Status SWARM Orange Summit, 1 Aug 2018

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Running Swarm on RedHat OpenShift

Description

In this presenation from Day 3 of the #SwarmOrangeSummit, Nikola Jokić from DA PowerPlay presented a run through of the technicalities of deploying Ethereum networks on Red Hat OpenShift solutions. The aim of this talk is to show possible steps to achieve production-ready Web2 software deployment.

A

Hi hi hello, I forgot to put my name on the slide, but my name is Nicole Aoki I'm CTO of digital assets, power-play. We are based in Croatia, but we are really present in the Slovenian seen at least in cooperation wise. We are focused on algorithmic trading, but today we're not going to talk about that. We are going to talk a little bit about running swarm on RedHat OpenShift.

A

Has anybody heard about open shift perfect? Nobody? Almost nobody. Has anybody heard about the docker Dockers? ah Yes, so you have Dockers at the bottom layer. Then, if you have multiple Dockers and you want to orchestrate them, then you have kubernetes. If you want to deploy kubernetes in a more serious environment, then you will probably use open shift, which is basically let's say you wrap around kubernetes stuff. So the talk today is how to run basically any blocking software in sort of an enterprise ready production deployable environment.

A

Well, if your code is production ready, then this thing this thing should work.

A

So how do? How do you do production ready deployments of ev3 software if you rooms for whisper I think these are the most three important software's piece of software that you can find today. I'm, not sure art is the software production ready, but I think I, think Red Hat overshift is so ready. Autopia shift is an open source company. They started in 1993 with a fork of Linux, they they're fully open source and they are working.

A

Some billions I think their market cap they have around 11,000 employees and they're focused on like an open source but enterprise-ready solutions. You will mostly find them in banks or some bigger, bigger institution which have a bit more serious deployments deployments procedures. We we deploy our platform on the on the opposite opposite platform. Some parts and parts are bare metal. So what's the idea idea is when you develop some piece of software? That's one part of the problem. Your software now works nicely. It passes tests, but you must somehow put it in production.

A

What you can do is maybe SS kind to a machine then do a git pull build somehow then run it. But it's not. The way that you should do it, maybe you have some rich Jenkins procedure that will deploy this thing. But then, then there is the question how the production is on one state I do some development did I'm somehow need to switch the prove the old production out and then put my new dev branch or something in in production. These kind of questions are well anybody's problem that deploys some software into the production.

A

Openshift really really helps here. Since we started to use it, we kind of sped up, maybe five times in terms of in terms of our efficiency, so why? What would be nice to hear from your own deployments after it would be nice to easy to scale, so you can maybe click a plus button and get ten more etherium nodes that you need. Maybe because the the attack on your on your back end is a bit higher.

A

It would be nice to easily test different topologies, maybe to have an environment which you can deploy some kind of network topology in your private network, then maybe test out what happens when the network splits or maybe what happens when the Wendy Wendy Wendy some hard work happens. If you're you doing it maybe manual it's kind of tedious process. We were previously on digital ocean or AWS, and it was like you go into these ten machines.

A

Then you set up this kind of situation and then you see what happens, but it's really hard to reproduce because you're not just copying the source code or the build off or project, but you need to replicate the environment.

A

Ideally, your deployment process would be infrastructure independent. So it means that any kind of hardware you have you can you can deploy on it. We have on-premise hardware and built from from the hardware layer, so we styled OpenShift on our hardware and it really it's a non-trivial installation, but it basically it's it's. It's it's a way to go, I think so. What's the part of the technology stack its record Enterprise Linux?

A

If you are for a more of an open source solution, you should go with CentOS, which is basically an open source fork off of Red Hat Enterprise Linux, docker, kubernetes and OpenShift. So you have some text here that you can read: what's what's this field all about, so what we did? We, we kind of were surprised that you cannot own opposite. You have ready templates to deploy almost any kind of software. If you need the Possible's database, you can like yes, yes, yes and you get your own persistent process, Postgres database.

A

If you want some kind of a Kafka cluster, if anybody uses patchy Kafka there is a stream see project Red, Hat recent. We just started to work with them. It's basically, yes, yes, yes and you get tens of Kafka nodes connected and working together. So it's really really easy way to get more complex deployments of your software up and running and and template that so you can reuse it and in in some other projects or other environments. So what we did is an experiment. First, we started with: let's, let's get one parity node.

