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

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Workshop3 : Argo Rollouts

Description

The speaker will be mainly presenting\talking about how to achieve Progressive Deployment(Canary/Blue-Green deployment) using Argo-Rollouts

A

So, thank you all for being present here to give a quick introduction about myself. My name is nina desai. I started my career as a database administrator and then got eventually into the day of srs space, so I am working with infracloud as a senior site reliability engineer at infracloud. I mainly work to help client to get onboarded on cloud and handle the need in regards to infrastructure design scale. Modernization as uh using.

B

A

Standard cncf tool stacks, I would say, like kubernetes clouds, terraform and monitoring stacks like plg elk data and, yes, we are hiring at intra cloud as we are scaling. So if anyone is looking for a more better opportunities, please do hit our website and check on your career page, okay. So just before, starting with argo rollout itself, argo rollout is not the only argo project. There are.

A

These four main projects are go workflows, argo, cd, argo events, ergo rollouts and argo project lab is like a lab r d kind of environment where new products are getting uh developed and so on. So this argo workflow, basically uh just to give some hint, I would say it- helps you to create and run advanced workflows entirely on kubernetes, uh mainly this gets used for etl or ml related job processing.

A

Then uh argo cd, as just like a flux. You know this is a cd tool which helps to implement githubs and always make sure that your, which would be like a source of truth, would be always in sync, with uh your clusters desire state. Even if you change some object, it will always try to reconcile it and bring it to the again or the same state as it is mentioned in your git uh argo rollout. We would be talking uh in.

B

A

The same topic today, it's a custom, kubernetes controller, that's what I would say at the moment and argo events. uh It basically acts as a, I would say, triggers uh based on the event uh using the kts as a platform.

A

So it can listen to any events uh like github or gitlab notifications or any file based or any kts events, and then it can trigger action like it can trigger your lambda functions or it can create your greatest resources. It can send some or other sort of notifications as well.

A

So that's, I would say some uh context about this. Particular these particular projects. Okay. So this is the agenda of our today's session. uh We will go through what the progressive delay is, why we are choosing argo rollout itself, uh how it works its installation. We will do lots of hands-on around four different areas that I believe would be able to help you to give you a proper gist and understanding around how argo rollout can help you to achieve progressive delivery outside.

A

uh So how I came to know about this term called progressive delivery or argo rollout is. uh I happened to work on one of the healthcare industry's client, who came to infra cloud with the ask that they would need help in achieving minimal to zero downtime deployment and with possibly uh minimal inter uh human intervention. I would say- and that's when, uh while working with our solution architect, I came to know about this progressive delivery term, and so eventually our rollout.

A

We, if I am not mistaken, they had almost 16 17 micro services with their application and we helped them to achieve the candy deployment using argo roll out for the sake. So uh I think it's been close to like last five years, mainly, I would say we have been all hearing quite a lot of time. Our times about cloud, devops, ci, cd pipelines mainly, and I think uh every single company that has been using ci cd has seen the benefit of it. I recall the at the start of my career.

A

We were doing the deployments like once in a two months or three months from there till the time uh recently, I saw like at least seven to eight deployments in our day on environments being carried out. So this is the outcome of using ci cd process. I would say continuous integration, continuous delivery, and uh I think my personal views are the world should move ahead from cd specifically and look for a solution like progressive delivery.

A

I predict maybe at least a year or two every company would start to now get onboarded onto the progressive delivery side. Some of you might be feeling that we are just starting with ci cd, then. Is it again good to start some or adopt some new other methodology?

A

I would say yes, because this progressive delivery is based on the core tenets of your cd.

B

A

So the reason why I would say uh one should go for this: progressive delivery is with ci cd pipeline continuously, delivering the I would say, software updates or upgrades to your system, though we are trying to bring more customer satisfaction by bringing some more features at the same time as uh as it is also been mentioned in uh google's sre book as all right. Most of the systems instability related issues does occur because of the change.

A

The change in terms of these software deployments that we do. Does that mean that we should stop doing those deployments answer is obvious? No, we should continue doing the deployments of your software upgrades, but while moving with the pace, we should have some sort of control over the same, and I think that's where this uh that was the underlying thought behind having this progressive delivery as a term, uh I think james governor from redmonk was the one who coined this term actually.

A

So this is a kind of you can say. Modern software development uh life cycle that's uh again been mentioned, uh been built on a core tenants of ci cd itself.

A

It basically allows you to pursue cd or continuous delivery with quite safer way and the big giant companies like google, facebook or microsoft amazon. They are already employing the progressive delivery at scale. uh For example, you might have seen sometimes this used to happen that you would be able to see some feature, but maybe someone of your colleague is not able to see that that's because they are rolling it in a progressive way and how these big companies start is.

A

They will basically start to roll out their new product market by market to a small percentage of users in each market. They will often start with the small english-speaking countries, like lexington, new zealand, before rolling it out to a larger market.

A

This help them to do a b testings as well, when I say a b testing, as in very simple example, if I have to put it like, let's say they are not sure like.

A

If we should put a login button in the center- or it should be at the right side, what they would do is they will roll out both the versions where, in one version it would be on a left side in one version, it would be at the center to a different geographical uh geographically based users and based on the user uh assessment, I would say, or uh the metrics they will derive from that they would be able to decide. Okay, I think this version would uh scale better.

A

This is what the users are liking more okay, so how progressive delivery is different than continuous delivery? I would say uh main difference is with progressive delivery. uh You get a ability to gradually release your feature to a progressively larger percentage of audience, and that start with. First your internal teams, then maybe better testers and then subsequently to a large number of audience, and uh if you detect at least even a one percentage of errors or performance, spikes or user backlash, you would be immediately able to roll it back here.

A

Impact would not be that high, like blast radius, would be always minimal because you would start with one percent of audios. If you are able to see. If any issues are there, you would be immediately uh roll it back automatically and uh that's like only one percent of maybe really lost, and maybe one percent of customer satisfaction issues compared to if it would have been given to like all the audience in one go.

A

So that's the core, I would say characteristics of uh progressive deliveries.

A

uh I hope everyone is able to get some just or understanding around how progressive delivery is bit different than the continuous delivery.

A

C

I have one question uh so when we try to do a kennedy deployment and we use header base to make sure only the specific changes has been uh the new change right during the canary with the header base. Only a few people using those headers will be able to get the new release. So is it something part of it progressing? What it's still a cd? It's.

A

It's a part of progressive delivery. I would say it will be part of your story right yeah. It would be a part of your progress because.

C

We have only one region, so we cannot target uh like region, ways right, so other alternative. What we are thinking is only to a small people for the beta means. Like a testing team, we use that so that is considered as a progressive itself.

A

Right right, uh based on that only I have added one hands-on as well, where I would be doing similar stuff what you said: uh that's how even we did for uh internal testing when we were uh delivering this solution for our client, where we use this uh header-based routing, so that the client's qa team would be able to use it and assess it uh way ahead before rolling it out further yeah.

C

Got it just wanted to know the difference.

A

Okay, so uh before going in depth about, uh I would say, uh argo rollouts. uh Let's understand how and why uh we need to have something like argo rollout uh when kubernetes itself is helping us to get rid of many pinpoints. So in kubernetes we all know rolling, update, execute. These are the two strategy.

