to the green application right. It is not that blue part of the green part directly it'll go by a service layer. Only correct, typically, we'll have a service layer on top of pods.

B

A

And we hit our service there, only which would take me to different parts behind the scene.

C

A

B

When does it comes into picture.

A

Like see the same is here only right: this is the cluster right right, ah which is inside the cluster.

A

This is, we refer, the cluster.

C

A

Which is available only inside the cluster okay, one more point I want to mention is this: is we talked about one role of q proxy to get the excellent traffic inside the cluster q? Proxy plays one more important role here. It basically connects the uh services uh to the back end part. So what's the actual traffic see what's happening is whatever mapping you are doing here right. This is all logical mapping in a way, but you actually need like a packet transfer right.

A

So when I kind of connect right, uh I need to have a packet flow saying that if, uh if I'm hitting this particular ip, I want to go to these parts behind the scene, so they're kind of some kind of a networking rule which enables that- and that is what cube proxies- also help us doing that. So what q proc does is q proxy configures, the firewall rules?

A

I assume that you know firewall ip tables on the linux environment or anywhere so q proxy, basically configures a firewall rule, saying that if you hit the cluster id, if you hit the cluster ip this ip, you would go to any of these parts behind the scene. So this is the network connectivity or how the packet would flow right. It has also been configured by the q proxy on every node, so q proxy is available on all the nodes.

A

As I said earlier, and this mapping what you have would also be need to be done on every node saying that if you hit this ip, I would take you to these pods behind the scene, any of them correct. This is again q proxy's job to configure it and.

B

You can there, I have a question here yeah. So what is our algorithm? It uses a random. It's a random algorithm. Okay, it is no. It is not as specific.

A

Yeah yeah you'll not configure it as well. It's a default one. Okay, it randomly sends to the available box correct, correct yeah, but.

D

A

Not kind of try to just uh bombay.

D

A

B

A

In a very decent way, so there's.

B

An internal logic they.

A

Have it? Oh, okay, thank you. Okay, this is what we have as a type called cluster ip.

A

uh Then we have uh so this is only for internal purpose, but now, let's say for kind of expose these apps publicly right. So let me simply do one thing here: I want to actually set up a cluster on distillation as well, so that I want to show you some more things so meanwhile I'll go that lab. It would also this will also get over.

A

So I'm just going to log into my cloud provider and I'm going to create a.

A

Service type of oh sorry, I'm going to create a cluster here, so I'll go here, say, create kubernetes and I'm going to create a very minimal cluster just for our testing purpose.

A

This is how I'm deploying the managed cluster on coupon distribution. So let me go here: choose my whatever the version type, the location, the nodes, I'll specify very basic one, and then just the hit button. So it's going to create the cluster for me. So let me kind of go back and explain you. What node port is and I'll come back here and show you a few things? Okay, so what we did earlier is that is how we have exposed. I mean, in the previous case, the service, where available only inside the cluster.

A

Now, let's say I want to make my apps publicly available so here I am deploying the service type of node port here right so now, in this case, um I would basically create a service and say service of type node port here and I can so what happens the type node port here, one port is being blocked on all the nodes. So if you have run this docker command right, which says like docker, contain that on minus p 8000 to 80 right. So typically, if you have run this command of docker.

A

We don't have docker installed here, but I'll just show you the command.

A

So if you run this command write, docker container run minus.

D

A

Right, minus d, minus p, eight thousand to eighty and let's see an engine x right, so what this particular container would do is we would it's going to deploy the cutter of nginx and map a port 8000 with the port 80 of the container correct that is what's going to do correct so, uh which means 8000 is your nodes on the nodes you're going to hit? So similarly, here in this particular case, I'm basically exposing a port on all of my nodes here.

A

So I'm going to expose a port on all of my nodes here and when I hit that node port. So when I say that I want to deploy my service blue of type node port, a port is being blocked. Like 8 000, we saw on a single node think about the same. 8 000 is available on all the nodes, so it kind of exposed on all of my nodes and when I hit my nodes ip on the node port, correct that traffic is going to come to my service layer and to the pods.

A

It is not necessarily that I need to have a pod running on this. Third node, because I am hitting that node. No, you can hit.

D

A

Of the node on the node port and from there that traffic would come to your service layer and to your parts.

A

D

Let me show you the example.

B

A

So, in our case now in the case of service with the type of node port, so here.

C

What we're doing.

A

Is we are we already have a deployment created, so we can skip that part now here we are creating a service of type node port here type output, and we are hard coding, the port number here right now. This port number either can be hard coded. It should be between 30 000 to 32767. Some ranges there or if you skip it, then from that range the random port would come up.