A

We use parity client for development, so, let's, let's have it just deployed. This template is open source, so you can see it and then we kind of started to experiment with more bit more advanced, openshift possibilities. What we did is we deployed one validator node in a proof of authority, Cohen set up three load-balanced secondary nodes, which are not mine and one block Explorer that basically requests get load balanced across these secondary nodes, and this validator know this is actually actually mining the box or signing them.

A

Whatever you like to say, what's annouce nice about this, we'll go check out? How does description of this deployment looks like, but basically you kind of describe I have some service that service has route outside behind the service. You will have three parts. Each part has limitations of X megabytes of RAM or this much of processing power. This all editor lock has a bit more. It has also persistent storage, so you reserve some some space on the hard drive and then you can easily easily easily redeploy it. What we can do is basically.

A

What we can do is maybe look look how it does it. Let's see, let's see, how does it look in in open shift when you try to deploy it? So, let's, let's sit here and hands on this- is open shift user interface. It's an open shift online solution, so we can open up an account like an Amazon Web service or something like that and pay some money for for the resources, but basically you can install it locally. I recommend mini shift. It's like a small open shift that you can run locally, and here you can.

A

You can do a bunch of things. Ideally, you can start any of these things behind there every each of these things as a docker image, with some description around it. That describes how to run this. This docker image this is already had supported. What you pay with record is basically support there. You can have somebody on the on the line. Please help something is correctly in my system and that's what you pay for. The code is basically open source. So you can, you can deploy a bunch of stuff, but you can.

A

You can also import your own templates. This is the one that we that we developed and open source. It's a piratical searching starter that we have called it. You can go next, it's a bit description! What's what's this all about in basically now you have some configuration phase in which you can now give the name for this console. You can give it network ID, you can set block reward. You can define set step duration. If anybody looked into Genesis JSON of the Cohen Network.

A

These are basically the parameters for the Genesis Genesis block, so we can here define who's, even the validator. What's the validator identity, how many gigabytes of memory should the validator have for a persistence?

A

What's the RAM limitations for for the part, because you can also set some limits if your project goes crazy or you have some kind of memory leak, so it doesn't eat up the resources of the home project but stays contained in his own limitations. So it's it's kind of kind of nice. You can even set your private key, so it immediately starts to starts to mine box. When you click create, basically wait a bit and then what happens is we can go to the deployment?

A

Now? Let's, let's go to this view. So what what happened? We we have a template.

A

We have a template. Basically, we have a block Explorer, which is forked I. Think if there light or something like that, the one that can show there are links so I can I can send you which, which walk Explorer. Is that but it's an open source solution we just packed it to to a docker non-root docker image based on where ahead Enterprise Linux, what's also deploying here is the secondary, knows those load-balanced one from the deep, deep power-play parity, one, nine five 10 plus image.

A

We did our own, because almost anybody today that writes docker images does so in route way. So at the end, the user that starts the the container is root user and, although you're in a container, this is not the right way to do it. Even platforms like cocoa shape, don't complain. So when you write your own doctor files, please take care that we at the end you kind of switch to a non root user and then fix some access access for your users. The project can work.

A

We are kind of talking with parity to fix their own CentOS image, but for now this one is enough. What we deployed is free pods, so it's free copies of the same parity parity parity secondary node, and here is the the validator node, the only one which actually signs the science D blocks. What we can also see is basically services, so you have different services.

A

That say you know: I have some service that I want to expose to to to the outside world, or maybe between those parts and I say this service exposes on port 3000, some kind of selector and then inside I defined, on which parts the service will load balance and what are the rules of load balancing? You can define some weights, so you spend like 90 percent of traffic to one pot. You can do even crazy. Things like have 10 different rests. Api is all load balanced, although they are not the same project and not exposing.

A

You will not get good results, but you can actually actually do kind of crazy, crazy things. There are also routes. So when you define a service, you can define the route and say: ok, create me route that will go to some target port. To that service service exposes, we can maybe try to go and see.