A

Rolling update is a kind of uh default strategy, and uh so, even before that it was having a big bang kind of uh deployment strategy, I would say in the in the regular world where we were taking down time, we were kind of eliminating old stack and recreating new stack to avoid this kind of downtime uh kubernetes employed this strategy of rolling update, where, let's say when you create a kubernetes deployment, it internally, create your replica set and that replica set will eventually create all your ports, okay and uh when you let's say, want to go from version 1.0 to 2.0.

A

What eventually happen is, once you update your deployment, it create another replica set with version 2.0, and this replica set, I would say, create another uh pod with version 2.0.

A

Then it will do a health check based on the readiness or line space or anything that you would set and if it is, if it does find that it's looking uh good, then in that case it will start uh receiving traffic and one part from your older replica set would get deleted and that's how one by one, uh this rolling update strategy, work and eventually, in the end, you would be able to see all your parts from previous replicas had been terminated and from the new replica sets, are being working and receiving your uh larger traffic okay, so uh there would be a question like so this is already helping me in certain way to do a kind of uh minimal downtime kind of deployment, then why we would still need to have something like progressive delivery.

A

The reason is, I would say, even though rolling update has already helped us, but there are some of these limitations like you would not control directly the speed of rollout. Second, you would not be able to uh control the traffic flow to your new version compared to your older version.

A

For some scenarios, uh your readiness probes are kind of, I would say unsuitable for deeper or stress test. uh You would not be automatically employ uh ability to query external metrics to decide whether your update has been completely working or not, and the main thing is, it cannot automatically roll back.

A

These are the some of the limitations with rolling update strategy and that's where uh the original contributors of argo rollout uh thought of developing something that would help to overcome these uh challenges with rolling update strategy, and that's how argo rollout got into the picture. I would say so what it is. It is eventually just a kubernetes controller uh and a set of crds.

A

It does provide you deployment strategies like blue, green kangaroo, kangaroo with analysis recently they have added experimentation which would be more or used for destructive tests right, and uh if you want to compare what this argo rollout is it just completely, uh you can say, dropping replacement of your kubernetes deployment like for deploying any of your microservice. If you are currently creating kubernetes deployment, you can create a rollout component for the same it just exact, uh I would say replacement of your kubernetes deployments.

A

uh So these are some of the use cases about the argo rollouts like you can run last minute, functional test on your new version. uh You can compare the performance of uh old virgin vs new version. As I said right during the a b test you can do and you would be able to control how much traffic you would want to redirect to the new version of the service. So this is how the I would say, architecture of argo uh rollout is about. uh I would just try to put them in very simple words.

A

I would say like when we deploy nginx right. Nginx. First thing we deploy is a nginx controller who basically control all your ingress in terms of engine x. Similarly, for rollout as well, you would be deploying a rollout controller.

A

Job of this rollout controller is to check if there are any resources, been deployed in your kubernetes cluster of kind uh rollout and make sure that this rollout state is always been as it is uh that you want it to be, and uh apart from that, this just like in deployment right when you create a deployment it create your replica set and that replica set creates your ports.

A

Here, you can say it create a stable replica set that would be running your older version and when you create an new version, it create a kind of candy replica set, and that manages all your pots.

A

Okay, uh ingress. uh This is not directly, I would say part of argo rollout, but just to make it more understandable. uh Ingress is any uh english any service mess. I would say it is your nginx controller or any service meshes that you are using.

A

uh Those would be able to help you to route and uh roll out traffic in percentage way for your targeted micro service and about this analysis template you would be able to understand this concept more during the hands-on, but for doing analysis.

A

Ultimately, uh during your rollout, you would need to create the analysis template uh in this analysis template. Basically, we will define what kind of analysis it should do to decide whether it should roll out ahead or it should roll back your deployment automatically, and for that purpose it create a kind of internally analysis run component and uh that analysis and component can do any kind of analysis.

A

It can use your promises matrixes or you can create your custom kts job as well, which, if I would say, can do some or other testing that customized testing that you would want to do before deciding whether it should roll out ahead or.

B

A

uh Clear to everyone about how this works, yep, okay, uh thank you. So, uh let's get our hand dirty a bit. Let's do argo, rollout installation, it's very simple uh installation. I would say you just need to create a dedicated name space as we do as a standard practice in kubernetes and deploy your argo rollout there. So.

A

So I hope uh you all would have got access to these lab setups right. Just click on this.

D

We need to share this link with them. Let them sign up and then do this just share the link on the chat. Okay, I'll share the link on.

A

The chat sure sure so nipendra would be sharing this link with you. uh It's a platform being developed by cloud yoga team now very handy. I would say in terms of using for doing all your tests or learning yeah.

D

So it's a it's an environment which you can try it out again: there's no compulsion that you should be doing that, but it's like your entire workflow and this would happen with very ease. So we appreciate, if you kind of log into the platform and then open that link to kind of uh get the workshop. So we'll give you guys, maybe a couple of minutes to sign up and then take it from there any questions. Meanwhile, that would be good to have.

E

So, with this uh rollout, we won't be able to control the traffic right. Here is the the number of ports that is in ready state. That is what um this rollout controls right.

A

uh Yes, the number of parts plus you would be able to control the traffic as well in certain way it has that strategies and mechanism available, using which you would be able to control that traffic.

A

But in the essence I would say yes, it control, basically the amount of pod you would create and yeah. It helps you to route the traffic to the new version. It would be able to control that traffic flow as well.

E

A

So once you would be able to access the labs, you would be able to see this kind of environment been provided uh provided to you.

D

A

D

Give it one more minute uh in.

D

Just say yes or thumb, plus one on the chat, as you have all set up, so we'll wait for at least five thumbs up to kind of take it forward so till she got it. So, please, let me know, let us know at least we are waiting for five years before we continue um yeah. I think we are good if all of you have triggered the lab go ahead. I think uh you can just talk about briefly about the how the labs look like and then we'll continue on.

A

Sure so uh guys you would be able to see here. uh Some of the I would say, facets about this platform. I will show you like here. You get this terminal where you can run all your regular linux command. Anything that you would like to do.

A

Apart from that, it does provide you the id as well. Just like a vs code editor, you would be able to use it. We would be using it anyhow in uh one of the hands on here. You will be able to see it will open you a good uh code editor just like this.

A

Let's install a yaml, I would say extension: it would be helpful for us later from visual perspectives, so this root, for is something that you are currently able to see here on our labs here. Only okay, so I hope now all are logged into this console right. We would be going through these all instruction set. One good thing uh is all instructions I have already copied here. uh You don't need to run any command at such shear directly.

A

All that you would need to do is just click on these green signs and they will do the job for you. They will execute commands for you there throughout the instructions. Apart from that uh to access the argo rollout console now, we have already been open for the same. You would just need to uh click on these urls right click and open it into another tabs.

A

Okay, so I have copied uh all the argo rollout examples that I have created into this uh git repo. This is a public repo, so you can fork it uh and if not, then you can just clone it directly.

A

Let me show you how you can do it, it's just that you so right now you are able to see these were like a sample things that I had created in past. Now. Let me clone this repository.

