OpenStack OSATX, 7 Dec 2012

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: OpenStack Austin Meeting 12-6

Description

Meeting: http://www.meetup.com/OpenStack-Austin/events/91683102/.
Speakers: Matt Ray (Opscode) & Josh Kleinpeter (ATT)
Topic: having hardcore discussion about Chef (first 30 mins) and Deploying OpenStack (second 30 minutes).
Slides: http://www.slideshare.net/mattray/chef-for-openstack-december-2012

A

Check that in to your repo uh pork, it you know dip it patch it. You know all those fun things to do with uh you know with your regular soap. So what this allows you to do is rebuild your infrastructure from scratch. If say, your data center burns down or you need to relocate your application from you know public cloud x to public cloud y or private to hybrid, or you know, linux containers to openstack whatever you need to do so chef makes it easy to do that.

A

um You know, and we have a big community. That's already built lots and lots of stuff, a very liberal license, which means people use it for just about everything um and the best part about it is our community. We are apache licensed, which means essentially patent protection, open source. That's it you know, so there's no! You have to share it. So if you don't want to share it, you don't have to, but luckily for us we recently crossed over a thousand people have contributed code to chef um outside of obstacle. uh Over 180 companies.

A

185 today have given their employees permission to contribute to uh chef um companies like dell, hp, rackspace, calvada, vmware sousa, you know, openstack companies uh and what is they contribute mostly cookbooks. A lot of patches to chef cookbooks are how we encapsulate a particular service or application, and you know there are over 700 examples of on our community side at communityappspace.com.

A

There are plugins for every single product out there, and uh but let's talk about openstack, uh so chef chef openstack is a community. uh It is a project, it is not a product. You know, there's not like hey, buy this skew and you get magical openstack. uh What it is is it's a community of people who are deploying openstack uh people who are interested in learning, how this works and putting it all together. The reason we had it is there are lots of people using chef for openstack.

A

You know, there's rackspace, there's hp, there's dell there's greenhouse calzada, there's att. All these folks are doing it and they weren't really sharing a lot between each other. um So what at the april summit, uh I was asked, hey it's great, because people are using chef, but let's try to get them all in the same code base or try at least so chef for openstack is an effort to get these people collaborating somewhat it's having moderate success uh yeah. But let's talk about what that means.

A

uh What is it? It is a chef repository for deploying openstack and it's documentation for uh for how that all ties together. uh It is currently seven cookbooks, keystone, glass, middle horizon, switch quantum and center, um and it is a nice plugin for deploying stuff on top of building stock.

A

uh Where is it? We have a mailing list. uh It's a google group, ops code, chef openstack. We have an rc channel, openstack chef on irc freenode.net, usually about 40 some people in there uh they're the github repos, for where everything lives.

A

uh Who's currently involved intel has sponsored the project. They wrote a check for me to write an intel cookbook. It works well with openstack working on that uh rackspace is contributing a lot of code. They had a version of the cookbooks that they forked from previous work with dell. That was forked off of work, that I've done all the way back to the bear release. So uh rackspace is uh very actively developing on hp. A lot of the code that went to early versions of hp's public cloud came from the same codebase very diverse.

A

Now, because I haven't gotten to look at it lately a dream host their public cloud is in beta. I think they're shooting for the end of this year, so they're running out of time, um but they're using the same code base. uh Obviously ox code.

B

A

You know uh some of the stuff in crowbar uh came out of this uh bare metal cloud is working on this as well uh marantis. You know, they're a large openstack employer cloud scaling has submitted some patches back, but they have they're using it right now, uh calzada submitted. uh um So what is it.

C

A

uh It's currently essex s6 works just fine uh on a bluetooth 1204. You can go to kvm right now. Let's see my sequel, uh you get slightly hcp with floating ips. It works really well with test kitchen, which is a test framework for from ops code.

A

That works really well with today.

A

The same code base is what's behind rackspace private cloud, their alamo project. If you go to github rcb ops, chef, cookbooks find them the same cookbooks. You know they're kind of a parallel fork right now. I've taken out all their operational things and those and take their patches and take them back and check for openstack.

A

um They have some opinions about how your openstack should be deployed. I want to let you do it. I realize um yeah. Maybe you have different monitoring choices, uh but if you haven't checked out our model, it's pretty nice way to get started uh for just you know, kicking the tires on an openstack deployment.