A

Is the box or up and running Internet is a bit slow, so we'll give it some time until this loads we can go check a bit about, let's say: config Maps config Mac's are basically your configurations for the for the Dockers that that you deploy. So we can maybe look at the Genesis. That was that was that was defined. So some step duration was defined during the template creation. Some block reward was defined using the template creation, the dependent whatever you decide to expose.

A

Basically, you can set up in the initial setup if I edit this file some change some parameters and go to and quickly deploy. It will basically redeploy the thing you can pick between two two types of redeployment, so I could recreate so it crashes the new one who crashes the old one and then brings the new one up or you can do it like rolling deployment in which you kind of keep the current one alive and then destroy destroy the other one okay.

A

So we have the block Explorer alive and running basically no blocks who are still mind. I think that the nodes are still looking for for each other, although what we did is we kind of told them that there are some reserved peers, so there is one period that will connect all all other nodes together.

A

So what's the idea here, you can see that you have like a really nice control over the over the already process of deployment you can, if you like, you can basically do something like this like go to a secondary, node click plus, and it basically scales to four. You can define some rules if the CPU usage goes up, 10% and scale scale to five things and stuff like that. So this is the currently the only way to easily start some aetherium software on openshift.

A

We hope that the reticle guys will wake up and then kind of prepare more software. So this is about that I scratched service surface a bit with the let's see. Did it start to mine? Still, nothing? Ok, let's! Let's not wait, let's go back. So how does the template? Look? Let's? Let's maybe just check this thing out. So, let's, let's first look at the basis. Basically this is the docker file that that that that deploys the parity node. What we basically is, we get their own build of RPM.

A

We should basically build this RPM inside the docker image, but this is also okay and then, at the end, what we do is expose some ports and say user 1001. So it's not the user, user, 1 and entry point is basically parity. Any other parameters that you want to move forward to these doctor image. You will be basically applied to this entry point, although we can even change the entry point and stuff like that. So this is a non route doctor image of parity.

A

If you like to use it, you can use this one, because there are no other non route images out there. Yes, I can say I'm coming right now to the swarm part. Thank you for the push.

A

Let's, let's, let's move, why swarm an open shift? Okay, at the end of the day, it's all about the data. We need to store a bunch of data. What we do is we collect the data for our algorithms. We collect thousands of market pairs and it's a loft of data that we need to store somewhere.

A

Why do we want swarm? Because we think there is a problem who creates cryptocurrencies here.

A

Okay, do you have some kind of algorithms written up or you're like manually, clicking manually, clicking? Okay when you're in the algo trading environment? You basically see that bunch of guys are writing their own algorithms and collecting their own data and that's basically a problem because I collect your arc. I collect market data, you collect market data and this guy collects market data into basically the same data, and we have a problem because then the market is not efficient, the data is not free and it's free on the exchange level, which is really disruptive.

A

So we somehow must finish to this endgame open market data. We must all collect it together because it is public and today, if you, if you want to back test some strategy of Bitcoin from 2009, you have two choices. You can either buy the data which will cost you thousands of euros and you cannot verify it or you can maybe ask your friend or you can maybe like send an email to an exchange. But basically you are.

A

Although this data was free and each minute that the day goes, we have less and less trades collected and we have a bigger problem to the future. I believe swarm is the answer for the storage layer of this problem. We need to somehow publicly store this data and what else do we need? We need to somehow publicly they duplicated, because if five of us see is the same trade, ideally we would write only once a 2d 2d swarm, swarm layer.

A

Now an example of running swarm on OpenShift not going to happen. There is no swarm on OpenShift currently or this is Croatian rapper. He says memo a no-go. It won't happen. Why? Why it's difficult to find source code? Who can tell me where the source code of swarm is you've dries you're the maintainer yeah, it's probably on some kind of branch on gif I think? Is it I think so? Yes, it's difficult to detect, which branches should I pull to get a stable built.

A

It is difficult to filter issues because I really don't know what the you guys think an issue is even if I think an issue is I. Really don't know where to put it it's kind of up it's difficult to detect, which we are to pull to get some features. I heard some guys, like you guys, are doing great job. You know it's a serious big piece of software that she was developing and you're developing with another piece of software that doesn't have any relationship to this.