A

You would be able to see it is cloning, and if I do this, you would be able to see the it has already cloned the repository. You would be able to see all the examples that I have added would be available here.

A

Okay same you would be able to see on this console as well. You'll be able to see all those examples been added here and if you would have already used the yaml extension you would be able to see in in good way. I would say in this color formats.

A

A

I hope all are able to clone the repository right.

A

If, yes, then, uh we'll move ahead for the next instructions, so we would need a helm3 uh client. It would be used uh in one of the example to show how you can use having itself with the rollout and so just install the helm, client as well.

D

So, rather than the appreciative thumbs up and say yes on the chat as we kind of go right because uh being on remote, it becomes difficult for us to kind of judge, what's happening there. So if you be more interactively really helpful for all of us, but that's your choice here. Thank you.

A

Do let us know in case you face any issues, so if you would be able to see now we have install helm as well on our machines, you'll be able to see if it is being deployed, and we were originally about to do the argo installation right. So for that we would create a argo rollout namespace, as you are able to see it here. So let me just create a namespace for it.

A

You would be able to see it has created namespace, let's verify if the namespace has been created correctly or not.

A

If the argo reload namespace has been created, uh so argo rollout directly does provide a url from where it can install this install.yaml file. I have just copied it uh for my own purpose. You can directly use the urls from their official installation document as well. So let's install the argo rollout just prior that you would be able to see. Currently there is nothing running at the argo rollout inside or reload namespace. So now, let's deploy the controller, so you'll be able to see it, create crds and cluster rules, and so on.

A

You would be able to see now different objects been getting created here.

A

Okay, so if you see it create a replica set, argo rollout deployments and a pod, so let's just you can refresh it by just clicking again on the same command. You would be able to see its status, so it was in container, creating state.

F

A

Is in uh running mode, but it is yet to be ready yep now this is ready. So your argo controller is been installed now, uh just like to interact easily with kubernetes. We use cube ctrl plugin right to interact with argo rollout. You would need to have this argo rollout plugin, so we would uh deploy the argo rollout, qctl plugin.

B

A

A

Yeah, so you can see the argo rollout has been uh plugin been installed, it is showing its version as well.

A

Now argo, the rollout does provide its own dashboard as well. uh It listens on port 3000. So what I have done in this lab is, I have already opened it on port 3000, just click on this and you would be able to see uh argo rollout console available.

A

Just click on this url and open it into the new tab.

C

On young one quick question: we installed this ergo uh command line right, so this is part of cube, ctl, rollout or you're using the crd. It's a command line overall, because you're uh doing with uh cube ctl argo rollout right.

A

Yeah so like uh the only difference, I would say when we execute cube ctrl command, we do like cubes it'll get deployments right. uh This plugin uh is cube. Ctl argo rollout get uh rollouts, I would say so.

C

Because for argo cd we have a command line for argo cd right, so I thought it will be in that way.

D

C

B

Can do that as well.

D

Or basically, it's like that, you can insert the plugin that I think what the ad is doing is a very easier way so that you don't require to remember the rbo cd command and so on. You just go with the cube city elevator itself, so it's it's a personal choice. Processing, there's nothing changes per se, even if you go with the argo cd r1 command or you go with the cube, ctl command. It's similar. Okay,.

A

And uh just for sake of showing, I would say or you'll, be able to see all the commands provided by argo rollout here. Okay, we would be using uh them in our demo further.

A

Okay, so currently you would be able to see it as a blank, and it is just normal: uh don't get confused once we will create rollout objects, you would be able to see them here go back again, so we have done the installation of our controller. We will move with some more uh details about it and then again, come back so we'll move in this way we'll go into the slides, then we'll go back and do hands-on on the concept that we have learned. We have seen already how our goal rollout work.

A

We have done now the installation of rollout as well right. This is how the ui would look like. Once you create your rollout objects, uh you would be able to see it very soon.

A

So in terms of rollouts, these are different. Specs been available. uh All these. You would be able to understand uh much better way when we will be doing uh the hands-on mode- okay, but just to give a quick comparison uh compared to the specs of deployment. The difference is uh in case of our kubernetes deployment right. The client is deployment app. Our api version is also different. These are the two different and the main difference is the strategy you would be able to see here.

A

This is the strategy like you can use blue green strategy or in case of candy you would be a setup, candy deployment strategy as well. You would be able to mention the candy service table service, uh analysis, templates and so on, uh and for the traffic control as well right, you would be able to add this traffic routing parameter. I think someone was asking so argo rollout help in this way to do the traffic routing as well.

A

We would uh get to these things more in uh during the demo. I would say you would be able to understand in much better.

A

Okay, so uh moving ahead, so blue green uh is one of the I would say uh main uh strategy provided by argo rollout.

A

I think some of you might be already familiar with this concept of rollout, how it blue green rollout work is, let's say: currently you are running application version, 34, so you'll be having some load balancer and all your client requests are going to via this load balancer to your application version. 34 now, let's say you want to deploy a new version, so what you would create is you would deploy another set? Let's say right now for version 34, you are running three parts.

A

Now you will create another three parts which would be running your application version 35. Okay, so you would be having your old version and new set of application version. I just that it is not yet receiving the new traffic. What you would do is after let's say you do some tests internally to check that new version is working.

A

Well, then, you will cut down the traffic from old version to the new version, and once you are able to see that it is correctly working well, you can set a timeout value after which this your old version, I mean to say old parts of the applications would terminate and the new version would be completely running.

A

Is everyone clear with this blue green concept.

A

So these are the different specs. I would say in case of you want to do a blue, green kind of deployment. Main thing, I would say, is focus on this rollout type, api version and the strategy part. The rest, all part you would be able to see is similar to kubernetes deployment that we do. Okay in strategy in case of blueprint deployment, you would need to mention it, which is your blue, that is old version or active. Currently active version and preview service would be your new version of the service.

A

Okay and uh currently, in this spec that I am showing I have stopped the auto promote. So there is this parameter been provided by argo rollout so which will make sure that it will not automatically cut down the old active service and use the new service. Here we would be doing a manual intervention to promote this in a to promote the traffic from old version to new version. Only after we are sure that uh the new version is working well.

A

Okay, the only thing different with canary here is no live. Production traffic would be going to your preview or new version of the service. I would say that's the main difference between blue green and candy.

A

Okay, let's do some hands-on in the same lab that we did earlier.

A

So if you would be able to see here after argo installation, uh there is this option available for blue green deployment, where you would be able to see the specs that we have created, so we will be creating just kind of like the deployment we would be creating a rollout object. Where I have mentioned. I want to run two replica set of this version and it will be basically running one sample uh containers which would show you on a console, the blue version.

A

It is running on port 8080 and the strategy it is using is the blue, green and in blue green deployment strategy. You would create the two services. One service would be your active one, and one is your preview service one? If you see nothing is changed except their names, and uh here for demo we would be using nginx controller, so we would be creating an ingress uh which would route our traffic to our services, so how the flow would be from your client request. It will go to your ingress from ingress.

A

It will uh start routing traffic to your initially stable version, and once we promote it, it will cut down and start moving all the traffic to your new version.