A

um If you've got like 510 boxes lying around pretty nice, um so folsom is I'm currently doing like a three-way merge between rackspace uh att and dreamhost, I'm trying to pull branches from each of them, because each one of them has different features that are already working um like, for example, uh quantum support from dreamhost. They have mycera and obs. I believe att is also using nice here um and mysierra called me today. Asking me to help work on makeup. Look for them as well.

A

So next here is getting very well done, so, if you uh you want to use, uh you should be in good standing. uh Cinder dream host has cinder working on seth um I haven't talked huh hct is using netapp, so very soon cinder will support at least those two back-ends, probably other stuff, um I'm going to be going to boston in january, and uh the league hyper v guy is in boston and we're having a chef for openstack day in boston and then two days later, one in new york city.

A

He will be at both of those we hope to have hyper v. Support in these cookbooks by then um and rackspace is working on red hat support. So it'll just pick up red hat support, um it's magical! I guess you get fedora too. um Those should all be done before uh before the next release, in which case we'll you know, change everything again, because all those things will have to be uh changed out.

A

uh Documentation currently lives in uh my repo, uh but ops code has about two weeks ago, open sourced all of our public documentation. It now lives on doc's opsco.com, um we've been pulling everything out of the wiki and making real documentation. So if you haven't seen the chef documentation lately go to docslashcode.com, that's really nice and the openstack documentation will become official documentation, not the data product, um so ties it all together. uh Scaling changes how we deploy right now the cookbooks are fairly naive.

A

uh They are meant for compute, plus one or one compute, node plus in uh oh wait, one controller, node and then compute nodes. uh Dreamhost has some patches to break that apart. So you can have multiple controllers and multiple computer nodes. That's still a fairly simplistic topology, I'm going to be working on aha support, so probably we'll. Let att talk about what they want.

C

A

um The nice thing is all this is getting shared and documented. You know, so it's slowly getting better and better, and all these forks are going into production, so that's kind of cool and because it's all apache 2 licensed anyone can use knife openstack. Maybe you don't want to run your own openstack cloud, but maybe you have access to one and you want to deploy stuff on top of it. So knife openstack is our plugin for knife. Knife is chef's command line tool.

A

It allows you to interact with different apis, including cloud apis, so I can do things like knife, openstack flavor list and get a list of all the flavors on my particular openstack deployment, or I can say you know give me all the images that I have access to and what's cool about.

A

That is when you're done, you can say night open, stack, server create, and this incantation here creates a floating ip address, backed by uh this is a bluetooth really well before, uh with my local key and I'm creating instances from the cli without having to use the ui. So I don't have to click and click and click. Throw this into a shell script boom. I have 100 of them, you know, run it with parallel and I can fire up vms blazingly fast without having to use the ui.

A

And that's what it looks like on the dashboard and there it is logged in yeah, just showing you that actually worked as about 30 minutes ago.

C

So what does that? What does the knife command get you there that you're not getting by just doing a nova nova freak.

A

Four slides back next four slides and I will show you: oh you have the same command flow for openstack hp, rackspace, ec2 cloud stack, eucalyptus, vmware, hyper-v, kvm, lxc. Anything else.

A

uh Basically, about 37 different clouds, all fold that same pattern, so what you get is infrastructure portability use the same code to deploy to openstack as you would to kvm or to uh mpc or to you know, ec2 um yeah. So that's why that's why you would use that, um and you know uh knife openstack has been worked- has been tested with piston nebula covar tri stack, you know it works diablo through folsom and you know I'm sure grizzly will just keep working uh once I get a look at the new auth stuff.

A

The thing I.

D

Mean it's really cool to be able to create vm in the command line. It's even cooler than use the command line and play your app on top of it. Yeah not have, and have that then.

A

um If I throw a dash r, I can put applications on the end of this. So then I can say I need a hadoop cluster. You know throw up a new master and 20 hadoop from the cli immediately. um So that's that's actually why it's useful. uh I can actually deploy infrastructure without having to touch a ui.

A

Okay, so what's coming up, uh you know fingers crossed by grizzly. um We should have hyper-v, we may have postgres um some folks at a chip maker are working on that um other operating systems. I've been reaching out to the deviant guys.