A

So currently, I have given a hard coded number here, so let me apply this so this way I'm going to create a service of type node port here, which creates an internal cluster ip. Also by the way- and now there is- a mapping- has been done here so which means says that now, if I hit my nodes ip on the node port right, the traffic is going to come to my service layer on port 80 right.

A

So if I hit my node on the node port I'll come to my service layer, as I said that earlier right, I don't need to hit the node on which board is running. So if I hit my blue service, specific node port on this particular node right, the traffic would come to the service layer and from there the response would go back correct.

A

This is how the thing is being uh configured so now. What I'll do is I'm going to find out my node's ip address? So if I run this command here, cube ctl get nodes, minus or white. Let's give my nodes ip address and if I hit my nodes ip on the port number. So if I do a curve nodes, I p 10 0, 1, 32 and port numbers. 3 1 you'll see that I'm going to get the page here correct.

A

Make sense so you now go to hit the node type in the node port and.

C

Then you get your.

A

Applications now, let's see if our clusters up and linux there's still, I think, get created, but let me download its cube config. Meanwhile,.

D

Let me just refresh this page.

A

Yeah, so I'm just going to I just copied called the q config file here on the other thing, what I have so now, I'm going to go to this particular, for example, slash tmp do config, so I'm just copying the config file of my digital cluster paste. It here set the environment, variable export capital, q, config 0.2, slash tmpto config that and if I do cube, cpl get nodes. Now it should give me three nodes. If the cluster is up and running.

A

Yes still only one node has come up. Let's give it a few seconds, it would bring my other nodes as well.

A

Again I'll put a watch flag here yeah, so I see that now two nodes are up correct, so let me now deploy the service here. So if I'm going to this is my other cluster right. This is not my local cluster. This is my remote cluster on distillation, correct.

A

So I have got two nodes up and running. Third, one also will be coming up, but meanwhile I can deploy the application so I'll go ahead and deploy my similar app so apply. Minus f, um deploy.

A

Start this I'm doing a deployment creation, then I'll deploy my service here.

A

So cube ctl get uh nodes now, so nodes are up, cube, ctl get pods or should be up is running as well yeah, and if I look at myself it's also up right, so we have got our app get deployed it's available on the node on the node port here now. How would I access it? I'm going to find out my nodes id so I'll. Do ctl get nodes minus or wide here so now you can see that I have a keynote cluster.

A

They have three public iep addresses correct, so I can hit any of the node on the node port of 31.09.

A

So let me find out on which node the part is running and I'll hit another node in that case. So if I do this get port, minus or white, so it's running on the node, which is the third node.

D

A

So I'll pick up my first or second node's ip address, go to the browser and hit 31.009.

A

As you can see that your app is up and running makes sense, suppose you can't do this duo example, but otherwise you could have tried it there.

B

And the first thing which you did was the config exporting: uh can you they.

A

Just they're just kind of saying that I want to connect to my do cluster now at.

B

The uh digital ocean cluster.

A

Is correct, correct, yeah, correct! So how do you get that uh cube config? I just I just installed the cluster there right with by making a create button here right, and this is a way to download that file locally ah download config. That's really straightforward, right, yeah!

A

So now my clusters, so in this I'm running the cluster there, okay! Now? What do you see that? Do you see any challenge, so I have I made my app publicly available right now. Do you think that there's any challenge here or any issues? Do you see this kind of management of applications.

B

So we don't get the domain right. We are currently pointing it with the ip correct.

A

Correct https https, even even you, don't want to tell your customer right access. My applications by the ip address correct so currently the way we are doing it by this way correct so, but I cannot create a mapping of this with a dns name, for example, cy.com for example right.

A

I cannot make it because the dns name does not allow me to kind of map the iep and the port combination. I can only map an ip here correct, which means, with the help of type node port. I cannot make a dns name correct unless I create another layer on top of that which can be a let's say, a proxy, or something like that, I can do in that scenario correct, but not in the regular one. I cannot do it. This kind of a drawback of this thing got the point.

A

uh If I look at next is so this is this is of service type of node port. Then we have a service type of load balancer.

A

So in this case I can basically tell my underlying infra underlying infra to create a load balancer for me. So currently, what's happening is uh we saw the type node port where I need to access the app by the combination of nodes up in the port number.

A

Now here, what we are doing is we are saying, um expose my applications on a load on a load balancer, so this is possible only in the infra, where my kubernetes can communicate to my underlying info and request a load balancer correct.