A

Okay- and I hope all of us are already familiar that basically service is able to find out its pod using labels and selectors same thing happens here. Initially, your active service is having those selector labels matching using which it is able to route the traffic to your active version and what you would be able to see once we promote this deployment, your previous service would start receiving traffic as well as your active service would also would start routing traffic to your new version. Only okay, how we can do the demo is just execute this command.

A

What this command will eventually do is in this console.

A

You would be able to see in this example. All these objects, basically, it will deploy. We are doing cube, ctrl deploy only okay. So, let's.

F

A

Execute this command.

A

Which will deploy these objects? You would be able. We are currently just for testing purpose, deploying it in default namespace. So right now we are deploying, let's say your blue version of the service. uh It is creating two services that we have told him to do. We have. It is creating two parts, as we have requested for two parts.

A

You can refresh it here. Okay, and now it is in running mode. You can access the console of this application by just right click and open it in the new tab.

A

So right now just for a demonstration purpose, you would be able to see the container that we are running. It is showing you all these blue dots, okay, so that means you are running. The blue version of your application.

A

Let me know if everyone is able to follow it till here. You guys are able to do this demo so far,.

E

F

A

So now, let's uh now, let's go back to that argo rollout console. If you will refresh this page, you would be able to see it uh running on.

A

So you would be able to see earlier. It was a blank console now it is showing the rollout objects that we have created here. Okay, it is currently running revision, one so going forward either you can use this console, argo console or you can use the command line as well. I would prefer command line doing and from here you would be able to roll out further. You can promote or you can abort roll out as well. So now, currently those two blue pods are receiving traffic.

A

That's why you are able to see all the all those blue dots on your machine on on the client front. End. Okay- and this is currently your stable service. So what is happening is uh from ingress. It is uh going to your stable service from stable service. It is getting uh routed to the parts running the blue version. Okay.

A

So now, let's move here and deploy the green version. So what right now I'm gonna do is uh you would be able to see this way? uh The use of cubesat argo, rollout plugin? If you would be able to see it has created a replica set and that replica set has created two ports. It is running with that capacity. Only now, let's change the image so that it will create a blue version of your green version of your application.

A

Currently, you are able to see it is running two parts. Now, when I will deploy the green version, it will create another version of your application but running the green version. I would say.

A

B

If I will run this again,.

A

You would be able to see it is creating the new uh two new ports which would be running your green version.

A

Okay, so these are runnings, you would be able to see on a console side as well. Now the new revision has got created. It is showing the new revision is uh able to run these two parts that we were able to see there, but in console site it is still showing blue version reason, because we have not yet shifted the traffic from blue to green. That's what I was referring earlier, that your live production traffic would not be.

A

I mean you would not be able to do the testing of your live production traffic directly uh in the blue green version. I would say, unless what you can do, if you are uh going to off the blue green is uh we would be able to use a traffic routing mechanism like nginx or any other, and do you can use the custom headers to pass the traffic to the your new version preview service and able to do some tests on it?

A

We would be doing demo on that. So now, let's do the promotion. Okay, when we do promotion, what will happen is traffic will be routed from blue version to green version? And let's do that here once so, I will do the promote okay.

A

So the moment we do promote here.

A

You would be able to see now your green version of the application has started receiving all the traffic client traffic compared to your initial blue version.

E

I have one question yeah: when I check the labels of those boards, it is showing rollout dash blue green and have two services right. One is um selecting the uh blue green, oh, both are pointing to blue green, only got it.

A

So so the the way I mean this argo rollout does, is it use? This part template hash value? Okay, you would, and it changes the port template hash value so that this service, that your active service that was running there would also now start pointing to your new version of the box.

E

A

That's how it works.

B

Thanks yeah, okay.

A

I hope everyone is clear till here.

A

Now you are, you are still seeing uh both the parts and uh if you see after the promotion, it has now started terminating the old parts as well. That's why the revision one parts have been uh removed. Now you can set a timer value for it for how many long hours like that would be a standard practice.

A

I would say make sure that your old version is still available so that even in case of some issues, you would you don't need to again create the application or the old version it just that you need to shift or roll back the traffic routing you would be able to see earlier. There were four parts. Now it is back to two parts because it has deleted those older version, parts.

A

Okay, so let's now delete this setup, so we will do cubesat will delete and it will basically terminate all your parts. Everything will that we had created so far.

A

So we don't need this now.

A

It's a blank one only again.

A

So, let's go back to slides again, we will see now the concepts around. How can we deployment does its work? Okay, so just before moving that I would say these are some of the pros and cons that I see in terms of blue-green deployment. uh You get client apis consistently. That's a plus point, I would say your previous stack can be tested even before your live traffic goes to that and rollbacks are immediate, because your ports are already been running as a standby mode there.

A

So in case of issues, you can just switch the traffic and roll it back and you would be again running your older version, but yeah. It comes with the cost of running 2x of your resources, and you cannot do a fine grain canary kind of deployments with it. uh This cost you can reduce, as there is one parameter available which would help you to make sure that how many number of parts of your new service you would want to run.

A

You can limit it uh in certain way to reduce the cost, but yeah it will still incur a bit of cost for you. So moving ahead with the canva deployment concept, how it work is again the same uh example. Let's say you are running your application version. 34, you are receiving all this traffic where your load balancer.

A

What now you do is you will create a new version of your application. Let's say again a green version, but you would not be routing all the traffic you would be initially creating only one for of your new version. You would route it a one percent, ten percent or a candy traffic.

A

If you are able to see those requests are getting correctly routed, then you would increase number of ports and also you would increase the percentage of traffic that you are routing to it, and once you are happy with it, you would be full-fledged using the new version here. The cost factor will not be affecting much because, let's say currently, you are running five uh ports as part of your application version 34 and you deploy the application was version 35. Initially, you said you would need only 20 percent of traffic to be routed.

A

It will create one pod alone and it will terminate one pod from the old version, just like in case of rolling update. So that's why your pod count will be always what you are. You have asked as a desired one. Okay uh in terms of spec earlier we had uh defined blue green here here you would be defining canary uh in case of rollout, and you are using traffic routing mechanism. You would need to define both these things.

A

You would be able to see it in the third demo that we would be doing, and here, as I said right, you would be able to control the traffic that you are routing. Ultimately, the number of ports that you are creating. Let's say I am running five ports and if I set weight initially to 20, what it will do is out of old five parts.

A

It will terminate one and uh create a new pod of new version, one uh you can add up manual pause, as uh we did in earlier case as well, and you can do some testing on your new pod. Only one percent- or, I would say 20 of your traffic production live traffic- would be going to this new version. Okay, if you are happy with the results, then you can set further weights and accordingly it will.

A

We create new ports at the same time, start terminating the old parts and uh you can set up the duration as well for the same so yeah. Let's.

C

A

This more using uh doing some handsome, so here you would be able to see uh this is this option b right this? These are the specs that uh I just now defined these you can set a weight, whatever amount of steps you can add in that right now. In this example, uh we are not doing nginx or any other traffic routing mechanism that we would be doing in uh last demo that we will be doing okay yeah again, we would be creating one in case here uh for this demo.

A

You just need to create one service and that service will route traffic to both.