C

There's a guy who's using chef, debian.

E

And he's the deputy maintainer for openstacks, so we do votes with our uh oh yeah.

A

You guys do too. We talked about this yeah yeah yeah, I mean there's, there's postgres support floating around. You know I just haven't rolled it into these cookbooks, yet okay, um yeah and then uh dreamhost wants postgresql. So yeah.

E

A

This conversation six weeks ago.

D

The cookbooks that he's talking about are cookbooks right there. It's not specifically postgresql right.

A

E

Community events.

A

uh The eventbrite went up today or boston um anyway, uh and then the nice thing about chef openstack is: it touches a much bigger ecosystem than just openstack. You know the cookbooks that are built to support this are things like galaxy rabbit, mq, hyper-v, kvm, post postgres, my sequel python. All these cookbooks have been improved just because they had to get fixed for this, um so the community actually benefits a lot from him uh test kitchen.

A

You know our test framework um has been pushing a lot of the openstack stuff as well, uh because we dog for this internally. We use openstack to test all of the community cookbooks um that are out there. So those 700 some cookbooks are running on an openstack cluster that we maintain.

A

um We haven't published that much yet, but it's coming um you know and then projects like librarian and berkshire, which are cookbook management tools, um those have gone while it works spice weasel, which is a tool pixie dust which is similar to crowbar and razer, and all these other ones that manages pixie stake. um You know crowbar they're, obviously pushing a lot of this as well, so it has a bigger halo effect than just you know. Just guilty stack, which is kind of cool um and splendid cloud launcher might have something next week.

A

So if you weren't paying attention last 20 minutes, it's a project, not a product lots of contributors, real deployments, essex works, folsom's merging, and if you want a feature you show up with code, you get your feature so atod. That is a big slide.

B

That's yours, that's your slide. That's my slides great. What do you.

A

Know so how many data centers is att rolling out.

B

Five zones: we have.

B

More than that.

E

A

Do you want me to just ask questions and.

B

So so uh so we've been using uh chef for a long time. um Our first couple of deploys were with uh cloud scaling. um They use uh chef solo for their stuff and then they built stuff. On top of that, to do all the main.

B

um And so uh we started working on um our chef cookbooks uh and have been working on for a while. We did one iteration that was very specific to what we were doing um and, uh and we realized we weren't, really getting a lot of use out of what the community was doing, um and it was also hard to use what the community was doing.

B

It's kind of like one big repo and you can't it was kind of built to be copy, pasta right, because it's one big repo that isn't like sub-module or anything like that. That's probably that so, let's work on that. You know, let's put everything out in sub modules. So if I want one piece and not another, I can pull them and contribute back to just the sub models I care about.

B

So that's one thing that we worked on a lot.

B

We did a lot of a lot of basis for what we did was built from the rackspace cloud. Builders repos, but their repos are very opinionated you're, going to do networking this way and you're going to do volume this way- and we didn't want that. So um you know chef has lots of capabilities for variables and things like this. We add all that stuff in so you know we try not to be opinionated about anything as much as possible, um so uh a lot of our work has been around refactoring.

B

You know everything that's going on where you know. If there's an opinion, we try to get rid of it and let people say okay, you make your own choices, um so uh we yeah we've done a lot of work around that and um that's going really well um we're working on. um We finished, I think. As a friday, we finished kilometers. We added that in um as well, because that's another important bit, although unfortunately, the way silometer is built, it actually ends up being a recipe.

B

That's part of nova today, um just because thermometer, slumber metering uh framework as part of openstack, and so um it's useful for billing systems right it listens for events pulls them in and then you can query an api and say: okay, tell me how much this tenant has used a network or disk or whatever, um and so uh right now it's really dependent on the nova code base, and so it ends up making more sense to have it be a recipe, that's part of that um the salometer guys are actually working to pull all that which one of my guys is important tribute to that.

B

So he's one of the guys working on pulling all the common code.

A

Are they not using the apis or.

B

Just there's some stuff that they can't get to yet and yeah. So that's why it is yeah. um So there is a nova commons or openstack common product project that they're pulling a lot of the stuff, that's in nova and pulling it out into the separate modules so that you can use all that code separately and so they're working heavily on that project. Just pull that code out and then what you do, then you can have a true cookbook for salon.