A

So now um here if I am running an app in my kubernetes cluster, through which I can access my underlying infra and because load balancer, then what's going to happen, is my infra is going to give me a load balancer ip or a dns name, which I can access publicly on the internet right so similar thing load balancer would create a load port service to the parts. Okay.

A

So now I'm going to deploy a service of type load balances. So here we change the type of our load, balancer correct so again in this in in this particular case, this would be in the pending state all the time, because currently, this private effort does not support the load balancer.

A

But if I run the same command on my distribution cluster right where I have support of load balancer, what's going to happen, is under my networking section.

A

Currently I don't have an lb here if I go, but when I apply my application on this particular distortion. Cluster you'll see that automatically a load balancer would come here. So let me do that so cube. Ctl apply, minus f load, balancer.

A

Correct I've applied it and now you'll see that automatic load balancer would come here.

A

Come on, as you can see, it has come up right so now, once I would get this load balancer, then I would just get an ip address which I can publicly refer or map to a dns. Dns thing I am all done.

A

B

Yes, so ssl will be applied on the load. Balancer.

A

No not like that. You need to kind of provision any something on top of that. So currently there's a cell right over here, but.

B

That provisioning will be done this via.

A

There are different ways to achieve this. Actually, there's no one answer to it, so somewhere lb can be done somewhere. You can say I want to configure the pass through ssl or somewhere. We have in case. There are many other ways of doing it.

D

A

So, as you can see I'm getting through, I will get an ip address and once I have this ip address, my.

C

A

Would come to any of the node of the node port service to the parts correct, so just finishing up? Okay, let it come! Yes! It's been created. Now I got an ip address and if I, the ip address you'll, see that I have my page up right, pretty straight forward right, I can just deploy any of my app on kubernetes make a load balancer and this is available.

A

Similarly, if I remove my load service here automatically, the value would be removed from there.

D

A

So if I go here, it should be removed automatically. Okay, yeah. So now, do you see any challenge of type load? Balancer I mean we can deploy the app on type load balancer. But what do you see a challenge right? What what kind of problem do you see here?

A

Do you see any problem or not? First of all, if you see what it is.

B

One problem would be like uh we are stuck with the like particular hosting provider like if you want to like you, can you will have to use one single provider gcp or physical you're using this cloud.

A

No, I don't think so again. There are different things. Then we have some of these glbs and all those things. So those can.

D

A

I mean taken care into account. There's another issue service, nationally kind of a thing. There don't go there, but.

C

A

What what's happening here is um so currently I'm creating one load balancer per service, correct, which means, if I have 10 services, I'll, deploy, talk, load, balancers right, which may be expensive in some cases correct, because you're on the cloud and so on, you have to pay for every load. Balancer you create correct if you're getting an ip address from them.

B

A

And there's some.

B

There would be some quotas about the external.

A

Yeah correct correct, so those kind of things would come up. So that's where we have something called ingress which can help me do things in a better way. So there you.

B

Will provide the rooms and all the places.

A

Yeah yeah, so there we have so there only we have one load balancer and that can uh map it to different applications and so on.

D

A

I think the workshop, what we are doing on istio right uh in the I think. Third week, I don't know, there's no schedule there there. I think uh the team would be covering that. What is that and so on.

A

Okay, so that is something which I want to kind of cover in this training this session, um any any comments, any feedback you have.

C

F

Couple of things.

F

Only thing I want to know this recording will be provided and, and second will, the lab that we created. Can we use it after the session or develop all the all? It will be disabled. No.

A

No, so you have access to this kind of for four months. I believe, and the labs are there for 40 hours, as you saw that you have chosen number of hours, correct.

F

A

So until you finish 40 hours, it would be there correct.

D

The body will give it.

A

Will even publish on youtube as well. That's what our plan is.

F

Okay, the the recording right yeah.

A

F

A

Yeah you can. You also give a course feedback here, uh there's a course feedback formula if we just fill it up, that would be really helpful.

D

C

A

Just want to highlight one more thing is on our clouding website. We are also uh publishing different, hands-on labs so other than so what we saw as we kind of deliver these trading sites. Some of the content was to deliver is primarily focused on finishing some kind of toc and so on, but then what you deploy? What we need in production right? They would be more than what you just teach or being taught in the classroom, trainings and so on.