A

It's like if helium swarm, two pillars, two totally different scopes in the same you get repository, come on, don't me and then the end it's difficult to collaborate. Yes, you know why? Because I cannot I cannot find the place where to talk with you guys. You have getter channels, you have everything, but you think you don't have the core. You don't have the get repository.

A

Do you have am I wrong.

A

A

Is there a repo coming like an organization swarm, /, swarm.

A

I cannot really find the github repo containing swarm code. I can find the etherium go, aetherium repo fighting if helium source code and then you'll have some convoluted way, finding the source code of swarm in it. And then, when you build it, then when you build it, you I must pull the whole assyrians repository and then I get five binary result and then I need to clear everything out and get just the swarm binary. If I want to run it same long.

A

Isn't that true, it's kind of it kind of its kind of not right way to do it. You know it's if if this was just a feature of etherium I would say: yes, of course, you know develop it, but this is like even more important pillar than than etherium.

A

A

Yes I: this is this this. This is what I'm sharing I'm sharing get a separate vista bleep, oh because this is the only one thing that you need to do to speed this project up because then, as a CTO I can pinpoint to my director look, this is warm. This is the source for delay, can verify. Then he goes okay, nice we can. We can set this to these guys to check about swaram. Well, is all about because we can read the source code. It's really hard to read the source code.

A

I always like to read the source code. When I see your Apple, because, let's let's go, let's go: let's go github github, let's give me a sec and then what's it what's big s know if they're ether spear? Oh? So it's not like this.

A

Yeah, so you know these kinds of fishes and I think that I really don't have any other objections to this project. I. Just think that we need some some exact point where everything that's happening, ah but I did contributed my energy to think about this for the last six months. Yeah it's a difficult to thank you so I think swarm a separate from gifts. I think swarm has defined, build and deployment processes that are exact for swarm. I. Think swarm must manage his issues independently, because it is an independent project and I.

A

Think swarm has provide an honor dr. image, ready to be run because developers of today are all about containers like five years ago, they're like virtualization and now it's containers give me a doctrine and it's up in five seconds. I, even I didn't test as hard into a machine for I know how how much because you get a docker image just runs. What they do is just apply configuration to it.

A

A

No I'm not saying this is easy. This is hard problem. What I'm saying is that each day that we continue converted with some other code base is going to get them more difficult, and my attitude and developed me some something: doesn't work. I, take an axe and I just chop it down and it goes down and you build it again, because you, you don't need it's kind of wasteful of energy.

A

So my message is at the end of the day: please you did the perfect nice job with swarm and I think that you need to refocus from implementing features first form to helping the community help you because they're, a bunch of us I have what I can offer is time. I can offer developers and then I can offer storage, because we have terabytes of storage free for swarm and I can offer some money, but I think the money is the least of our problems for this project.

A

So please allow us to gather around the codebase and please stop hiding, because there is a sentiment. Let me just wrap this up. I talked with people on this conference and when I talked with people, I said lockley. Do you have some issues with the get your planet everybody's like well here there is clearly convolved it way and ask it, but why is it like so and like few people told me you are doing it on purpose? You are flying under the radar and I'm. Not sure.

A

First of all, is this true because I'm not sure what would be your motivation to fly under the radar and I just think that we are all focused on swarm as a product and like what the features are and not what ecosystem that it will grow.

B

A

B

A

B

Unfortunately, that I have to disappoint you, but the truth is probably less mystical and and more close to this, we just probably suck at certain things we suck or or let lets, let's put it in a kind of milder and more to the point like objective way. We don't have been do it's for proper community campaign like evangelism.

B

We don't have bandwidth for for holding together about this, like apparently so any any contributions in that regard, even if it's like bullet points of what you find as pain points as developers to connect with us would be very helpful, and once once we have that kind of spurt out, we can start targeting these problems.

B

Maybe you can contribute even developer. Efforts like in.

C

This was exactly my onboarding experience as well. I'd spent several days going down one path, only to then learn or discover a wiki page over here. That pointed to another one. There are several wiki pages that are outdated and that just need to like at the top say, hey this wiki page is outdated.

