Kubernetes Community, 19 Nov 2015

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes Community Meeting 20151119

Description

We have PUBLIC and RECORDED weekly video meetings every Thursday at 10am US Pacific Time.

https://docs.google.com/document/d/1VQDIAB0OqiSjIHI8AWMvSdceWhnz56jNpZrLs6o7NJY

This week included demos of updating the control plane with the Deployment object (Mike Danese) and Open Contrail on GCE (Pedro Marques from Juniper) as well as a 1.1 Release postmortem discussion and an event report after Kubecon!

A

A

See boy watchin watching you ask my journey part of children. Oh.

B

Ok good morning, folks, let's give it may be another minute or so and see those drawings.

B

On just as a note, Donna's recording I'll say that a couple of times during this meeting just so, we remember that we're being recorded.

B

But does everyone have the Hangout document the working document for what our agenda is today.

B

Or enough of you that are not unmute, that I know that I'm broadcasting to you Sarah question: okay, I'm.

C

What's the what's the plan to do with the recordings, are they going to be published somewhere or is this for archival or what's the deal right.

B

The answer is yes, the plan is to publish them somewhere, I, don't know if you guys noticed I just updated. We came to be hangouts right to say this is a public meeting and it is recorded. So I'm laying the groundwork for google legal to let us Cesar.

B

So my hope is that by the time we get to the first of the year and or possibly sooner, I will have permission to publish what we've been recording internally.

D

E

F

So I came in a few minutes. Late did the the calendar invite wasn't updated this time? Was it or did the calendar invite disappear? Are we going to keep you doing that so.

B

There's a calendar invite that it was everyone who was on the old calendar. Invite has been added to a cooper, Nettie's community video chat group, which is at the top of the cooper Nettie's meeting document. And if you join that group, you will always be invited to and seeing the most updated calendar items. So we were starting to get into a crazy list of people that was hard to manage.

B

There are a couple of people since who have been added to the calendar item directly and I'll, be adding them to the google group, but basically this. This is just the here's. How we keep the calendar item for people who want to go to the video chat, nice and clean I.

F

Thought they had joined it now, I just click through and I head. So that's why I didn't see the video chat I'm that the meeting invite.

B

Well, welcome welcome anyway, um so we have topics since we've got people gathering I'm going to go through the one specific topic. That's super short next week is the tectonic conference and the CNC f, the first cnc f, not meeting the cloud native compute foundation meeting is also on the third. Are there enough people that want to have a meeting next week or is everybody's sorry, two weeks next week is Thanksgiving. Two weeks is the and CNCs meeting on the third?

B

Do we want to go all the way until the tent in order before we have our next meeting, or does someone want to host the meeting on the third with topics that are listed there?

B

Do we have enough people that will be around on the third to participate, I.

F

Will be traveling so will be difficult for me. Yeah.

B

I will be as well I can't host it on the third I'll, be at the CN CF meeting.

C

So done: I'm traveling men too, okay,.

B

So this sounds like a large number of people will most likely not be available. I will bump content from the third until the tenth and then we can pick up again on the tenth awesome alright. So that was my easy thing. So I have two possible demos, for today is Mike Denise on yeah awesome like do you want to go ahead and do your control playing demo yeah.

G

Sounds good just a second I'm trying to figure out how to present to the screen.

B

Believe it to thinker screen screen share button is.

G

That good working, yes, um okay,.

G

Do you guys see a terminal I'm doing my best if I full screen it? Do you guys still see it or does it go away? Great, alright looks great, sounds good. All right here we go so.

G

Right today, I'm giving a demonstration of a bootstrapping a cluster with the controller manager and scheduler running on the cluster, and then I will show a upgrade of the control plane using a deployment object and if you are unfamiliar with the deployment object. It's a new object that sin. It's a controller of the replication controller and it manages rolling updates of replication controllers all right and feel free to stop me or ask any questions during this. So right now, I have a cluster deployed in GC.

G

It's just a normal cluster built from head up step forward. It doesn't have a controller manager scheduler running right now, so it's actually a pretty useless closed cluster. So what I'm going to do is I'm going to start a controller manager I have a controller manager packaged in a pod. Here's the pod template.

G

Just normal pod template it's running an image: Mike Denise, cube controller manager. V1 k is my shortcut to cube control, so k create FHA, KCM, pod yml.

