Istio Networking Working Group, 25 Oct 2018

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Istio Networking WG meeting - 2018-10-25

Description

- CNI Extension demo
- Discuss Traffic Ownership / Controlling Policy Governance with Istio Authorization

A

All right, hi everybody welcome to the East your networking community meeting. We have quite a few items on the agenda today and my problem. My proposal was to allocate about like 20 minutes to each of them.

A

The first one is CNI extension demo and John will give us a demo, the reason also an Associated document, but I'm, not sure John. If you wanna like preview today or wait till next time. It's up to you. The second item will have to. We will have to make a community decision around the traffic ownership, controlling policy governance with its authorization so and the third okay. We actually have four items now discuss sidecar lifecycle manager, a management by Robert and also the east of CNI status.

A

So we basically have two things: one is around like CNI plugins and all the all the features we can deal with by using CNI plugins and the second one is around like virtual services and scoping, then ownership and policy control. Okay, all right! So John. Are you ready to start with your demo.

A

Donna, are you online.

A

All right, it may not be yet here so I wonder if we can now discuss the traffic ownership first then, and you are you, are you hi.

B

C

This is John Key here, okay,.

A

John, okay, so then we're going back in the original order. I'd argue.

D

A

You able to give us the demo yep okay, great! Thank you please, to present yeah.

C

Do you tried yellow slider? Can you see our plight.

A

C

Do that now? Yes, they go.

A

Is anybody else, a builders and okay? Now it's loading? Okay,.

C

So, let's go through it. That's good good diagram, not a slide. Number three significant, just a couple of slides to set the context, so Oh lost thanks to Tim and Robert because of extending their work. We're leveraging that pretty heavily so the whole one do is show how to sorry it's not a generic Network proxy using network namespaces and do some rewiring. They will think I had some interesting tweaks to how we keep the five boxes into Sto next slide. So what we've done?

C

The architecture is really, if you think of, on the left hand, side what you have today is envoy running as a sidecar, so what we have done is created a new network namespace and deployed into that Network namespace and then rewired. The network for region V eats the IP address, is still the same. Ip address that's allocated by kubernetes. There's no changes! All it is, is really a bump in the wire of traffic going from in and out of the pod. This could be a couple advantages.

C

It means that all traffic is guaranteed to go through the proxy. There's no need to do IP tables mucking around to try and figure out how to ensure traffic clusters with TCP UDP ipv4 ipv6 all go through the process. Namespace Negra. So.

D

It all goes through proxies namespace. Do you have to do anything in particular in the envoy network namespace, to get it to go through envoy, good.

C

Questions we haven't done anything on ball yet so we've just use our sample DNS proxy, which is busy a bump in the wire, and we don't do anything. Basically, we take packs in pack aside, I, don't think the things we've done to on voice I need to go and look at it. That's one of the action items we have because as far as envoy, seeing is just seeing an interface.

E

D

Listen to interfaces aren't going open, sockets yeah.

C

So it can open a socket. Quite it's just it's just a city in namespace, so packets will come in looking for connections. Envoy is listening. It will connect to them more.

F

Complicated here because because it's a packet go to a beep and envoi right now, it's listening to a particular people, but I think it is possible, with with some changes to have way to to and with net have been privileged to bind to a particular heap and cultures on traffic.

F

C

Okay, I think that's one of the big things to figure out how to make this generic and what has to be done because there's other proxies opossum envoy, ng acts and bunch of others who are I, think you know tiny few millions generic as possible. So it's not tied to one particular technology and we need to figure eyes the.

G

One is that we interface all areas so I feel good. Please, the be is bad one in the picture.

C

G

What you know that doesn't have a name yet.

C

No no no IP address, so the IP address is still on v-0 or so still a same IP addresses allocated by kubernetes. There's no changes to the networking which we thought was really important to keep as simple as possible.

F

H

Though right I mean, if you're doing this, then the original destination being requested, this only one of the cost, isn't it what I don't know if it is not done? Okay, because.

C

Also still work, it still still aspect. It says out to the bridge the bridge says: oh, if you want to get to this IP address. This is the Ethernet you go on. They just followed it, but.

H

That's not what I'm asking you I mean good lord, this lady soo regional destination, which the colonel actually preserves in place. If I would traffic was captured and read output to something else, then this was the action. There was the question by the collar and that's what we actually used with. Unlike mine, I.

F

Can confirm that it's preserved I mean I've done similar things with done tunnel and, and it is preserved.