B

So it's kind of unfortunate just because the way codebase is, but that means aluminum only works with both of them today um they don't have swift support yet, um but they have support for everything. I was actually talking to one of the leads on that julian today, um they've got networking all the network in and out they've got volume.

B

um They've got cpu.

B

All that so any any kind of data transfer stuff. So it's it's all there right now, the granularities to an hour um and what they're working on now is. um You know getting better granularity if you want it um and then doing this to an abstraction project um and adding swift as another component, because, obviously that's that's an important thing um and yeah.

B

So uh that's that's gone pretty well, um as of now, we have push button for us for full, just one command and then deploying everything, um I'm not sure how h a we are yet um there's a few things that we haven't gotten yet, like you know, software and balancing you know we're dependent on the f5 at this point.

B

But you know we like to be able to see the capability of doing software are.

A

You all building a public cloud, or is this a private internal, uh we're doing private content.

B

I mean it means I go to the command line and I run one command and it re-pixies every server it. uh You know, I think we have like one chef server running and then uh when we do the push button it it goes through and re-picks. These everything installs the base, chef, client, chef, client checks in his chef server says what the heck am. I doing and it says deploy, and so at the end of that I can start spinning up vms and getting ips so a single command to do that, the setup being.

E

B

So you know right now: um you know everything is run via jenkins and we, you know, make it make a code: convincing, conservative, hey, I'm going to go, redeploy a data center and it refixes everything and make sure it's.

D

All works right um and but does every deploy like a clean, install.

B

D

It's it's. I mean it's from.

C

The environment.

B

Right so it tears it down and goes back because we we need to make sure that it works from there. Now, of course, once you start doing continuous deployment you're going to be doing that plus you're going to be.

D

Doing additions and making sure that they can, you guys, focus much on the like the upgrades of live environments, yet.

B

That's yeah: that's the next step, get the initial deploy going and then start doing the upgrades, um because yeah.

C

What classes of users are you exposing it to it's primarily for developers or all across the spectrum? um It's everything we have.

B

We have uh I mean our cloud is private right. The products may or may not be public right. So one of the things um the developer program- uh api.apt.com that runs on our staff- um the speech text, uh translation text, speech, speech, yeah that stuff that runs on our stuff.

B

So there's uh you know lots of different programs internally.

A

So you have production workloads on openstack by the way yeah.

B

We have production report, I mean we've had to really increase our capacity.

E

Because of some of them, because people are like, I like this.

E

Spinning up the data somewhere, you know expanded just a little bit quickly.

B

So we've done that work.

E

B

Lot of it's a lot of it, I would say: well, I actually know I don't know if all that's not new so like some of it is like that news is, um they were on. You know hardware they wanted to upgrade, um so they came over with some of these projects that have been around for a long time. Hardware.

B

It's you know it's a huge pain in the butt to find a data center that you can deploy and buy the hardware for it, get the funding for all that kind of stuff yeah, and if, even if you have the funding, it's still painful right. Okay, whereas for us they're like there's credentials yeah, because it's easy so their turnaround time for our you know, developers their product teams are way faster.

E

So you have an upgrade, upgrade your physical to the virtual line you give. They give you the servers and well.

B

They, no, they give us their budget, we don't, we haven't. I don't think we've gotten into a situation where we're taking other people's servers.

B

It's possible, I don't think anybody servers. That's a reasonable survey.

D

So, are they the apps that they deploy on top of the stack? Are they also shaft deployments? So they just point at the moon.

B

We push automation um pretty heavily uh like as a matter of fact, for a long time, we didn't support snapshots right. Do you guys support.

A

B

Nope, don't doesn't work right, you guys have, uh and you know we would do things like. uh Don't have volume support.

B

You know, uh part of it was well running with diablo, um but uh but yeah I mean we, we didn't we didn't support or didn't say we supported certain features to kind of push people to like automate right, because you know, if I'm not and I every time I would talk to a new. You know, project I'd be like all right guys.

B

You know these are ephemeral, vms, they will die, they will die they're going to go away, do not manually, install anything, don't know the knowledge state of that cm. Don't know who touched it last automate automate all right when I get I had somebody a couple weeks ago that was like.