G

Alright and that pot is created but pending, so the reason is pending is because a scheduler doesn't exist and when you create a pod, you need somebody to write the node name, which tells a a giblet, sir anyway, to fix the feedback.

G

It tells the cube 'let to pick up this node, so we have to do a manual, bootstrapping or manual scheduling of this note. So what I'm going to do is I'm going to run cube, control, patch and I'm just going to pick one of the nodes, CZ 8k and I'm going to schedule that controller manager.

G

G

Watch my pods that bootstrapping controller manager is already preventing, so it should be good to go now. What I want to do is I want to create some deployments for the controller manager and the scheduler, so these are deployments that you can think of them like replication controllers, that the actual API object is very similar to a replication controller and I have one for a scheduler as well so I'm going to create both these.

G

So that was the controller manager and there's the scheduler. So now, if I get pods or if I get deployments, I.

G

See the two deployments, the 0 updated replicas right now: I get our seas and I query over the label control plane because I labeled, all these are sea is so I see two to replication controllers running the both currently have 0 replicas, but that should change soon as the deployment loops and decides that I gave it 3 replicas.

G

So now we have three replicas, but if I go back and look over here, I see that the controller manager created the pods but their impending, and that's because there's nobody yet to schedule a to start scheduling pots to start writing no names, so I actually have to write one more node name. So I'm going to do the same thing that I did with the controller manager and I'm going to manually bootstrap a scheduler by manually scheduling it sending a patch to write the node name.

G

So there goes and.

G

It looks like the scheduler is running is starting to run, and we should see that the other schedulers go pending once the scheduler starts, and also the controller manager should should go from pending to reading as it gets scheduled.

G

G

Since they're, just pods I can look at the logs. Oh.

G

No, that's a bummer.

G

Well, I must have pushed a bad revision of the scheduler.

G

But what I can do it is. I can save this demo by.

G

Just start scheduler so glad this guy's. Let's.

G

G

So I'm going to stop the deployment of the scheduler because of my schedule. It appears to be broken.

G

G

Just may be suffering for the same one, yep well, I think my demos is just about done, but I'll explain to you what I was trying to do. So I pushed a bad revision of the scheduler earlier this morning. I should check that, but I had another meeting is to go to this morning. Of what I was going to do was I was going to update the deployment object by running a.

G

Cute control, edit deployment, controller manager and.

G

What you can do, once you have these running and deployments is I, could flip to the new version of the control manager by modifying the deployment object. This would trigger the deployment controller to go and spin up new replicas of my new version of controller manager.

G

Unfortunately, I can't show you that today, one aspect of this was that the controller managers and schedulers were running in a in a HJ leader, elected mode and how that how that's implemented is using rotation on end points. So, if I went and I looked at the controlling manager, I.

G

Can see that there's a annotation on the controller manager, endpoint that says the bootstrapping controller manager is the leader and the other controller managers respect this annotation and they don't begin their sink loop until they are able to acquire this annotation.

G

So this allows during a rolling update once the the leader is deleted, because he has an older version than the version that the deployment is rolling towards a new controller manager with version v2 will pick up from this from where the older version deployment controller, left off and I and finish the deployment. So it works pretty nice and seamlessly sorry that I could not demo that to you guys today.

H

Right I saw the demo last week and it worked last week, so I got like a one on one version and I can attest to the fact that it worked thanks.

G

Jay thanks Jay. Let.

F

Me say that even though it didn't work I think it's awesome, absolutely awesome to start seeing humor Nettie's self host its own components, so that's very, very cool one idea was: is there any way to run the scheduler in like a one shot mode right where, instead of actually having to run that, you can say, hey scheduler like do your thing for one loop and then exit.

G

Right, that's, we have a run once for the cube lit and we need to implement this a flag or they do it. Just starts up, starts the containers and kills itself. We need to implement the same thing for scheduler and controller manager, but definitely a good idea and then.

F

Sell posting the API servers, obviously.

G

You, like you, know a next step out of that um it's a little trickier, but we have some ideas on how to do it. um Definitely a little bit trickier yeah.

F

The trickiest who you are you were actually.

D

Oh um yeah, the trickiest bit is getting the qubits. Are the cute proxy to find the API server in the API? Is perverted finance, ed, so I so.