I

The destination ID should be preserved. There.

F

Is no money question of the packet that mean is just a routing table that is circled.

H

Route, would it not destined ation this akkad option that we actually use? That's what I know is a strategy. Okay, you know the board might be the virtual listener. Yeah hold it. You.

D

Only need to call original desk because we have iptables rules rewriting the packet, and in this case there aren't IP tables for all screw room. So.

F

This is gonna, be replaced by this other option, which is binding. Each listen as directly to so, instead of having a single bound on 15,000 would have bound on each particular.

F

H

How do you restrict the mean today? We people actually uses include IP ranges in all together, an escape hatch if they have one too many service and freeze I mean one too many external services. They would like to let you know so on context, assets and every I mean all traffic will definitely enter envoy, and so you need I believe.

F

I believe you can still use IP tables to redirect stuff I mean you you, you still have control to say some packets, based on whatever will go directly to et tu V 82, serial interface. I cannot problem and and you're out some specific um you can you you can you? Can it's basically specify different routing table for four different pockets?

F

G

My question is this right: the IP address is associated with using a zero imprint. The namespace and my proxy will be dealing with lyric. How does it obviously able to make connection with you know, with pilot Oh.

J

John, do you have like working code for this yet yeah.

C

We have what we have a working with our be so sure we were sure demo, but just now the questions with envoy are very apropos and we need to say figure. Those out, but well just want to show us now is the fact that, with the extending you know Tim and Roberts work, we can actually do this fairly simply and we have a sample to the sample code.

C

It's basically just a packet, a map pump in the wire that lose packets from he means one east zero to each one and out to get to so just a simple demo to try and show that the mechanics of building is fairly simple. I think there's a lot of good questions here about how would actually work with envoy- and you know, TCP boxes, I, don't think, there's any major problems. You know Devils are also in the details. So let's go next slide and this Christmas really quickly. Then we talk more or I.

C

Think Roberts item is probably katashi lawless as well. So a lot of stuff here is we're doing the same things that Tim Tim Scott did.

C

Yeah we're only using on voice only in a network namespace, it's also in the existing pod namespace. So some discussion last week about how to get keys and mount keys so envoy would see the the same amount points in the parties every else. These questions for the Google people is. They are this question of CNI. You know for a cheeky e, but this not be touched on just throw it out. There's a discussion point.

A

C

We remove the need to mean to the IP tables, so if an app is not really well behaved, we don't really care because it's his own network namespace with control this isolation.

F

Was softly, prom technically technically on a PK misbehave application, Emily forget modifiers are out some kind of my person. Processes doesn't know, but does any IP table requires the same capability, so whatever you are doing that money placed IP tables that you fall into manipulating around you should do it with a privilege, yeah.

K

C

So you see a little more control. um Yeah it works with is. This is just some direct stuff from Tim. It works with NEC and I. For us, it's interesting cuz. The CMI gives us some interesting information like the bridge name and IP address so using CNI. What's really well compared to some of our previous approaches, we'll try to do yeah control of all traffic works. My peers, ipv4 ipv6, the proxy can I do so certain things with it can have a little more privileges because it's installed from someplace else.

C

It can have the ability to use packet, a map or SRV. Other things would be interesting. I thought with chaining boxes together, I think Koster mentioned the email, a bad idea because of late mistakes totally agree, just throwing it out. There's a you know has a thing to think about: yeah, okay, next lista kind of just how the flow works. The really additional step we've added from Tim in Tim's work.

C

It's really the proxy insertion, and this is where, as personal discussion we'd like to have so basically you deploy given set for the cni the same as Tim did.

C

They need apply some cross insertion mechanism which we'll talk about a little bit. What's a pause deployed, we it fires off the C&I, then it starts creating to insert first created, never name space, and then it runs the proxy in that namespace and then starts receiving traffic so fairly simple, fairly straightforward. We can make it I think complete transparent to the user apart from having either one or two deep insights running on the node. Next so example, just this we have, you know couple parts off each other before without anything and the Knicks.

C

Now we have, you know a couple of Dien spaces and we have our vns proxies. You know in between the traffic to visit, so we can show them so Sharad, you wanna start the demo.

C

So this is first thing we did. Is he took Tim Yama file and.

C

We use it pretty much, as is for certain the C&I.

C

So they see their issuer nodes, there be sure, yellow file.

