IPFS IPFS þing 2022, 10 Aug 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IPFS Operator - IPFS Cluster in Kubernetes - @cory - IPFS Implementations

Description

IPFS Operator - presented by @cory at IPFS 2022 - IPFS Implementations - https://2022.ipfs-thing.io

A

I should give you a warning uh this. This uh talk will discuss uh ipfs running in kubernetes, so if that causes you to wins in pain, uh I understand uh okay, uh this is a talk about uh ipfs operator.

A

Ipfs operator is a kubernetes operator uh which I will explain, uh but in brief, it's a way to manage uh ipfs uh and put in also ipfs cluster uh in, like a smart, easy to use way uh on kubernetes, so uh yeah we're actually trying to target kubernetes uh as the as the in installation, and you know like leverage some of its benefits.

A

Okay, this is the agenda for the talk, uh we're gonna, we're gonna, I'm gonna explain uh who is it? uh Why would you do this uh and also uh I'll give a little explanation of how it works uh and I'll uh go over the run run. uh Excuse me, the road map and uh some of the pain points uh that still exist. uh This is the team this is who is working on it. Currently, uh myself corey uh hector, who presented earlier on ipfs cluster itself.

A

uh This is being uh developed uh in conjunction with uh red hat, uh so we have two uh people from red hat emerging technologies who have been uh really helping us uh to develop this thing. uh Okay, uh before we get started, I want to. I want to give a little bit of context. I just put 10 seconds up there. uh What is kubernetes, I I feel like I just need a little bit of explanation uh in brief. It's a it's a cluster operating system.

A

uh If you're not familiar with it, uh it's a way to basically get your applications running and there uh there's uh like an api that will let you manage things in an abstract way, so load balancers, managing your storage and managing workloads. Things like that are sort of like wrapped up into this common api and another thing. Another reason why this is a good target that makes it so popular is that it is uh it's ubiquitous every uh cloud provider that you can think of.

A

They probably have a kubernetes implementation uh and you can take your kubernetes applications and run and run them there. um Okay market share. I wanted to add this here, just because I want to give an explanation for why kubernetes is a great target, like 3.9 million developers, that uh that is a huge amount. uh If you're thinking like who uh you know who's running uh kubernetes, the answer is: it's everybody.

A

uh uh Yeah 31 of uh back-end developers globally are using kubernetes, uh but, uh and we want to make this a a target where they can run ipfs. uh So that brings us to really the quintessential question.

A

uh If you are one of these people, which is everybody who is running uh kubernetes, uh what do you do when you read about something cool uh like ipfs? How can you explore ipfs?

A

I would bet that there are probably in this room. uh You know. Half of you could probably come up with a a decent implementation, but the thing is that we are very familiar with ipfs and my guess is that uh if we that uh the implementation that we would come up with, we would struggle quite a bit. So anybody that's just a standard.

A

You know bog standard web 2 developer, that's running their own, like application, if they want to if they wanted to run ipfs they're going to really have a bad time, so um I think we can. I think we can make that a little bit better, and so that's the reason why uh kubernetes is a good target and uh developing an ipfs is a really uh good idea. Okay, so running on ipfs, I've got a little gif here, but I just wanted to kind of show what it looks like live. So the way this works.

A

uh Let me let me open collab. Let's say you want to run a collab cluster. uh There are there's excellent documentation on the ipfs cluster website. It involves setting up. You know new peer ids, generating you know, random data and using that to generate keys. uh You know like ssh into a bunch of boxes and like causing them to join each other. uh We can make this easier on kubernetes.

A

uh Basically just I have this uh really simple, like template here, that you can just simply follow so the way that you would create a ipfs cluster in kubernetes world uh coupe ctl apply. Let me do the other one cluster 100.

A

And there you go, it's created, it's not actually created, but it's just doing work in the background I'm going to come back to this later. This is going to be like a big surprise where we're going to be like.

A

Oh there's a cl, while we were talking a kubernetes uh run, uh collab or cluster was uh formed, uh uh yeah, so um running on kubernetes, the the takeaway is that it's easy, like you just saw me, do it in one command, it's easier to set it up, one if you're somebody that's already familiar with kubernetes, then setting up the cluster using an operator like this uh is easier even than setting it up.

A

You know like manually like following the instructions uh yeah so yeah, particularly for people who uh already are familiar with kubernetes the learning curve on this. Is you know it's one command? It's it's something they're already familiar with. um Okay, uh the other, the other point of context. uh What is an operator uh if you're familiar with kubernetes, but you haven't actually run it yourself. You might not be familiar with what an operator actually is. uh An operator in brief is a method for extending the kubernetes api to add additional functionality.

A

So you can it's a piece of code that you can run in the cluster. uh It has access to like create things, but you can add your custom logic to it. So what I was talking about, like creating uh cryptographic, keys and like creating cluster secret and all that stuff, all that can be done in the operator uh you don't have to like even know that it's happening or know that it exists uh and uh that custom resource. That's that file that we just defined.