F

Here's one thing: that's always bothered me a little bit. Nha scenarios is that we've said quite a bit that hey it doesn't matter if the API server reboots, because they're everything else just coasts right and so a little bit of downtown with the API server is all fine, but as we move towards doing more and more name resolution and sort of discovery in the API server with things like the endpoints API, that does become critical path.

F

So is it worth actually similar to board with bns thinking about sort of like, like factory now naming from the rest of the clusters, then then running the API server you know in the cluster and having it actually sort of float around becomes much more yeah.

D

So we're have a couple of things in the works that I think we will look at that. I mean I. Do agree that the more logic you put in one component, the more likely it is it that component that component at some point at some point its employees, bad update or you know otherwise, get into a persistently bad state.

D

So we're we're actually working federating API servers right now, so we can actually split out logic into multiple API servers. So I think the endpoints logic would be a prime candidate for moving out and we're also thinking about moving the DNS logic down to the notes into Q proxy or something like that and eliminating the additional instance of FTE. That's required for that, and also eliminating the serialization of in points themselves to that's to s.e.e and then potentially constructs that data is a view / cached pods and the new the representation of external services.

D

That's an works, so so yeah. You know I think over time will eliminate some of those dependencies and factor that out.

B

Awesome other questions before we go on, so we can get on to the contrary mo all right. Thanks, Mike um I, oh good, you did stop sharing your screen that lets meet. My next request.

B

All right so way fix thanks is watching so again. Oh.

I

We can do it again. Yes,.

B

Top marques these every week, so the control demo from Pedro uh and juniper Pedro are you about on here. Yes, so.

A

Yes, yes, excellent.

B

A

B

A

B

You'll be doing the demo awesome sorry about this I'm.

F

Getting an echo, everybody else should mute, then, if somebody else might be so David.

A

Okay, so let me share my screen here.

A

Okay, so before actually um you know going to the actual demo, I'd like to share some slide deck on. Can you see the dick good? Thank you. So essentially, what we've done is open. Contrail is going to provide some of the niceties of overlay, networking and managing overlay networks, as well as the providing multi-tenancy and tenant isolation in the existing OpenStack words we are trying to. Basically, you know, bring those niceties that are already there as part of the OpenStack deployment into Cuban IDs, which is basically enabling orbs to basically have isolation.

A

So when applications get deployed in parts, they basically have private eye, peas and the eyepiece, basically another pod basic. You cannot talk to each other unless there is an intent for the developer crucial and that intent is basically a exposed wire annotations by labels that we actually apply. So the the good thing about the open contrail is basically, there is a community.

A

It's open source and community is basically developing some of the components for it, and it is, you know, deployed currently in some of the service providers as well as enterprises like medium large enterprises who are running SAS applications.

A

The the model that we actually followed in integrating with Cuba nities is replacing the cube proxy and then providing the video outer as the functionality to basically you know service networking further for the pods, so the value prop for open contrail is basically it's a distributed. Router. It actually can serve not just the content. The network for the workloads that are containerized but also can actually provide networking for the existing workloads that are either running on the virtual machines in OpenStack, or you know, plain vanilla, kvm based verse.

A

You know cloud worst cloud, meaning basically it's just using the word tools and standing up the VMS and as well as bare metals. You know some of the customers that we actually work with have workloads that cannot be virtualized, so you know with the contrail. It's basically, you know can provide networking seamlessly across all these.

A

The other thing is basically multi-tenancy full isolation and fault tolerance, so, basically having any kind of false, not being you know contained if at all, there is a compromise of a particular part, then the failure is basically contained to that that pod itself so going here. The implementation wise. What we've done is: we've basically developed a cube network manager that actually is running on the master talks to the queue API server and the southbound.

A

It basically communicates and updates all the objects that are required for control and then on the node it may we have a cube, let plugin and then the kuebler plugin is what interacts with the V doubter and the video outer. Basically as two aspects, one is the kernel module and there is the user space model, so the key components. You know what: how does it actually tie into Cuban at ease it basically encontre? We have virtual networks and a network policy, and these get exposed as labels so named. Basically is.

A

If you look at this, one is more interesting. Is you know a pod definition when we define a part part in the in the metadata label? We basically have a name which actually relates to the virtual network and uses is basically you know that part needs to use some of the service, so it actually exposes that services uses class, and then you could. Basically, you know have that policy tie-in between two parts, so essentially two parts cannot communicate if it all without the users cause.