C

Here's a link to the widest version, so you've done one of the things that I consider doing for a part of my talk was kind of collecting together I mean this is like a project manager, role that you've taken the initiative on it and done I'd, be happy to take your slides and put them as issues and to help manage the actual like pulling it together because yeah these were real challenges.

C

It's it's a I, think we're getting to the point that we're hurting ourselves, because new members of the community can't on board and contribute in meaningful ways, and so I'd be happy. I. Think I.

C

Think at this point, I do know where we can take this and we can put it and we can organize a little bit more and I definitely agree that Victor's time and other core contributors is better spent continuing to do what they're doing, and we do need probably part of the community to step up and take on more of the community management. The organization, the maintenance, those smaller tasks that every engineer hates to do, and so it's also a really good way to orient yourself around.

C

The project is by taking care of these maintenance tasks, so I'm happy to volunteer and help you get this implemented. Just one.

D

Other thing I was going to add, though, is that sometimes the decision-makers, the people writing the code itself are most aware of breaking changes or things that become obsolete, so it it probably would be good for them when they are aware to just at least like you know, post something on that wiki page. That says this is now outdated whatever, because it may not always be possible, or at least notify the community manager, whoever that happens to be right, which sounds like Doug just volunteered for that.

D

C

E

Track of where all the wiki pages are as well right,.

D

I think there may be just pockets of awareness that aren't fully overlapping right. So you know.

E

Yeah I just have a question more in like technical devops side, because I was using not OpenShift but like burning T's for a swarm cluster, and all of these tools seem really great at hiding thousands of micro services and containers in front of one load-balanced web interface, because that's what the web 2 needs. But when we want to run a whole cluster of swarm nodes, we want each one of those swarm nodes to be able to communicate and appear to peer network, which means they have to have an individual identity.

E

You know an address and a port, so they cannot just be copies to each other. So all of these great tools that could burn his automatic replication and scaling we couldn't use because they just make identical copies, and so we had a lot of replica sets with replications of one and then had 50 of those because it just didn't fit into the kubernetes mindset to have peer to peer nodes with individual identities. You know, as Nick Johnson told me, yeah and kubernetes think cattle and not pets, so I was wondering he openshift better at that.

E

Can you scale their spin up, 20 nodes and have them all be slightly different, different address, different port, but otherwise the same.

A

These kind of problem is opposite, aiming to solve and is basically solving it's a bit more convoluted, so it's not easy, just pressing, plus plus plus, but there are solution there are so called state rule sets. There are so called more more advanced stuff that you can do it's it's not it's not ideal, and you cannot move the manual work out of this. You know what you cannot move. You can easily move images, but you cannot easily move configurations. So this is something that there is basically no solutions.

A

If you need to, if you change your configuration and dev environment, you can, you must either write custom script to move it or you they just manually retype it. So this is not gods and solution, but it does help you to test and construct different topologies and then describe them and give them to somebody to easily replicate them, because it's just a JSON describing it so I think there are values in it and when you know what kind of deployment is optimal for you, you can easily push it to production environment.

A

A bit more I have a bit more faith in Red Hat setup of Linux than I have, in my own, so I kind of kind of like to have the nice nice button on which the thing is running on. So this is not the god set solution. There are challenges exactly like that. One. What happens when you have twenty thousand nodes with each different configuration, then you're?

A

Basically- and you have thousands of config Maps, similar with Apache Kafka for each topic, you get the config map and, in the end of the day, is like five thousand files. Configurable configuring, your own topics out there. There are challenges, but I think this really helps to build an iterate more easily, maybe not for the development. Maybe with many shift.

A

Yes, but when you decide to push it to production more seriously, then this really really really helps, even if you have manual work to set control configurations, but there's always ways to automate it if I can just and so to finish things up, I opened up a swarm, OpenShift repository, which is currently empty, just initialized. This is the place that I will hopeful.

A

In the next day's deploy some things because it's possible- it's not impossible, he's just converted, unfortunately not I'm here with my daughter, so I'll move out to home, but I would like to include you know online communication and basically thank you for your time and I hope that I wasn't too out there. But this is something that cooked in me for a long time and I just need to get.