A

So clear to all: uh let's just do a demo of it again uh this example. You would be able to see- I have already created here. This will be available for you. Even you want to play around it later.

A

All these english and all are being available here.

A

So, let's just deploy this candy version, you would be able to see the it has been created.

A

So we have defined five as a replica account and that's why these five parts are getting created here.

A

These are running you again can access your application by you go on this url.

A

Again, you would be able to see the blue version is running.

A

F

A

Able to follow till here, let me know in case you are facing any issues with uh with demo. So now we are initially running again the old version of our application. Okay, now again, if you want to access it via gui, you can go and use the rollout app console.

A

So you would be able to see again the rollout objects been created here. You would be able to see what things we have defined there. Already five parts of old uh version is running. We have defined the steps as well for canva how it should progress ahead. Okay and now, let's do the promotion of new version.

A

You will be able to see it is currently running replica one which is a replica set, one which is running these parts. Now I am changing this time image from blue to yellow, as you are able to see it here,.

A

B

Has been updated.

A

If you would be able to see here now in terms of pod, one old pod is terminating and one new part is getting created here. That's why it will always maintain a replica count of five.

A

You would be able to see your new version. uh Yellow uh dots are also coming here in terms of revision. As you see here, it is waiting now on pause. Initially, you were running five ports right now. One part from old version is terminated and new pod has been created with the new version, and that's why so out of five parts, four parts will be kind of showing the older version. One part would be showing your new version and that's why you are able to see very small amount of yellow parts running in here.

A

A

This is already been shown there and now, when we will do a full promotion, I would say it will start terminating all those old parts. Basically, it will move ahead with these steps. We are kind of doing unpause here for uh this step, so let's just promote it.

A

Keep an eye on in terms of pods. You would be able to see it is stimulating one by one old parts. You would be able to see the it is increasing. The pod count here best way to look out, for it is via gui.

A

You would be able to see it has created three parts, so it is terminating one part. It is moving ahead with the versions you would be able to see. Now soon it is running most of your yellow version. Only.

A

Okay, since this is just a demo, these uh weight size are been kept in, like in terms of seconds uh in production. You would increase uh this value to a much higher number. So now, right now, your entire new version is running.

A

You would be able to see it will turn it has terminated all the old ports and created the new set of ports.

A

Okay, so that's how the canary does work it. You can control uh to certain extent uh the traffic routing.

A

Okay hope all are able to do this demo.

C

A

So let me delete again this setup that we did created just now.

A

You would be able to see it has been removed now.

B

A

Let's go back to slides again, okay, so, as you have seen, the candy is kind of simple one. It does not require any special service mesh, a bit better advantage than doing rolling update, but uh yeah. You will not be able to do a fine grain candy wait without running number of many pods, okay and compared to uh blue green, where you would be able to roll back immediately, because uh your old version is still running as a standard in case of candy.

A

Roll back is comparatively slow uh because, again to shift from, let's say, yellow version to your blue, that is old version. You would need to terminate one part here then uh create one new part there, and that way it would go. Okay and yeah. You would need to. uh You would be doing a production traffic testing only there, but these advantages can can be overtaken by using the analysis, steps here uh for the candy purpose, okay and that's what we would be doing for the so argo rollout. Has.

A

This very amazing feature called analysis where, let's say right now, someone did ask that they do by passing the custom headers. They can do analysis here. The manual intervention is there. You can instead of ask vargo reload itself, thanks to this analysis, feature to do all those testing for you and uh based on that result of your analysis, you can decide whether it should roll out further or roll back the new versions, deployment, okay, and there are different ways available.

A

You can do a background background analysis where an analysis objects will create and it will continue running in the background and unless and until your deployment is completed.

A

uh Second approach is your inline analysis, where you would add that analysis as a step during your uh during your uh deployment and for bluegreen related deployment, you can do the bluetooth, pre-promotion analysis and post-function analysis again. The sole purpose of this analysis component is to let you automatically decide whether to roll out or roll back, and in what scenario it should do that.

A

Okay, like in case of camry, you would be able to do this way. You would be able to do a background analysis where you would define this component here.

A

You would add a template name, so this template name will ultimately create a new kts component for you and whatever things we will ask him to perform as an analysis it will do and based on that, it will pass the result again saying: okay, uh it is good to go ahead and so on.

C

Yeah one more thing: this analysis can I put it to some prometheus and check if I'm getting more 200 or 400, something like.

A

That, yes, yes, you can do that. uh I would be showing that I mean the specs of that as well. Okay, so you can use any of your monitoring tool. uh You can use uh prometheus datadog or any other. In my case, I had to uh initially use datadock. Then client uh had some other requirements, so we created a custom job as well. So as part of this analysis, you can run create your own custom, kubernetes jobs that you can trigger.

A

Okay, so compared to uh background analysis, where it will run continuously in case of inline analysis steps. You can add it here as a steps only so it will not move ahead unless and until this analysis is completed, okay, suitable, it is more suitable for doing this benchmarking kind of test.

A

uh In case of uh blue green, you can do pre and post promotion analysis where, uh let's say once your new version is available.

A

As uh you can add this as a pre-promotion, where you would be running your templates where this template, you would be running either your custom job or let's say you have some monitoring tool from that monitoring tools. Whatever you are checking as a 200 or any metrics, you would be able to do that so that you, once you are able confident enough that okay, this uh new version is been rolled out correctly. Then you would do a traffic shifting from old to new version, as well as as a part of post promotional analysis.

A

uh Even once your new traffic is been routed that time as well. You can do the analysis part again, so it has both of these capacities available with it.

A

Okay, so I think this is what you are asking right. So as part of this analysis, I can so they have providers for prometheus and for any monitoring tool. You can say and uh like in this example, we are using prometheus provider and via this provider, we are particularly.

F

A

Query that one would normally run in directly by logging into the observability tool to check if a new version is correctly running or not. Okay, like here in this case, we are defining that if it gets some 500 related errors and if it's percentage is more than uh like 0.95 as in you can say, one percent more if it.

A

If the new version is getting errors more than one percent, then we would tell our uh rollout to stop that promotion ahead and just let me take back to the older version, you can run the custom jobs as well the provider difference here like here. In this case you would be able to see prometheus as a provider.

A

Here you would pass job as a provider. We will be doing this as a hands-on now, okay, so I hope this is what you are looking for. Right.

C

uh Yes, correct.

A

Okay, because I had done the similar work where for datadog I had to install a datadog agent and basically using that it would connect to datadock, it will try to run those queries and decide whether to promote ahead or not.

A

Let's do a handstand of it in this hands-on. We will be running a small single job based on the. If that job is successfully completed or not, it will promote. If not, it will roll it back. Okay, so we will be doing both kind of demos.

A

So here I have added a demo for the same as well, so what you would be able to see here now, let me go back to this console.

A

So I have added an example for canva with analysis that you would be able to see here. Roll out service, ingress objects all remain almost same. Okay in rollout object.

A

We would be adding uh step to do the analysis. You would be able to see it here, so I am doing an inline analysis where I am asking rollout to go back use this template, so this template is been created by using the analysis, objects, okay and here just for demonstration purpose. I am running a busy box container.