A

This is an example of the guestbook application that was that I'm going to actually in a run in some and now that actually has named Redis and then the front end, which is the guest book app basically has uses, and if you use this Redis so without that uses, guess boot will not be able to communicate with the Redis cluster. So standing up is basically very simple, so it's as simple as basically providing expert network underscore provider equals open country lens and Duke you up.

A

So the salt changes for getting all the Quantrell components are already in place. There's a there's, a there's, a PR which is actually pending. So hopefully you know. Tim can look at it, it's 12 Tim's queue and then now let's actually run the demo. So what I've done is I've already already just this morning, sometime back I already have this cluster, which is done so, if you see here I, just actually van cuba with a tent Cuba, the environment variable actually set. As you know, you put a wider open country, send.

B

You I think you're still just sharing the despair sharing the presentation, not the terminal window, okay,.

A

Let me I could.

J

Just a second sort of in them.

A

Here, okay, here you go, I'm going to show you.

A

Can you see it now.

K

A

Better good, thank you. So actually what I've done is, if you look at this, so all it takes to actually stand is basically you can export.

A

A

Provider, open country alright, so once this is done, then I can basically go to wherever I've deployed and then I can do a cluster of.

A

So this would actually essentially stand up the cluster I have already. I already have it running, so I don't need to basically you know stand it up so in this deployment. What I have is this default three nodes, one master and in addition, I basically have a gateway, node, and all of these are basically again running in as containers. If I do, you get parts.

A

So if you see here, these are all the contrail components which are basically providing control functions as well. As you know, updating the video return of module. They are, essentially, you know, running as containers, so these are the three nodes and there's one open content will gateway, which is actually a separate vm. So one thing to mention also is basically I think cuban it is basically have a container vm which is provided, so this is actually running on a debian 79.

A

It's a debian 79 virtual machines that spawns the master, the minions and the gateway node and just on the gateway node. So let's actually go to the gateway node.

A

A

So this is the gateway node and here, if you look at this basic links will be during the deployment it actually goes in seamlessly installs the kernel module and the kernel module is the one that actually has all the IP routing so for all the next hops broad tables and all that stuff, and then you basically have here.

A

This is the the V dot agent that communicates with the kernel, module and updates all the routes and all that stuff.

A

So we have three such nodes and then the gateway, node and the reason for the gateway node is basically to provide north-south communication. So any traffic that is coming into the part I mean needs to actually come. We are the gateway node, so it's like a gateway for receiving external traffic coming into the part.

A

So now, let's look at the one of the apps, so I have the guest book app and the diff here is basically, if you look at the Redis master controller, so under the spec selector template metadata. So this is basically the name.

A

This is the the name which actually boils down, which translates to a virtual network. So this part, if you basically go to I, didn't actually deploy it. Let's actually quickly deploy ok, 38 minus this book is still controller.

K

K

Think pending come.

A

On okay, so here you got that and now service for it.

A

Will have slave controller.

A

K

Okay, k, minus f.

A

Okay yeah, so basically, here still it is slave lettuce, slave service.

A

Okay, so you have that and then you get basically the guest book controller.

A

Okay and the source, so what essentially this guy now is service actually does is, basically you can say so there's this external IP that actually gets assigned as soon as the guest book application.

A

You know, gets deployed, get this book as a service actually gets deployed, and the reason is basically, this exposes the whole external traffic to actually come in. So now, if you basically king this, this is essentially coming where the gateway node, and we actually pinned it to the gate we know to. Let me now show you.

A

Okay, so let me now show you: can you see this now? Basically, you have just booked up and okay, sorry, so what I need to do is basically create the tunnel again. I have the tunnel here. Okay, so if you see here now, basically you have this kiss book app running and it has a connection, so you could actually say one, and this is basically the way that it gets routed.

A

Is there are replicas for the guest book as well as replicas for the for the Redis, which is actually running and the way that Quantrill is basically doing is basically ICMP ecmp to the the parts.

B

A

Yeah, so the whole deployment actually is just so simple and easy to deploy, and you know once it is stood up now. The beauty of it is basically that you cannot have pods or talk to each other unless there is an intent by the developer to do so and when you're developing an application, the the developer. Basically, it has this name and uses that can be provided into the pod definition to allow secure access so to say so.

