Knative Demos, 15 May 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Automating service delivery with bindings - Knative demo

Description

By Evan Anderson, software engineer at VMWare.

A

So this is just a little application that I hacked together that show that does a lookup in the dick in the dictionary using dictionary, api's.

A

A

So if you want to, you can look at it, but I, basically, I've got it running locally. I had.

B

A

Running locally.

A

So it's just a little doe application and you can load it and define a word so banana and it will go to there's a couple different sources it'll pull from, but it prefers the oxford english dictionary. So here's some definitions for banana, which is all well and good, and you know if I decide. You know. Oh maybe I should say style.

A

Yeah, okay, so I, don't remember how to do inline Styles. This is not about how I know or don't know how to do HTML, because clearly I don't. But so you know my local development loop is pretty quick. You know just kill it and restart it, but then, if I want to go and actually run this on K native, you know I don't know. Can people read that? Does it need to be a little bigger.

C

A

Up a tick oops, that's too many ticks. You know, I have this wonderful bit of gobbledygook. That is a two-phase dockerfile. If I want to actually build this stuff and then.

A

Let's see and then I can do a docker build.

A

And you know you've all seen this and it takes some time. You know I happen to be using go, but you know: there's lots of different languages and almost all of our stuff, unless you're using Co kind of looks like this, and what's even better, is now. If I want to push this I realize. Oh wait. I forgot to tag it so.

A

Okay and then I have to dock or push it.

A

And it takes some more time and basically you know this iteration cycle, you know kind of works but kind of annoying and then I have to go into my llamo.

A

You found the money you are oh I will I will be finding a bunny later, but you know now: if I want to go in and actually you know update something I then have to run. You know: KN service update, I.

A

Image is whatever thing I got here.

A

And then it turns and does its thing and.

A

Eventually, it will show up here, I've been having a few issues with SSL, so in some case, as I might fall back to.

A

And because I didn't remember how to do this, there's my background. Color I'm gonna start using blink tags instead, because those are really obvious.

A

But but you know, that's that's sort of your typical workflow, but wouldn't it be great if your workflow could look a little bit more server-side, driven and so I'm going to show you a couple of tools that were written by a variety of folks from VMware when they were under the pivotal umbrella. So some of these come from project Grif and some of these come from kpac.

A

So first let me show you k pack. So, let's see.

A

A

This is a image which is a kpac construct that lets you do cloud native builds, so I can make this a little bigger for you all and.

A

When I create this image Hamel, it will see it done. It's done. I'm also gonna show off octant for those of you who haven't seen it it's kind of a nice UI on these things, so I just created an image and right now it's not ready yet, but it will create a build each time that things change and you can see that the build is currently progressing.

A

We'll leave that back there doing its thing, we'll look at what's actually in an image, so an image defines a source I've just pointed at github and master branch and a builder which gives you a set of build packs. This will actually build a lot of different languages. This is this default. Kpac, 1 and.

A

Cluster, so here is the cluster builder and you can see oh.

A

That's the definition of cluster builder I always forget when I'm looking in octant, you have to load the object and then the ammo, but you can see, there's build packs for node and go and dotnet and a lot of other things, and so this will using cloud native, build packs, auto detect which language, you're, building and running and then go and build a container.

A

So probably, if we come back here- and we look at our images- we'll see that there's a latest image now, which is the last it the most recent image that's been built, so that's great I can go and we'll note that this starts with 480 I can go and make a change over here.

A

Let's see I said I was going to do blink. So, let's see.

A

A

A

Link more yes,.

A

Yes, do all the saves push this up and in a minute or so we'll see another build kick off. Another way you can watch that is with Cube control at build. W is just a watch on builds, so you can see right now. We have the one build that was created at the beginning of time, but we'll pretty soon and see another build kick off as the image notices that there's been a commit, and these are various stages of the build making progress, and so it keeps updating the status.