D

B

Try to restore my vm, I was doing a lot of development on it felt bad for the guy. So I like had somebody go and like find it, you know what was wrong.

B

A lot of times it's like somebody says: oh, my vm died, I'm like sorry.

C

Did you find any resistance or push back to say no to features because one of the things we face is they want us to have? They want us to build amazon before they'll, even look at it, and so it's kind.

B

Right uh yeah, I mean some people like we had one tenant early on. It was like. I want amazon, org, yes and build that. When can you give me that, like are you.

E

B

How fast, how fast do you think I could build that like that's crazy and they ended up they're running rds in amazon and using rpms, and it actually works.

B

That sounds terrible. It.

E

Does something.

B

A

E

um What do you guys use in terms of like hypervisors and networking uh we use kvm.

B

For hypervisor and networking um we have used nasira and we've also used the nova network, we're using quantum mechanics, yep yeah. That's our essex data center.

B

Yeah, um but with continuous deployment, we're kind of like hamstrung a little bit there, what we do like we want to do continuous deployment, and but how do I you know anytime, you're bringing a new vendor? Are they compatible with what changed in the last several weeks, like that's kind of problematic, and so you have to like push.

B

Can you talk to your guys here about that see if they're, oh yeah or the uh sorry are they is that something they're open to trying to stay in sync with trump um on money basis, yeah.

C

B

I mean they know we want to do it, so they know the way to get into the game is to figure that.

A

Out to figure out how to do it and then they called me they're like we need you to help us make the atp happy.

A

I was like you know: I'm working for hkt, oh.

E

A

That not before the call.

A

Instances being ephemeral, I've been talking a lot of it management lately and they talk about cloud.

B

Stuff- and I point out that anything you run in the cloud you have to consider- is the federal you're going. What that's right you can't depend on it.

E

Continuing to run there, you've.

B

Got to be ready for it to die at any second, it may run for an hour. It may run for a month, and it's.

C

Interesting because.

B

You know at a t there's a lot of like really long term. You know 20 30 year, architects who have always designed you know they know where they are going to be in the data center and they know every nut and bolt that's going into a server right and they're speccing. Every little piece out.

B

You know so they're building their hardware to never fail. You know and spending ridiculous amounts of money on that hardware. Make sure that that never fails, and then they write to that software and you're like in having those people come in and say: okay. Well, you know how do I? How do I do this? It's it's a very interesting conversation because it's going.

E

To go away right so you're, pushing your users to use chat files and no automation, so automation.

B

I I don't. I try to remain neutral to that. I say we use chef, we use chef for our infrastructure, we recommend chef. We can't help you if you do something else, but I mean I've. I told people like look if you want to write a bash script and pass user data when you start the vm, that's a start and we've had several tenants that have started there um and and that's fine right. You figure out so.

E

The idea in that case would be that you have me, and the bachelorette.

B

That's basket: I'm gonna run it okay right, so you can just do app, get installed and then start this process change this thing. Whatever um so yeah I mean I've.

C

B

People do that at the very least, because it's super easy right. It's like I, don't there's no learning curve, but if you know bash, you can do it um and that's worked out because people have been like oh wow yeah. I can see how this is useful. Bash is terrible. I'm going to do something else, so I think I don't know of anybody using like cf engineer puppet. I think most people do chef.

E

A

You guys are wrong. Some people use hosted separate chef servers for each data center right and they've. Automated the deployment of chef.

B

Right, so you expect your users to deploy their own chat server. Absolutely.

E

What about other managed services load, balance messaging.

B

um That is, uh we've helped people out in some cases, especially early on right, they're like oh. I don't know how to set up aj proxy. All right, I'm sure we can help you set up aj proxy we've learned to not do that because then you're stuck with that patreon proxy forever.

B

um But yeah I mean we at this point. We do recommendations right. You say: okay, you want load, balancing, we're, not doing load balancing the servers. Yet here's some options. You can use right, you use apache or nginx or h proxy set those things up. Did you enable enablement like clever.

E

Pre-Made, clients, images or chef recipes.