A

B

You want to take a couple of questions if there are any send you sure.

A

L

So I had one question, which is: what are the plans for namespaces and do you plan on supporting namespaces? Do you already support namespaces? Yes,.

A

Yes, that's a good question, thanks Brendan, so we do support namespaces and you know the. Whenever the pod gets actually deployed across names cases the same pod, then it provides multi-tenancy for the namespaces.

L

uh That's and so can I cry, can I do a network across a namespace or is that just explicitly forbidden, no matter what? Yes.

A

A bedroll is basically here, so he can actually answer that. So.

K

At the moment, we have two things that are implemented: it's a very fault when you say when you use a name label, you get a network inside your namespace. We also have a different annotation that creates global networks. For instance, the DNS pod is in the cube system namespace. So to connect the cube system of the connected DNS pod with everybody. We have a different configuration that just says well that the cube system Network, where the dns service is, is available to everybody.

K

What we don't have today is sort of this in between where the user can create.

K

You know, boat and a name and then have a white list of who can connect to it, and it's not so much a implementation issue in contrail, but how to model it in communities, because you know what we're trying to do is try to drive it from the labels that are, you know visible on on the existing coordinators objects.

K

Does that make sense, yeah.

L

That makes sense than any other sort of comment. That I would have. Is that in general, I think that we have a preference for people using domain names as routes on the kinds of fun like magic labels, that they add two systems, um and so it would be great to see the name and network labels turn into you know: juniper, calm, / name or some open, contrail org, slash name or something like that. Just so that we don't have collisions right.

L

You can imagine that if three people want to use name for the same magic thing, suddenly, they're gonna have to choose, do I use open, contrail or do I use whatever else yeah.

K

L

K

L

It's a committed, a perfect yeah.

K

I'll I'll do that that'd.

L

Be great um and also I, think it'll be great at some point. Community-Wide you just kind of have a glossary and taxonomy. These kinds of things have some way of sort of having everybody have a way that they can look up and see what what is possible. But that's.

A

Very cool stuff, so.

L

Thanks for trying, yes,.

A

So what the voltage and is, basically, we have open, concrete blocks that actually talked about. How do how to stand this up? We had in cube con. We had a hands-on workshop and we are also planning for a meetup to basically and create people to just try this out and give us feedback on what features they would like to basically see especially around you know. You know multi-tenancy, making it fault-tolerant having zones define az's if at all, they would like to which probably the schedule is already doing, but yeah.

A

Those are some of the feedback that we're definitely looking forward and.

B

How can you forget about the view on the internet if they can't come to your leader, so.

A

I we have asked contrail, basically, which is an alias- that anybody can actually send questions to. They have these blogs that we have published. So you know comments are basically welcome and will actually you know get you no respond to the comments as well. As you know, getting unique as stiff at all this necessary perfect.

B

Thanks much Sanju, oh.

A

B

On the plus side, just to keep with timing, we actually don't have a cig update today, because the home for a che is still being discussed. So we will have that update in December. So let's jump right on to David or uncheck. Talking about the 1.1 released post-mortem, and thank you to those of you who answered my google form request for community feedback on what and how the 1.1 release went and what you'd like to see from the one point you planning so David hi.

E

All um David Rajic I run product management for GK hmm internet ease thanks.

B

For yourself, yeah.

E

I, just uh I I, don't know how many I met many of you, but you know can't can't hurt to introduce again um so the the top of lying thing is I'm, a huge believer in what's called Kaizen, which is continuous improvement and the idea around uh every release. Everything that goes out the door.

E

Is you empower the entire team to provide feedback and help explain ways that they see our opportunities to get better integral to that is the idea of these post, mortems and kind of an evaluation after you do a significant release, and you know I just want to go through what we saw so far. This is the first one and in the spirit of eating your own dog, food and and kinds in um you know you this process can absolutely change.

E

So you tell me what what works and what doesn't as far as reviewing this stuff and we'll figure out what um you know, how we can make it better. ah That said as part of the post-mortem Saracen out as well as you know, just talking with folks internally, we found kind of a bunch of things that both people liked and went. Well and things that potentially could go better in this particular case.