A

You can basically define like a an api object, upload it uh and it'll be handled by our custom code uh yeah. So this is. This is going over what we just went over uh yeah uh I've got I've got this ipfs object here. I'm basically saying that I want to create a collab cluster in this case, and I want to follow these things uh if you're, if you're paying attention you'll notice that this is all of the collab clusters uh that are uh on the like ipfs cloud cluster uh collab website.

A

uh Yeah, uh this is what the this is, what the operator is doing: uh yeah, okay, so now we're at the now with the section we know what an uh what a operator is uh now we're gonna talk about this particular operator, so um these are components that get created uh this. uh This is this is basically a layout of what the pod looks like in kubernetes. A pod is basically like a unit of work. It's essentially like a bunch of pieces of software that run in a way that, like they have the same networking.

A

So typically when you set up ipfs cluster as hector described earlier, you have an ipfs, node or excuse me, a kubo note, uh and you have uh ipfs cluster that runs next to it and talks to the local node. Here we did the same thing: we're we're not doing anything uh fancy here, we're just you know like making kubernetes wrappers around it. Basically, so we have an ipfs node that ipfs node has some block storage attached to it and we have a cluster clip service that is able to talk to it on you know, 127.0.0.1.

A

Anybody that's running ipfs cluster. This will look very really normal.

A

This is, this is something that was actually. It was a pain point we fixed it now. um This is this is all this is something that I think anybody that was not familiar with lib, p2p and ipfs would uh struggle with this would be something that is not very obvious.

A

Basically, once you create this pod, you've got some sort of local ip address. It's not really very obvious how you can achieve like external dialability. uh Do you yeah? I put some. I put some examples of things that we like explored. um Like yeah. Do you use node ports? uh You know not obvious at all.

A

uh We. What we ended up doing is uh there are some. uh There are some relays that are out already out in public, and we also uh allow you to run your own relay. If you do run your own relay it'll be an additional pod that appears in the cluster um and the uh logic is set up so that you know like once once you.

A

If you are running your own relay, then all the ipfs nodes will be like configured correctly to use it and the result is that it works, and you didn't have to really think about it or know anything so hole punching works.

A

You get content, that's addressed to the network correctly using a network that is uh that is dialable and also you're, not doing something silly like spinning up a bunch of load balancers and using up all your ip addresses. So that's really good uh yeah. Let the operator operate. uh All the stuff gets set up for you. You didn't have to think about anything.

A

Okay, yeah. uh The other thing that the the operator handles is uh changes to the cluster after it's been created. So uh let's say: let me just go over here. Cube c cube, ctl.

A

Get get a list of my ipfs uh clusters. uh Let's look, let's edit this collab cluster.

A

uh Well, let's set ipfs collab, okay.

A

This would look a lot better if uh it was a little bit smaller, maybe yeah. Let's say we're running out of storage. uh You know we don't need four, we don't. We only have four replicas right now, but let's make it six. uh Oh of course,.

A

Of course, there's gonna be a bug. Let me let me edit the other one.

A

Yeah, so this one this one I was doing a hundred, but you know probably this will probably not be as obvious but yeah I'll, make it like 110 or something like that, uh but it'll it'll allow you to like adjust parameters on the fly uh and it'll just respond and create the thing that you wanted. uh That's the that's the overall point that I'm trying to make, but what this allows you to do is you can control the cluster through an api programmatically.

A

uh Yeah demo uh yeah, um let's see uh fingers crossed. Indeed, uh I wanted to do a quick demonstration. I have this file here, which is uh hello, uh cluster or ipfs cluster ctl.

A

If we add this file, okay, we get a cid back. It's pinning.

A

A

There we go uh yeah, um okay, so now we're onto the road map section. I've only got like a couple minutes here, uh there's a uh there's, a great diagram here for basically uh how to judge any uh kubernetes operator.

A

uh This is, uh I think, comes from uh red hat's documentation, uh we're we're not all the way up to uh level five yet so uh we're. This is still very much in active development. uh It's uh in fact not even really released yet, but uh we are uh working our way through to make it so that it's more capable uh more more able to like handle the entire life cycle of the service.

A

uh So the, although all the lists on this uh uh capability map or on the roadmap uh in the in the next couple weeks, uh we are going to uh push at least like a beta or something onto operatorhub.io that'll invite people to use it.

A

If you do want to try it, however, there is code available on the github repo I should have listed here, but I'll go back to the front of the slide. At the end, um the the code is available. There are helm, charts that you can use to install it uh and uh or you can run through the instructions and just do like make install and it'll also install into your cluster uh and yeah expect here in the next couple of weeks.

A

uh This will this will be on operator hub.io, which is like a kind of like an app repository type of thing, but for kubernetes operators uh and uh yeah. If you have any questions, uh I'm corey on slack uh there's my email address also for anybody here uh feel free to contact me or come come, walk up and say hi all right. Thanks.