A

That's where we're kind of building these uh different, hands-on labs, so they are complete on its own. So, for example, one talks about a bit on http further on kubernetes correct, so we kind of touch upon that in detail, so you can also go and practice these hands-on labs uh on your own um and also we are looking for contributors. So if you would like to contribute, uh build uh help us to build more hands-on lab, suggest the topics and so on.

A

So you can also suggest that, and we welcome, if you want to author some of these content, uh which would be available to other people as well. So we already have a few authors on boarded who are building the content with us, but it kind of you can also do it. It will give you more visibility and create more impact in the community.

A

So that's something which you can think next um anything else. I want to cover yeah such feedback. I think, put on the chat here. Yeah sure.

F

One question I wanted.

A

F

After this 101 um do you have any or or if you can suggest like what are the next, um how to proceed if somebody wants to get into ml ops right, you know, I mean like. Are you coming up with courses where you have machine learning, and you know something on kubernetes and and something like that because I just wanted to understand like this is this is very basic, but then, what's next for someone who wants to learn more but specifically into ml ops,.

A

So amazon- I don't have an answer for that, but mlaps is something like that. So, first of all, you understand more and kubernetes terms which we are not covered. I mean because the time constraint and so on right, so you need to kind of look at one of the full course, maybe which kind of help you out with, like maybe prepare you for cq or ckd, which kind of gives you overall picture about about it right.

A

So like of course, of course, we offer these courses, and these are available at some prices that you can buy them. We also do some of these. uh What you call say offline sessions, but we primarily go with and teach corporates uh and train their employees there rather than doing uh open batches, but if there's a demand we'll do it, uh but otherwise you have seen the platform you have seen how we deliver so remaining content on. Other courses are delivered like that, so you can.

A

They are calling, if you want to take this, or there are other sites available, udemy and many others, so you can pick up whatever works for you, so that's something which I'll leave up to you, but especially for mls or my labs. I have not his specific course there, but once you understood the basic concepts envelope will just like a piece of kick on top of that or icing on top of that sorry. So, just like once, you understood that mlas would be also pretty straightforward.

A

On top of that, so I don't see any challenge per se specifically.

C

A

Okay sounds good, uh and if you want to guys want to get in touch, I have share my linkedin profile here. We just want to be in touch I'll, be happy to connect thanks.

B

A lot uh it was wonderful and the last part which, for example, uh someone asked the question. I guess about the ml lobs.

B

C

B

Done next I mean we missed it. I guess.

A

No, so actually foreign specific input on that. I just want to say that first, we need to kind of get aware of it overall uh kubernetes stuff in general, right like like the not say about the course, but I think that ck or ckd going to give you a basic ground at what you need to know right. On top of that amalops.

A

I don't know specifically what you should be doing. On top of that um to be honest, but I don't see any difficulty on that once you understood that concepts, I'm a loss would be just like uh something some some bit on top of that, nothing to.

B

Be worried too much about yeah, so the ckd or cka like that, would give a good, solid foundation. Yeah.

A

Foundation would be there, so you can of course take my uh this thing. Yeah uh courses there you.

C

Have the pdx.

A

Courses there or.

C

A

Is a course which we also offer you can take them or any.

C

A

Course is also fine.

B

Definitely definitely I'll uh look forward uh and also I'd like to contribute uh for this uh to author, some of the course on cloud. Okay, I'll reach out to you. Yes,.

A

They cannot complain, they can be simple, uh hands-on labs.

D

B

Yeah, how much.

A

B

Because I know that everybody's.

A

C

B

Difficult tasks but.

C

uh Yes, I mean.

A

It's not a lot of time. It's not that difficult! You can't do it just that will take a lot of time right, but.

B

If you have like.

A

Places couple of hours in a week we get a song right and you.

C

A

Work an object, pretty straightforward and.

C

A

Content also pretty straight- I just want to if I want to show you quickly how you can build the content other than so. If so, basically, once you come on board, we'll give you an author role and then you can kind of simply come here. Okay, I just want to say that what topic you want to put in.

C

And then you say.

A

That I want to have a hands-on lab. You.

C

Choose whether.

A

It's a type of linux or kubernetes labs and then it just kind of started your content. It's like pretty much like a notion.

D

Thing right, you just copy paste the content.

A

B

A

Can upload an image system.

B

A

And so on, easily the content yeah. This.

B

Is pretty nice? Okay, super.

A

Let me know if you want to color.

C

Yeah yeah sure sure.

A

Okay, I think thanks.

B

Thanks very very effective session: okay.

A

Great look for other sessions also and for the.

B

Yeah, I'm thinking about that and you try to engage more okay, sure, okay,.

A

Okay, thank you.

D