E

You know some of the things that we saw went really well were you know we released pretty close to on time as well as getting it out to door quickly. There was a little bit of a slip, but that's often to be expected. I, don't know how much better we could do there. um The team was incredibly quick to respond to changes. For example, we introduced the concept of the beta bar and things like that fairly late in the process, but folks we're ready to snap to it.

E

We, it didn't feel like we, you know let something out the door with a lot of compromises, which is a really big thing for us. We do want to be the the scheduler that is known for an Orchestrator that is known for being the best highest quality and things like that and not letting you know things sneak out the door when they're not well tested.

E

And finally, um you know I didn't know how many of you were able to go to q con, but there was a real sense of excitement around the stuff that we did release auto scaling the huge benefits in scale generally, um you know just a whole litany of things that we were able to get out the door that people are really excited about it, and you know I think that excitement is, is you know, going forward even further on the negative side there, but there wasn't a lot that people um mention, but probably one of the ones that that had a thread through most things was the idea of a little bit better forward-looking planning.

E

It still felt like 11 was a lot of things that came together a kind of piecemeal. You know we really want this and we're going to get it in, but it doesn't have kind of a unifying theme behind okay, we're going to make a big investment in scalability or auto scale or for ease of setup, or things like that, and so that is something that, in product we're going to take very very seriously myself and Sarah have already discussed a few times about.

E

You know presenting a kind of roadmap both for the 12 planning, which we expect to come in q1 of next year, as well as potentially even looking even further than that for 13 planning, which would be a second quarter of next year. So we are very serious about that and we're you know. This is not something supposed to come from: google or red hat or any particular organization. This is a community effort. We want to make sure everyone feels very positively about that. So we will be sharing.

E

Something in the not-too-distant future depends on whether or not the post-thanksgiving meetup happen or community meeting happens. uh It would be in the first one that we have post-thanksgiving. It's.

B

On the schedule for 1210, okay,.

E

Great so then the second thing is around publicity. uh You know again, we had a lot of great features and we did get good coverage, but I think we can do better.

E

Part of that is that the CNC f, who owns Cooper Nettie's, is still kind of finding its feet and we're going to work very closely with them to make sure that we highlight all the hard work that the folks are doing and that can be in things like community are in Cooper, Nettie's, proper or in the many add-ons, as we just saw with with open contrail. You know those kind of things and we'd love to highlight those and the company's contributing to that.

E

So we're gonna have a more formal plan around that it's not as under our control. Unfortunately, it's it really is a CNC f thing. We were very serious about that and so we're going to work with them as they find their feet to work with them very closely and then, finally, again, um this is kind of a process thing, but it is something that we'd like to tackle, which is really having a little bit better checklist for anyone looking to take on something as they bring it to production.

E

So if someone says well, you know I want to write a new load balance or a new schedule, or something like that. We should have a way to say. Okay, here are the steps you're going to go through to bring it to production? You know, by the time it reaches beta its to look like this by the time it needs to reach production. It needs to look like this. You should, you know, sign the CLA.

E

You should have your folks, you know, read the following documentation whatever it might be, so that if you want to contribute to us, it's extremely clear on how to do that and then once you meet that bar, you can be a contributing member of the community and and we'd love to highlight it. So again, that's something on us to really highlight and make it very clear that you know when you want to contribute, what's involved as cleanly as we can. So those are some of the things that we heard.

E

My email address is just my last name. You can see it there and zoom at Google, I'd love to hear more and and especially again with the idea of Kaizen and constant continuous improvement. Don't don't just wait for the next milestone to tell me how we can do better, please let us know now, but you will see some things addressing these, both positives and negatives. You know, within a matter of weeks, there wasn't for thanksgiving it. Be it sir. This next week is.

B

There any conversation that anyone wants to have currently around this, since we have david's time so.

F

um Just quickly, I think you know, through the 11 release process being on the outside, not being at Google things looked pretty opaque stuff, like all the discussion around alpha beta experimental. What have you that stuff was all Google, only um I'd love to see you know as we're as we're winding down to a release, regular updates in this meeting any big decisions that are going to affect the release be communicated widely. It felt like a lot of that stuff was happening in you know: Google conference rooms in Google, hallways yeah.

E

And we hear that loud and clear: oh man, thank you, make sure that got every.

H

E

Let me let me just speak to that for one second, the way that we're gonna, try and change. That is exactly like you described so there's kind of two roads that we need to improve their uh road. One, sorry I'm getting you're going to you're.