B

We we don't our all of our os images are vanilla, we don't install anything on them, except once it comes with a very vanilla server, don't start anything except for sshd, because at least in my experience in a lot of other people's experience, if you install anything on that thing on that box, that runs, they don't know what it is. They don't know. What's there, they never look for it. There's a security hole right there waiting waiting to happen. So that's not good.

B

um So that's why we leave it super vanilla and then we're we're actually talking about like well. How do we do something like a chef as a service right or because we want everybody to maintain their own check? Server right and some people use hosted chef and we definitely tell people hey, go use, hosted, chat right. You got an account yeah, you do it. You go, do that, um so there's lots of different ways to do it and there's so many different groups, and they do so many different things.

B

I have different skill sets that you know I try not to be prescriptive. Heat's going to be interesting to see how that um that plays out and see, if that's, you know a useful mechanism to do it. I think I saw the the screencast that the guy did today on on heat. It was, it was actually pretty cool. I mean he was.

B

uh He is like a tool that allows you to like it works like cloud formation in aws, so you can like point it to a json file and that's your definition for what your your virtual data center looks like, and it will like set that up for you um and it was cool because, like through the web ui into the cli at you can pick uh actual cloud formation files from amazon and it would set it up on openstack and and then they also have their own github repo, that's specifically built.

B

So they implemented the the uh aws cloud formation api and then they also implemented a new one for heat as they're eating so kind of following.

D

Them so you're clearly contributing back into the chef streams right. um Are you also contributing stack upstream? I mean speedometer clearly yeah.

B

So um so everything we've done with chef we've, we've pushed the mat and these I'm sure.

B

All the chef stuff we've done is has been contributing. um We have a cornova contributor uh uh who's done. I don't know a lot of writing he's like he works in uh research uh labs his phd and he thought he was having fun and he was like he's starting. He started doing all sorts kind of cool stuff. um uh John tran does uh works on cellometer, full-time jay pipes. Obviously, he's got crazy contributions all over the place.

B

What else will probably be contributing our cli that we're using for deployment this thing called substructure so we'll see whenever we get that ready, but that's kind of what we use to to invoke a lot of this stuff um and what else um I think kevin's got packages? Oh yeah! Well, uh we've got actually a whole ruby team that is contributing a lot to fog. They fixed a ton of stuff in fog um and uh yeah so that those guys are doing that and then um and then kevin's an encore, documentation team.

B

So he's done a lot of stuff with documentation.

E

B

We do, in fact we just we just upgraded to the full symporizing for our.

E

C

Speaking of upgrades, how are you handling uh upgrading from version of version you're open attack? Are you just ready to play or hey?

C

E

Oh you've got your one button.

B

Oh yeah: well, I can avoid data center. I can't.

E

So we are going to have to do it via migration.

B

And like helping helping tenants migrate to vms snapshots pc, um so it will be. You know, here's.

E

B

Setups all right guys move over, and hopefully I only ever do this one time, but I mean we've got guys running they're running production, applications, they're or diablo.

B

You know it's it's just hard to go and take all those guys down, and I mean some of the teams go through ridiculous processes to certify their application before they do their production like months of certification. So.

C

It's just short: I've got.

B

C

That take two years to do a bill so yeah I break a certifiable.

B

Yeah so I mean they take a long time to do it, and you know you can't be switching them all the time. So hopefully we get them moved over and then we start actually trying to upgrade folsom continually.

B

D

Going to do that a lot because so are you going to pull from me? Do you are you relying on a distro for that, or are you going to switch ultimately use source course? Okay, so you pull from source. You need that now we're doing a distro.

B

Now I forgot what we.

A

D

Yeah is that is that nice, I have a lot of merch, because yeah we've got cookbooks to do. Pull from source also be interesting to combine that.

B

Yeah, and in fact I think the way we've got it set up is that you could pull from source.

D

D

A

D

A

Yeah, that's yeah! That's what we're was.

B

A

Based off of carl's work of a dream house, because he was doing quantum.

B

First yeah like very possibly I know we, I know we talked a lot with dream hosting.

A

Yeah because yeah.

D

A

Carl and dewey- and I were talking about whatever a couple months back and that's when carl's like what I got.

C

They still doing the same things that are still doing the same thing. Oh yeah, I just touched them last week. So what does the windows crowd.

E

B

Yeah, that's the thing is this: what do you mean my vm won't.