A

uh We all know once busy box container complete successfully exits with status 0. So that's where hypothetically imagine uh we are our analysis. Template is telling rollout okay, I am able to do analysis successfully. Please go ahead and do the further rollout promotion.

A

Okay and later we will make it fail as well by just inducing a component which will make sure that this analysis gets paid. If we will just uncomment it, let's see both the demos again, okay, so what happens eventually is, if you recall the architecture where I mentioned right, the analysis, template create an analysis, run kind of object and that analysis run eventually create a custom, kubernetes jobs and that job would run whatever commands like it will run the create a kubernetes job that will run this busy box image.

A

So let me deploy it here. Okay, you would be able to see it has created, you would be able to see all the objects been created.

A

Again, the blue version is running okay,.

C

If I refresh it.

A

You would be able to see it is using canvas strategy. We have added a pause of 10 seconds so currently, when you are doing first rollout. It does not do analysis. Okay. Now, when we will change the image name or let's say we are doing an upgrade that time, it will do the analysis as a step.

B

A

Let me show you one more thing: you would be able to see those objects been created as well, so.

A

Currently, it is running five parts you would be able to see thanks to the analysis, template object that we have created. It has created an object, called analysis, template custom crd, you can say when we will do the rollout this analysis. Template will create a object of kind analysis run right now. You would not be able to see okay, it is showing no resource found, and this analysis run will create that kubernetes job for us.

A

Okay right now it is not showing. So let's do the promotion of uh from the old version to the new version. Here again, so, let's say instead of blue now I want to do a yellow version, so let me run the yellow version of the application. You would be able to see here. So this scanner and a check analysis template will now create an analysis run object once it will reach to that step. Okay, so you would be able to see. 20 of traffic would be routed here.

A

So now the analysis part is done. When analysis part is done, it is initiating analysis run. You would be able to see it again here.

A

Okay, this should create an analysis, run object, you would be able to see it created, analysis, object and that analysis run object created a job for us. It ran that job in our case, just a busy box container and it exited with the completed status as a exit. Zero, okay,.

A

It has promoted new and it is terminating old revision. We would be able to see many yellow, been there. Okay, I think it has promoted quite fast.

A

So if you see the new version has been coming and the old version is terminating back because we are using here as a family analysis.

A

Clear to everyone, anyone has any doubt about this demo or anything.

F

uh So nina, can we say that uh this blue, green and canary okay, it's all a set of procedures. So if you get an.

A

F

Here we can be able to do it for our own service and deployment sets correct. Would that be fair.

A

F

A

So you mean to say how I can use the rollout for doing the blue green for my own object or you are saying about creating your own custom. Crds.

F

No, no, no. We just gave here in this workshop an example how to do this correct. So basically we started with this argo reloads and you know we created the objects right. That is application, specific and all that, and then there is a technique by which you do this kind of a blue green and then you went to a canary correct. So if you follow the similar kind of steps and all that to achieve it, I think we can be successful in any other production, any any other customers as well correct.

A

Yes right, but as a I would say like I believe is that uh see the challenge with us is uh when you want to develop something it would take uh 30 of your effort, but maintaining it. It would take 70 percent of your effort. So my personal philosophy is uh use a kind uh use it as a managed service like uh if someone has already solved a problem for you either. You can leverage that but yeah, if you have some custom need- and you want to develop something like that.

A

Yes, using the same principle uh as you mentioned, you can develop something on your own.

F

uh Something or something I mean not something on your own: it is basically its water understood, it's it's a kind of a procedure right, you do it and you can replicate it or is it something more to it and all that? Are you saying that uh you, because your statement was mentioning that uh there is some 30 and then 70 is on the maintenance? What does that.

A

Mean uh okay, so initially I thought you are saying that similar kind of uh custom uh tool you would like to create on your own. That's what initially I felt.

F

No, no, no, I'm asking a very basic one see now, as you said very clearly, so let's say you know, I don't know. Maybe myself I'll put myself in this thing. I come from cd, only okay, continuous deployment. Now I want to have an idea of progressive deployment. Okay, I suppose I want to take like this and then suggest that yes, argos, argo cd rollout is there okay, it is open source and then that is also trusted.

F

Also so probably in our uh you know, let's say in a devops and all that we could try this out. We can have a demo, so we can create our objects here, application objects and all that and then we set it up like this and then we can have our blue green and then we can go to canary as well.

F

A

So this can be done right like following these steps. Yes, yes, following these steps, you would be able to do it. That's how we did it. When client came to us with the after giving proposal, we showed them. We picked one of their microservice and we tried to show them doing this way. So yeah I mean when you want to approach or propose this kind of solution.

A

uh Yes, you can follow all these steps and you would be able to show them uh how to achieve this deployment basically use this uh set of tools and do it.

F

And this this is open source or this is something like you know. They are giving some add-ons and then the page service is there circle solution. It's open source.

A

So you can use it as it is.

C

Well, with your experience right how many micro services have you have been used till now for using ergo rollout.

A

So the need that I had for the healthcare client right, I said it- it was having, I think, uh 28 or 29 micro services that we converted from deployments into the argo rollouts.

C

A

So there is no uh upper cap for that I would say.

A

Maybe what you as a principal you can follow, is whatever you are running as a stateless services right, you can go ahead and convert them into the rollouts.

C

Yeah, stateless right yeah, so no actually, my use case is where we have around more than 700 to 600 800 micro services.

C

Okay, so will it be a good idea, but now we are using istio part of it for doing canary.

A

Okay, so uh argo rollout has integration for istio as well available for traffic routing. So yes, you can do that. I would say- and I haven't seen any upper limit about how many services you can do as long as you have a good, robust clusters, and so on, I would say so. You can do that. My approach was like in my case as well like I did not directly went uh and did it for all 29. We started one by one service in a phase manner.

A

That's one thing I would say you would see uh been added as a learning that it's a continuously like iterative process when you want to shift from cd to progressive delivery start with. One then uh show the success of it then plan for a few more and that way you can go ahead for it.

C

Oh god! Well how about the response for the ergo rollout ui, if you scale it to that much extent,.

A

uh In my scenario, uh I did not get any issues. uh Client has, you can say, github action has been available so via github actions. Eventually, we were doing all these things. We did not have to use the console specifically for the same purpose.

C

You got it got you and.

A

It has a good community support as well. I would say that's my personal experience when, because it was a completely new thing for me.

B

A

Saw the community is quite uh good in terms of response. You would be able to get all the helps that you would need in case you get stuck.

B

A

Some of the companies like intuit itself right uh the one who is official maintenance of it. They are using the argo rollouts only for uh for all their micro services, and I think their number is, I believe, quite similar to maybe what you are suggesting.

B

A

Okay, so I hope it's clear till here now: let's uh delete this setup now, this time uh we will do a hands-on to see how it roll back in case it get unsuccessful right. So I have deleted the current setup that we did here. You would be able to see it has been stopped.

A

These things also would have been vanished by now.

A

Okay, so now what we will do is to make sure and see how it roll it back in case it find issues or during the analysis days uh use this ide console right here you are able to see. I have opened this analysis. You can open it at your end as well, and just uncomment this okay, we are purposefully here trying to make that our analysis should get filled because after running this command, let's say job will exit with non-zero status and that will make analysis believe that okay, my analysis is not successful.