A

The only piece that really matters here is whether it succeeds or not, and what the image is, but each time it gets updated along the way the watch will report new information, so this will go for a moment or two and then at the end we will have a new image and let's look a little bit at this image again.

A

This is the one we want. So if we look at build define here, the llamó status has this field latest image and if you all are familiar with duck typing, you may know where this is headed. Anything that has a latest image you can see, our build is finished and the latest image has been updated to build to our friends in project. Rif have created a created, duck type controller.

A

You can lostalready that is called an image binding and it lets you connect an object that has a status with the latest image to a resource that should be updated.

A

So right now, I can push, builds to get hub and within a minute or so I get a docker image built not on my local machine but off in the cloud somewhere. If I also apply this binding llamo.

A

Now this every time the image is updated, the user container inside the service should also get updated, and this works not just for services, but it also works for deployments. So.

A

We'll create that as well and then we can check.

A

And they're both ready to go so now I should be able to.

A

Forget how to do anything else annoying here, Oh what you can bold, bold.

A

And if I'm super smart I would actually check and see that stuff built locally, it does.

A

Yeah, do the stuff I still need to figure out the one click you know commit and push thing in Visual, Studio code I'm sure there's a way to do it, but now over here we can look and see our builds are still kicking off and running, and so, let's see.

A

Let's watch those.

A

So yeah this is some of this is early prototype code. Some of this is semi production code, none of it who's hooked up with eventing, so there's no there's no registration of a web hook or anything like that to kick the build process. It's just doing a poll on github, but github can deal with some polling so and here we go, it's kicked off and, let's see, while we're waiting. Let's take a look at that service.

A

So if we look at this.

C

D

D

A

Image specified here and when this is done, the controller should move things over. The other fun thing is one build, since we have two.

A

Since we have two image bindings, we can apply the same thing that we built to both of them, which is less interesting. If it's a knit web application like this, but if it's a sidecar you could potentially push a whole bunch of different side cars. At the same time, let's see we just built something that is in E. Oh look, it's already updated on the deployment. Let's go look at this Serkis and, of course, I'm doing a live demo, so the service didn't get updated here.

A

Well, the deployment got updated I would I would have to dig around to figure out why the service didn't unless somebody on the call has seen this before I.

A

D

A

To just get updated.

A

Well, the other thing, of course, is.

A

So you can see all the pods that I've got running on the cluster, and this binding system is where that binding happens. So.

A

If you folks, hadn't seen this already whoa, they run that command for you again, you can specify a deployment rather than needing to specify a specific pod, which is pretty handy when you've got. You know really long deployment names like in Kenneth.

D

A

This so that's a webhook reconcile, succeeded.

B

The initial binding, probably should have replaced the new tag as well. Oh.

A

B

A

B

Never kicked in for that.

A

Admission review: oh.

A

Well, that's not a useful.

A

Patch is a set of bytes I'm. Afraid I can't read that on the fly.

A

E

Is right below the the hex? Oh.

A

This I'm, not this is the same thing as the patch I.

B

A

B

Was recently fixed, okay,.

A

Yeah I have a slightly older version of the code, so maybe that is a sign that I should fix that. But let me show you where these things actually live in case you're, curious.

A

So this is the define app it's only interesting in so much as if you want to reproduce what I was doing. So this is pivotal capac. This is an integration with cloud native, build packs that does builds on the server side, and that provides you that image resource and.

A

Image bindings come from this project RIF bindings, and they have some nice documentation down here. That shows you what to do with them and all the different values you need to put in, but yeah the exciting thing about image binding is that it should work not just with Capek but with anything else that exposes a spec test image.

A

So, theoretically, you could query a docker registry and find you know the last version of a given tag if you wanted and have that auto bump, using image binding to resolve it down to a specific reference or you could say hey, this is my external store of past end-to-end tests or you know, past QA runs once something passes: QA then it can actually get rolled out to the next stage in production.

A

So I thought that was kind of neat, not my work but happy to show it off.