C

So will probably recognize this from your previous discussion. This is just you know whatever, but what Tim was showing us before? So what we've done is if these drop into the go code for we've extended the go code, attended and added, essentially the ability to create a new network namespace.

C

So when Tim goes looks farther the issue, your proxy container, all we do is passes namespace into some code. We wrote before that does essentially the East reworking and then once that's done, we now have a network namespace and then we call the add sidecar, which basically then allows us to insert.

C

The VNS into that Network namespace, so it's a fairly simple extension of the original work from Tim. So you want to show the error pieces yeah.

L

L

So this is our demon set, that is running, which will go ahead and install the CNI and the corresponding the quanta list. Files into the appropriate, decreased, so I think air in here.

L

So we have mounted.

L

So this is the cni that's basically getting copied over here, which is nothing but an execution of the code which goes and moves that creates a new name. Space creates a new Veatch there and does all the plumbing for the like a proxy insertion, and then it will go ahead and insert the proxy also right now and then what we are doing as part of the DNI invocation in the conflict is for the demo.

L

So you're creating a bridge right now, you're calling it like C ni, 0 first and then we are calling our c ni, which is the HT or c ni which we'll go ahead and attach to this bridge field and.

G

L

To this bridge, so that's our CN icon files that it looks like so I've already I have the client and the server running.

L

Get out of here so I have two clients and server running, which will go ahead and if I go back to the slides of it, so which basically are all running in this sort of scenario, where a new name space has been created and they have a like a vnf which is our currently bump in the wire.

L

Things is already running in this new network namespace and what it will do is like when Z from the client. Once we try to get a file from the server, it will go through this vns that we are running all the way out to the bridge and in here and get the file from here.

L

So for this. So if you look at the original client and server.

L

It only has like one interface like, which is each zero, and similarly here also.

L

There's just eat zero, but let me go to into the new network namespace that we have created.

E

I, don't need to.

L

Yeah yeah, yes, so now, I'm in the new network, namespace that was created also now I'm, basically inside here, instead of en FRR bump in the wire, should be running and if I do an IP link show here. These are the two interfaces, each one and each two that we have created so these correspond to in the in the diagram. These corresponds to be eat one and we each 0, as we have created your social traffic. Your.

C

L

Pie, client, which is running- and this is the server we actually tried running raw.

L

And from here let me try to get the file so the basically, the query goes through our bump in the wire BNF through the new network namespace into the server, which is also running a bump in the wire prophecy. In the new net in the network, namespace and Christ came to file up back to the client, so.

C

As a demo, it's got hard to also show the stuff, because it's all just TTY, you know we will post the code next couple of days like you'll. Look at it. If you clean up a little bit I'll post on to get help, so people can see a look at it. You go back to the slide again.

J

So the v-0 and one should really be drawn inside the new network, namespace right yeah. Yes,.

L

Doing slides so V needle should be seen inside the new network names. Yes, yes, so.

C

Yeah there's some open issues we see one is you know how do we label and annotate and pause in network namespace? You know we can give it Pacific names like history of coffee or something you know there may be other ways of doing it. There are some works that Chelsie Hightower did with initiative and initializers the lies annotations to allow you to tag various things you want to do say you want to initialize, which is, and then how do you do the upgrade downgrade? Now it's running its own namespace.

C

Yes, you there's probably ways to do two separate upgrade and downgrade of the proxy from the application how that works or the life cycle not quite sure yet, but I think it's possible. I became a an interesting way to upgrade the proceeds infrastructure with upgrading the application infrastructure.

M

Did you actually work this with envoy in that container or just a bump in the word, I.

C

Just know I've just hit the bump in the wire, because you know just to get just to get up and running. Now we have up and running. We can do envoy. Maybe we'll do that for next next week.

C

Basically, just you know taking simple steps, because we've done the bump in the wire for something stuff. We did it the device plug in Jesus, using as a simple piece of code without having to learn on boy I, debug ability. You know it's a challenge, how we debug this in terms of its now we're in network separate network namespace. The good thing is, applications, don't see it, but the bad thing is application, don't see it and how we do debugging of it. The last thing it's not list is just insertion mechanism.

C

I talked a little bit this list in the paper last night. Just now we're doing it fairly simple kind of using the same approach. Robert used I would liked probably more to use admission controllers or operators syllable dynamic, and it can isolate the permissions from the users and keep them into the daemon set. So it separates out.

C

Who has certain permissions?

C

Ok, that was it, although longer than 20 minutes, which no.