A

Let me pass that message to the rollout saying see: the new service seems to be not working as expected, and then argo ruler will say. Okay, then let me roll it back to the old version itself. Okay, so just uncomment it save it here. Only and now, let's run again the setup thing again, let's follow the same step that we did earlier here.

A

So again we will run the same commands.

B

A

uh Let's see if the objects has been created so yeah the new objects are getting created here. Okay, you would be able to see its agenda in six seconds.

A

Now the blue version is again there.

A

The new revision has been there. The only difference, as I said right this time, is that we have added a command argument which will make sure that our analysis will be failing this time.

A

So now, let's try to promote the yellow version, as we did prior okay.

A

You would be able to see the one part is termiting, and so now one of yellow part is coming. Okay, let's focus here, we would be able to see now the analysis is running. Okay,.

A

Ideally, this analysis should get filled. So if you see analysis got failed, so it terminated. The old pod created a new port of the old version itself, so, whatever initially you, the yellow version was available, has disappeared and again it is now fully running on the old version itself. You would be able to see those analysis runs getting failed here.

B

A

Okay, it was showing earlier successful.

A

This time it showed it as a field.

A

Would be able to see it is being rolled back to the older version.

A

It says here in the commands, if you see rollout about it, to update to version 2, because metric test failed and that's why, based on the failure limit, it has rolled it back to the older version, so I hope able to see the capabilities how it roll ahead as well as roll back without any human intervention. For us.

B

A

Clear enough till this time anyone has any other doubt.

A

I hope everyone is able to follow the demo.

E

A

Yeah, okay, that's it! So let me delete this setup again.

A

Let's go back to slides okay, so, apart from this, as I said right, argo rollout does provide you with traffic management uh capability as well, uh and the advantage of this traffic management is basically, you are able to reduce the blast radius.

A

That means in case of failure uh effect will not be as uh hazardous, I would say, or as a nightmare compared to if you would have done it as a 100 roll out of the new service, and uh there are various techniques either you can use a raw percentage where you will constantly sending only five percent traffic to the new version. Only second is a header-based routing, the one that we opted for our project and the third is a mirror.

A

You can mirror all the traffic that you are sending uh in production to your the new version as well, and these are the different traffic routers already been supported by argo rollout. You can use aws alb or you can use istio, nginx controller or any other smi as well. For the same.

A

Okay, we will be doing demo as well of it, the uh in this demo, the next demo that we would be doing. I have used a helm chart, so whatever we were able to see so far as a rollout services, I have just wrapped them under one hand, chart to show how it looks into the rollout and just like any other kubernetes deployment, where we uh use these kind of variables to replace values. You can do the same for the uh argo rollouts.

A

Also, okay, the key difference here when you want to use a traffic routing mechanism. Is this section you would add here and you would just mention a stable service thanks to these annotations, what it will do it it will create a canvas or a version of service as well, and here, if you see these, two parameters can read by header, can be by value so here for demonstration purpose as well as in my experience, we try to pass a custom headers.

A

So imagine our qa team is doing testing so that time, while doing the testing they can use this custom headers and get connected to the new version of your service and do whatever test they want to perform on it. This would be more helpful around. I would say in case of blue green as well as in canary as well.

A

So I think someone was asking about the same this way. You can add the custom headers past path as well, so we will be doing a demo around nginx as a one to pass these headers values and we would be able to see we are directly hitting the can new version of the pod. Only new version of your application service only rest all part remains same. The only section that you would add in your rollout is this one.

A

So clear till this okay, let's do a demo of it and I think that way you would get some fair understanding around it as well. So I have added a demo for it here.

A

Okay, so we will create a one name space. It could be of any name I just uh when was doing demo. I thought of this and so added it here so before doing that, one thing you would need to do is so create a namespace has been directed here now go back to your id okay. Here you would be able to see your demo been added called as a argo traffic management demo inside it in your values. Here, values, file.

A

Try to search for host okay, so I have added here this uh particular hostname. You would need to do. Similarly, for you as well. uh Very easy way is just copy this, your entire url uh that you have right and from that you would be able to get your header uh hostname as well. If you see here from this value of your url, just keep this particular name and remove rest of the all and pass that value update that value here.

A

So in the repository that you have cloned, it would be showing this particular hostname, which is of mine. You can replace with your host name how just copy this url copy from the host name. Remove this https part just copy from this section onwards, till the cloud gurus section remove this next part as well.

A

Okay, so, and after doing that, just save that change.

A

So here you would be able to see inside templates all those objects that were earlier we were using. I have wrapped them as a hem chart.

A

Here, uh when you are using the traffic routing you need to mention, which is your candy service, which is your stable service, uh mention the uh your nginx ingredients and these headers, and if you see in terms of values, what I am passing here as a part of demo is, I am just passing two custom headers and its value, as as you would be able to see here, I am passing one header name can be and passing value as yet you can pass it anything here.

A

The advantage is that when we are so for imagining that we are, we want to connect to our new version of service. We will be doing curl request and we will be seeing that those requests are always landing on your ports of the new version of the service.

A

Okay, you would not be able to see anything in uh ui. I would say uh this kind of ui because it is not built in that way. It is just a plain nginx uh container that I am running here so once you change the hostname here. Just do help install. I have given this name just run this command.

A

Okay, it will show you that it has installed this hem chart. Okay, you would be able to see it inside the hem list. This has been deployed.

A

You can see. The rollout is has created two parts. It is creating two parts for it right now,.

D

A

Yeah, so our two parts are available now, okay, so now uh to upgrade. If what we will do is initially, you would be able to see the version tag here as a 1.19, we will update with 1.20 and we will do helm upgrade once we do helm upgrade it will in the back end, do the same stuff. It will create a new replica set using argo rollout, which would be running this particular version. Okay,.

B

A

Let's go back here.

A

Look for the image right so now, instead of just uncomment uh comment out this one and use a version like one point, twenty point: zero.

A

Save these changes.

A

This is our old version of the service been running.

A

Now, let's upgrade it, we have changed the image version. Just hang upgrade standard commands. You would run it okay, so it has deployed a new version of the service we can see in the back end. It is initiating a new revision. It is creating a new ports as well. Here.

D

Let me see here.

A

So you would be able to see this is a new part been created of new version and these parts are of old version. Okay, so what we will do? We will do cube city log to open logs of this particular code, and we will be sending curl request to this host with custom headers that we are passing.

A

So you can just copy this command.

A

A

Copy this name of newport.

A

So we will be constantly checking logs of this part.

A

Ignore this for a moment.

A

Okay and now we will be sending curl request to this particular pod.

A

The way we will do it, we will pass this command. Can we we will pass this custom, header and its values here and again, just replace your hostname here and do a constant curl request to this. Okay, let me copy this go back to idea.

A

Okay, let me copy my hostname.

A

Okay and now, whenever we will hit the curl request, you would be able to see the request getting routed to this new pod that we are creating. Okay,.

A

Can you see this request is has appeared here? If we will do again see the output is getting added here.

A

If I do again, you would be able to see again. The request is getting routed here. Only.