E

B

Around that kind of thing.

B

And I I definitely complained about automation because I know earlier on, one of the operations guys was like I I was explaining to him. Well, you know what the firmware looks like, but it takes three days to install the software.

B

So yeah I mean I haven't, heard a lot about how people are doing that. Automation and I mean my team's linux.

E

Guys, they're not right right. Well, they're, not windows, guys. What are you running kvm on? huh What are you running kvm with the operating system.

B

With microsoft,.

E

uh And then have you had groups that used vmware prior and they're now using this stuff yep.

E

B

Well right, but when their vm dies then they're like right. What happened? Why isn't it still there like server diet, but so what yeah well.

D

You went with the cheaper solution, so okay, we've got a question from the remote audience. um There's one person that never tells, but uh he wants to ask you uh what needs fixing and fog.

B

What needs fixing and fog crap? I don't know so I happen to be very fortunate- to have a ridiculous ruby team, so I've got ryan davis, who's in spider and dr brain ericotle and tinder love uh aaron patterson, that's my rookie team. So I don't have a problem, so they fix stuff like whatever they're a bunch of lunatics.

B

They are a bunch of lunatics but they're awesome.

B

Is I guess the fourth member see I don't beat him in los angeles.

B

So earlier you mentioned getting data into a pm on. Where do you guys come down.

E

In terms of file injection versus metadata service versus config drive, what do you guys recommend for support.

E

E

I'm not sure how to answer that question. We use the metadata server.

B

Yeah um yeah, we use metadata and we we do everything we can to make sure those vms are like tight to like turn off password authentication, and you know make sure that we don't have any default security group, ports open or anything like that. So when you bring up that vm, you can't even get into it. You know.

E

And you use what you meant in all your base images then.

C

D

Yeah, what about that new cloudiness for windows.

B

I know we totally need that um yeah to try that out, because yeah there's there's stuff with windows that are like just you know, problematic to say the least, but uh but yeah I mean that would be.

E

Are recommended.

B

Is the metadata service? So it's like, we have a version of cloud in it that runs on all of it right right, um yeah that would be really good to have and we can we're working with microsoft on our images.

B

I guess so: um yeah windows is fun, I mean you're, like you know, you got the licensing issues you're like okay. Well, how do I talk to my licensing surgeons and get about to happen and that's kind of hectic and fun? And and then you have like we've had people complain because we have like 10 gig boot drives right and they're like well. I need.

E

More on my c drive.

B

Because the software that I'm running has to be installed on my c drive and it's like 80 gigs, I'm like crap, I don't want to give everybody any partitions.

B

E

Is going to be really nice.

B

With with tenant-based flavors that we can provide- and things like that, so you know that'll give us a lot um a lot better control over how we you know dole out different flavors, because people all want something different. You know, and I it's problematic, giving everybody a 64 gig image, because, unlike a public cloud, you're not really paying for it. So you're, like I'm gonna, run engine x, 64, gig machine.

A

When I was on hp's private beta, that's what I did.

A

A

uh No, so we don't have a good way.

B

Of doing a chargeback.

E

Right I was going to mess with.

B

Thing is, if I do any kind of like if somebody gave me their their corporate mx and there's tax and locations, oh yeah and all kinds of things right, so you've got to do it internally, and I don't know if we have anything like that: choose our currency.

B

Yes, um actually, what I'm planning on doing once we get the slumber out, there is start generating monthly bills for them anyway, right, even if you're there's not enough, and if I know like how much that they've donated to us, we can actually have that in the system.

C

Even if you're not actually building just notifying them, what you're, what they're.

B

Using yeah, I think that's what it is like, because I I mean I was looking at a data center today. It's like there's two 250 vms in it that they're not actually using anymore they're doing something else. They never took them down, because why would they I didn't think about it? I'm like guys using this, give me the hardware.

C

What does your hardware layout look like several wise and did it drive your choice in flavor sizes.

E

Or did you drive your hardware choices by your favorite.

B

Machine so it's easier to provide a 32-bit flavor when you've got 256 feet when you have 72 it's kind of a painful choice and in fact we did energy is standardizing on a single size, machine ish. I mean, I think, we're trying to figure out what the best balance is. At this point, and, like you know, do I need a 256 machine. Can I get away with 144?