H

E

To hear these for second, so road, one is a better right map, so we're going to do that. Like I, said we're going to have a road map published of the entire communities goals around 12 at the next meet at or at the next community meeting. So these are the big things we're working on. You want to add something on your own great. Well, show you how to do that.

E

You want to complain about something that that we again as a community we're trying to reach out to as many folks as we can in prior to that changing priorities and things like that. That's absolutely the time to do it and we're going to do that sooner. The second half of the thing that we're trying to fix on the road map is the way that a lot of these decisions- you know not just decisions, but points of conversation need to happen and and Sarah will help with that and already has helped enormously with that.

E

But but things exactly like hey, if we're going to have an alpha bar, let's make sure that the alpha bar is public, and this is how it looks to be an alpha bar, and this is all going to be in public. So you don't like it do a pull request to a committee. You know comment on it and we'll tweak. Based on that. um You know again. This is Mayock open on us.

E

You know we. We did not have formal community things in place before we were lucky enough to get Sarah on the job we do now and she will keep us honest, I promise. Okay,.

B

And I'll add the third half after I've just been pushed under the bus here. The third half of this is also that one of the things I'm going to be tasked with is getting this community. This core developer core contributor community and if you mean contributor more than just code, but the people who are core to this group together physically twice during 2016 I've, stuck my stake in the sand, and we will be having those conversations and getting those dates out to you. As soon as we can.

B

One preferably would be in first quarter, so that we can look at 12, which will be well underway and 13, and then the second one, hopefully will be in third quarter, wherein we can be a little over finishing 13 and 14 and 15. If we're trying to do orderly releases on these I'm.

E

Sorry, let me be clear: I'm not pushing Sarah through the bus. This is all me. So if you want to head over come buddy that.

B

Was a joke I'm not worried about it?

B

B

Oops cool, thank you David does anybody else have any more commentary? We got a little bit more in a chat which is a lot of definitely plus ones to more transparency. We also got it. It was better than 10, so progress a little bit and the translated to github milestones is something that I'm also going to be working on with David, so we'll try and get that sorted as quickly as possible on so that there's more transparency in it.

B

Any other commentary other than Joe's meta point agenda for the contributor meetings um not yet give me. This is my third week at Google with this position, so give me another couple of weeks and I will throw something on most likely. The first community meeting of 2016 will get you guys dates as soon as possible.

B

All right. It else add.

J

One comment about the issues around communication: this is Carl Eisenberg from mrs. fear by the way- and it kind of reminded me of issues that we sometimes have on distributed teams. We have a very distributed team here and often times if we have hallway or meeting conversations locally in San Francisco, the Germany guys in New York guys don't get any of that information. So it takes a lot of slack conversation and then the more slack conversation is the harder it is to keep up with.

J

Someone usually ends up having to like summarize that some of the ways that that we do, that is that we have a slack channel, that's for sort of important information rather than discussion, and we have discussion in one channel and announcements in another channel that way people can keep up to date on the day to day without having to read all of the conversation. That's going on.

M

Damage could we use the dead of mailing list for that, or is there some additional benefit of using slack for that kind of summary stuff? So.

J

For like big decisions, I agree that the dev email is the way to go, there's sort of a lower barrier of entry to the the slack channel. So you can just have a conversation and then summarize this the conversation in the sentence or two without having to give a ton of context, and it allows people to sort of read a summary of what's going on. Historically, obviously, if that gets, you know has the same problem of too much conversation in it. Then it becomes useless.

J

So there's always a balance there, but it was just a suggestion and a way, a mindset of treating it as a distributed team rather than you know, just a Google team and a bunch of other external contributors.

B

Summary sound important: we can figure out what works best for people, maybe trying trying both and seeing what works. Seeing what sticks.

B

Psyche is definitely hard to keep up on when there's a great big long discussion all right anything else. I.

B

Didn't have I had a holder in case we had lots of spare time about khoob con trip report kind of thing. Does anybody want to talk a little bit about their experience, a coupon for the next two minutes and then we'll go off, or are we all going to get six minutes of our lives back.

F

You can echo, says six minute of our lives back.

M

That's a really amazing echo. It's like it's like 10 seconds, too late. It's not like the normal fo all right! Thank you. Everybody.