C

Awesome, thank you so much for the demo. Now we have allotted some time to discuss this. Ideally we would be doing this on individual rooms, but we were not able to enable that for this meeting. So we can open the floor to discuss now or you know, share any comments or ideas. This may have inspiring you or questions.

C

There was sort of activity in the chapter in the demo. So if anybody would like to speak up, there will be the time. Oh.

D

It is I Alec, just a quick question about the binding. What are you changing on the image in the deployment or something else so.

A

That's changing just image in particular, it's changing the image associated with the container name, the inside Spectre template.

D

You mentioned some other tools that you were planning to use. What could be changed to make it I, don't know faster to pick up the changes or o.

A

If you connected the github event source and use that to poke the image that would cause the reconciler to kick off and decide to, you know, go and check and see if there's anything there and oh look, there is which would let you reduce the polling interval. It would look like webhook to something running on the cluster. You know to the source, which would poke the build controller and then would kick off a build that way and.

D

Entirely you could bypass github right. You could just send somehow what you modified either from your doctor, locker or just change and build it in your cluster right, yeah and.

A

Kpac actually supports several different kinds of sources: I.

A

Used to github one because it's easy and doesn't require a lot of setup, but if people want to see I can show what the other sources that KPAC can use are.

A

A

They have these source types and so a source can be a get or a blob or you can actually pull from a registry and they all you can take a sub path. So you can say not this whole git repo, but like some sub path of the repo or you know not this whole tar ball, but like this sub path build from there, so you can see, get has URL and revision blog is just a URL and a registry.

A

You can pull down an image, so you can package your your source code into an image, put it in your doctor registry and then pull it from there and you know. Obviously you need to get into the capex source to extend this further, but it would be doable if you move to something like Tecton. The fetch the source is actually a separate piece, and so you that's probably more easily pluggable. The listen for changes is a little bit harder with Tecton, which is why I chose k pack right now.

D

C

I think has a raised fund.

E

Yeah, hey Evan I was wondering if you could compare and contrast this with Tecton and you kind of started.

A

So k pack is is very focused on basically building a single docker image from a single repo. Tecton is basically a full-featured workflow system. So, if you want to, you know, use multiple repos or you want to deliver multiple artifacts to different places, or you want to do something more than just build your software. You know if you have a more involved testing process.

A

For example, you'll find that k pack will run out of steam fairly quickly, but it works really well for the simple case of I have some code and I just want to turn it into a docker image, whereas Tecton, because it's got so many more capabilities. You're greeted with you know, ok figure out what you want to do and how to describe that in a bunch of yeah Mille documents, whereas you know there were like three or four things: I had to like intentionally right for the kpac configuration it's pretty minimal.

A

If you look, if you look at their tutorial, basically, you create a secret, you create a service account, and then you create your image and you're done and.

E

It sounds like they might be able to join up or something.

A

They both use cloud there. They both could use cloud native built packs. That's a common tooling that kpac is leveraging.

A

Tecton can plug into something that can build cloud native, build packs.

A

Tech town itself is decoupled from any particular build process. So if you really want to use Tecton to deliver your mobile apps, it's a good tool for that to you, I guess, cool.

E

A

Any other questions comments. Oh.

A

I forgot to show one other cool thing that octant will do so. Maybe I will show that as well. Just so that you know it's look at all the cool tools that we already have because we're part of the curators ecosystem time.

A

So one neat thing about octant is that you can actually go in and look at a pod and if it knows what's going on, you can start a port forward from the UI which is kind of slick, and you can see all the events and stuff that we're related to it.

A

So I've been trying to train myself to actually use the pretty you eyes, rather than just do everything from the command line and then curse when it doesn't work well, so you can see now blink is no longer actually a valid tag, so so much for my making things really ugly. Let the blood tags are in there.

A

Don't need to port forward anymore. The other thing is that octant, if you look at it, is running off of localhost, so it uses your local cube, config and basically runs as you, rather than being something that runs in the cluster, and you have to figure out how to access control. It.