B

Is that okay and where's the you know, where's the price point? What am I going to pay for? You know? Do you need to get you know all these great discs for your? You know on your local computes, or do you not? You put a small amount of space and we're at the point. We need to start really dialing in what that is.

B

We do use many many vendors, many vendors. We don't use one with a genius, so that's fun and exciting, because you're like did you, try and use documenting.

C

To take out management.

B

Or, depending on, if that vendor provides something other than ibm, I might use that too. Okay, so you're not trying to so. This is like another chef. So um initially, when we did this, uh everything was off the pixie and we configured the servers off the pixie, and then we put the chef client on there and that installed it. Now we don't do we do as minimal a pixie boot as possible.

B

Don't do server configuration and all that configuration happens in check so setting up the raid any kind of buyers configure anything we can put off as long as possible. We do that all in check, though right and that's way better because like trying to have especially when you go to something that's heterogeneous, how do I figure out what kind of server this is and what I want to do with it, and the chef already have.

C

Tooling, to talk with all the different uh like uh hp regulating tool right now,.

B

uh No, I think, that's kind of really real. You guys have been.

E

Writing a lot of things.

A

B

That but it all depends, I mean you know, you've got.

E

You know if it's hp or dell or you know, cisco or.

B

Whoever super micro, whoever it is, they've, got something poorly at least got ip9 and you just use whatever works best for you. So.

D

What about quantum is that part of your deployment remember.

C

B

It in folsom we're not and do we know the network. Okay, that makes sense. Are you planning to.

D

Adopt quantum is that going to be a challenge for users to have to map into the networking.

B

um Well, I don't know if we're getting a lot out of it without.

E

Something behind it: it's private data center, so talk.

B

To so yeah- and we know the network is fine for us for now.

D

Is it going to be an issue that uh is being deprecated.

B

D

B

Once it is, but I mean yeah I mean we'll use quantum, but I think right now.

D

I mean everyone will use cleanliness, that's well! That's why I'm it's really a challenge on the upgrade story, as it is right because yeah, if you're not.

B

It's nice to give the tenants the just continuous ip blocking.

D

That's yeah. Well, I guess that's part of the real question is: is it going to pose challenges for adoption for you when it's you know if they have to then use network, can do networking configuration or.

C

B

D

B

I mean yeah yeah, I mean we'll have to figure out our automate around it.

A

um We recently released uh cookbooks for arista switches. We actually can install chef on them and manage them so and then uh dreamworks is doing that with uh cumulus based switches, also for sure actually.

B

Simon chin, another one of our ridiculous labs phd is working on what we call nas network operating system. That's kind of open source, that's going to help us register any switch into jeff so that we can watch all of our switches, not just our arresteds, we're just as easy like. There's linux back there, okay.

A

So yeah same with cubes.

B

But if you don't have that, then what you did yeah.

E

A

So is he writing so cumulus? Is they are selling just the os for bare metal uh switches? Is he writing uh an open source version of cumulus? I don't know, maybe okay, because I yeah because it's fedora based also so I don't know.

E

I don't know how open source it is yeah.

B

I don't exactly know what he's doing I haven't looked at it too deeply and he's working on it. I'm pretty sure he's doing it.

E

And look how long you can speak to one slide.

B

I did seven hours, I did seven hours with 70 slides.

E

Version message.

B

No, actually, that was all open stacking, much more gory detail about what we're doing.

E

Much yeah, so I think uh no, not you don't have to go home and you can't stay here.

E

There will be one next up the next month again.

E

We're thinking um actually miguel to do a such session on quantum uh we're also thinking of actually like lagging it all out. So for people to be able to you know, that's early thought process, but for people to come here and actually like get hands on and have an environment that we can all destroy.

E

um I'm pretty sure it will be not exactly a month from now, because you know it's january and everybody's off scheme, so uh we're thinking, maybe mid-january uh so yeah. But uh if you keep track of the meetup, then you'll get the notifications. All update is needed and so on support. So.

C

It'll usually mean a little later in the month, but december is a special case right. This is this was earlier.

E

E

Returning back to like the regular later in the month, kind of especially in january, contacts where people might still be sleeping whatever uh right so yeah help with the pizza. If you haven't yet, as you can see, we held it there.