A

Since we are using candy version, even if you uh like, I would say, remove header sometimes request- will land on the old version sometime it land on the new.

B

A

But from internal testing perspective I would say uh you can let your qa team to do these kind of custom, header based routing or any other traffic routing mechanism sometime, and that way they would be always able to connect to your new version of the service.

A

Any doubt or anything as around it.

E

I'm getting a page not found I'm trying to figure out what is wrong with the saying for server for not found.

A

Yeah, it's uh okay. uh I think.

E

The interest hostname.

A

I hope you are not checking on ui circuit, the four zero four expected to get, because we are not using any index html page right, customize page where it will go and land. So if you see on my machine as well, it is showing the 4.4 that is expected output. All that we are trying to show or see here is whenever you route past the custom headers, it should land on your new ports alone.

A

Okay, so that 404 is expected because we do not have created any landing page for this, just make sure to check that your new parts are getting the version of the service. Okay now like. If I have to show here now, I just now, if you see I promoted the new deployment which created the new pod, which was 66 seconds ago now, let's say if I.

A

Access that new point.

A

It's this part which has been recently created right, so I'm accessing its law.

A

And whenever I will pass the headers right now, the request has landed on the another tab. Now it will land here. Can you see because now we have two parts running of our new.

A

One time it will go to that pod, one time it will come.

A

Is anyone able to get the overall concept around this.

A

You have doubts around this particular concept or around the technique.

B

A

Me delete this.

B

A

A

People are able to understand this concept right traffic routing one.

A

Shall we move.

F

A

So this is the new feature been added into the argo rollout called as a experiment. Okay, what? Basically this experiment does differently? Is it create a set of parts just like in uh other uh demo that we did where it create a new version of the uh service, or, let's say new version of the pods here it will create those spots and you can set a kind of expiry date after so those would be like a ephemeral one and uh the traffic light traffic would not be going to them.

A

uh Even if you use candy version, no live, actual traffic would be going to them. This experiment feature is basically launched so that you would be able to do some more other sort of destructive testing as well. When I say destructive testing. That means you would not be doing the actual application functionality test, but would be purposefully running the test to check what are the point of failures of your applications.

A

Okay, or in case let's say you want to not only do a b version kind of testing but abc where you can create multiple experiments and those experiments will create a self-isolated version of your application services. Apart from the one that is stably running and against those isolated uh version of your application, you would be running a set of different tests that you would want to perform.

A

Okay, so yeah in terms of uh spec, I would say uh you can just add a another section here compared to what we did prior uh this section, uh you would add as an experiment and in this experiment again, you can run your analysis as well. Okay, so the main advantage is, it will not have any cascaded. If effect on your stable running service, those service will be running as it is. You would want to do some experimentation with your production running services. That's where this would be helpful.

A

You would create a version of the service. No one of your user would be able to understand that that you are, you have created a new service and you are running a different kind of test around.

A

Okay, uh I was not able to create a complete demo for it, uh but I have added an example for the same. If you see in the examples, I have added an example, you can just run those examples directly to get a feel of it.

A

Okay, and uh so yes, coming back to a question, let's say you are currently running all the kubernetes deployments and now from kubernetes deployment you want to uh go for the rollout option. What changes, ultimately, you would need to do. Is you need to change the kind from deployment to rollout?

A

You will change the strategy and you would need to create an additional service to route the traffic. Okay, so in terms of uh here, you would be able to see it was having deployment. Now, the same object is converted to the rule out in terms of strategy.

A

uh This is like a default kubernetes strategy that you can replace with this can and in terms of services. If you have one service, you will create another set of service, so I think this is what tulsi you are looking right if you want to do a demo to your client to show if you should go for rollout.

A

Yes, so these are the only changes so in terms of changes as well, uh you don't have to do something different, just that you need to as a step. One install the controller in that cluster and that controller is then capable of observing, keep an eye on the any object of kind rollout and do the actions for you.

C

And here's one quick thing so, while running, can you just go back to the previous slide yeah? So uh my observation right, I still still once I re-run that particular exercise, whatever you're given might be understand better. So in this step you have something called step weight, 20 and you are given a pause correct, so it stops over here. Only after I try to run the argo promote command. Then it goes to the next steps right right.

C

Okay, so even after uh step 40 I can give one more pause and I can hold on just to be playing.

A

Yes, yes, yes, it's up to the way you customize it.

F

Just for demonstration.

A

Purpose, I did it as a one pause. You can add as many as for as many as steps, not only you are limited to do it as a twenty percent.

C

A

On the number of replicas- and yes, you can do like you, roll out twenty percent traffic do pause. Then roll out forty percent do pause. Do some manual test then promote.

C

A

C

So whenever I have a blank without duration, only then I have to do promotion else afterwards. It will do by the duration, whatever it says,.

A

Yeah I mean whatever duration you would set. It would do after that right.

C

Only promotion, if I have a blank one promotion is something which we need to do it correct. Yes,.

A

Yes, because, basically here we are saying it or saying correctly, I manually promote you further.

C

Got it got it just sure? No.

A

I mean in regular scenario like in uh the use case, in my use case. What I did I was I there was nowhere. These walls were added. It was the part that we added and that analysis was basically doing this uh work of making sure that my new service is running correctly or not those projects so.

C

Both analysis with analysis, yeah that will change yes right.

A

To completely automated or to if you still in initial, I would say: uh do a manual test. You can add.

C

Okay, we'll go with this and then next step was to add analysis exactly.

A

Step by step incremental approach, only.

C

A

uh You have already seen this hand chart that I just now showed right, nothing different. You would be just templatizing it just like you do it for any other deployment object.

A

Okay and uh yes, these were my lessons uh from the work that we did is uh make sure that it shouldn't be a very short time to decide uh whether it should promote ahead or not. uh Based on what I have uh seen is companies keep it running for a long duration. When it's a long duration, like maybe a day, you can keep your older version as well running. Okay, uh second and most important thing is carefully choose metrics uh when you are doing auto promotion or auto analysis.

A

uh That metric should be the right one to do the decision, making uh consider that I would say right now, whatever you are doing as a manual checks, and you are confident that this is how I should I do a check. Add that as a part of your analysis, so that that automated process should do it uh third thing, I would say uh yes in one go, you would never get it correctly.

A

It would be always an iterative process right as the overall devops process in that way only uh start with small and then go progressively uh and try to improve something more in that. That's how I would suggest someone to use the can any kind of deployment strategy that you would use it.

A

Needless to say, don't directly, try it in production, start doing these things in staging and then go in production and the last but not least, uh I would say the don't try to compare your calorie result directly with exact production, but try to compare it with the baseline results, because sometimes what happens in case of like java kind of applications, it keep accumulating those results right. So better thing is try to compare it with the average results that you have. That's all I had from my end.

A

uh Anyone has any questions for me. Please go ahead.

E

Thanks for the wonderful session.

A

Thanks: listen. This gives some understanding around how we can move now from cd to progressive one and I hope, you'll be able to bring some or change in your existing works. Work scenarios.

D

Okay, all good guys see you. Thank you once again uh for putting all the efforts there uh and uh looking forward for more.

B

Collaboration like this yeah.

D

Everyone see you thank you.