A

Worries I think it was an excellent demo. So thank you very much and like this approach is definitely very promising. I guess that the next step would be to show how it works with employ. Instead of that, the vnf.

H

Funny I mean like it: maybe I missed this, letting it access to the pods I mean when you do this thing chaining. Can you actually access the pods secret mounts and stuff, so I you break up a little bit? No, so when you do this, namespace chaining, with like with bump in the wire stuff, can you access the board secrets and other secret mounts other things from the new network? Namespace? Yes,.

C

Because we've only treated network namespace, the rest of namespaces are the same, so the mount namespace is in Vienna for the same as the pod. Okay yeah. There.

A

Are different, UNIX or Linux actually namespaces the network and the filesystem, and there are a few others yeah.

C

All that is changing network namespace. Everything else remains remains the same with little reason: 13 act at the mouth like will.

H

I still be addressed his envoy using an IP delay because be there's also use case, an sto where we have HTTP proxy setup and as an application, I could just set up into the proxy and I'm and variable directly from to the local envoy and betters with the Vienna, which actually becomes completely transparent within the system. And so, unless there's be nice, if you can actually directly address the BNF if desired, I.

F

Think if you, if you're, if using sort of localhost, you use the idea that is so obvious, everything would be visible. So anything that is bound.

H

It does not have an IP, so it's just a bad like from worthless one source or other one side. Other side and the Vienna is literally a bump in the wire. It doesn't actually have like a full-on IP and a whole thing set up like I.

C

Think we I think we can also. We can give there the V's IP address of the part, because.

A

It's a namespace right, so you can assign IP address. Is there is.

F

There may be some problems with with admin portal and fifteen thousand, and fifty thousand one I mean we can probably manage that. It's yeah.

C

I mean it's a lot of details, I think to figure out. Oh yeah,.

J

I think it's probably I.

C

Think it's probably doable I mean this is a question. It really is. If we make it really easy, if it is too complicated, then it defeats the purpose, but if we make it really simple that it has has more value almost.

F

It'll be hidden from the user, I mean it's it's my question is: can we make it more secure I mean try to figure out. How much can we isolate an application from from from the site current, whatever secrets and things that it can do so in this design? I believe you are just using the network namespace and, and we can can run in a completely separated.

F

You know, identity, everything basically except a space yeah.

C

We you have, you have the option options of changing other namespaces as well, so we don't since so it's a question of if you want access to my name space to get secrets. You want to keep that the same. But if you want to hide things in a different name space, maybe it may be interesting or approach so.

F

For students I'm working on a proposal to kind of change a bit, so we don't use mountain secrets if possible: okay, yeah.

A

So it's definitely like I think we definitely need to continue the discussion, maybe in two weeks right with whatever other like progress will be made, maybe with endpoint. Maybe if you can so.

N

Like your proposal.

A

We put everything together and we'll be able to evaluate how well it is the requirements right like we still have to figure out the debug ability aspect like I guess that applies to any any of the CNI option, whether it's like with.

A

O

F

With with the other I mean you're going to submit a PR or no.

I

F

N

John, let's I was gonna. Ask about that so I think we can just I think there's about two other topics for the CNI on where you know it's under the sto ecosystem org and we want to potentially move it and we've got CI that were working on for that. So yeah I think he could put it in a branch for now, but it it looks like he's off of an older version, so John, let's get together and figure this out for how we came yeah.

C

I mean well I'll, get we've done the been illegal legal work for submission to a CEO. So what does he figure out? Push it to a github and then we'll figure imagine and figure out the the best ways to yeah get your get your latest code and sort of modulized a little more I think the thing we didn't change very much your code. We just did some.

N

Yeah so yeah! Well, we can not to let so, let's just.

F

See I think perspective. What exactly is I mean I. Suppose efficacy is not usable for this. So.

N

Yeah I have that as the topic on the meeting I don't know. If we want to people want to look at it, Tim.

A

I am very sorry but I think we'll be running out of time by the time we get to the fourth: that's.

N

A

Bit over so if that's ok, yeah, let's take it offline and definitely like discuss again in two weeks old, all the aspects. Alright, sorry I don't want to give a chance to the next topic. So.

N

Yeah, it's a male on all the points. I would just go. So if people just want to attack that male, then we can do it off on the email. That's fine great.

A

Great, thank you. Thank you very much John, and there was I.

P

Think somebody.

A

Else who did the demo but I missed the name so Quran? Thank you all right. So, let's move on to the second topic, which is discuss: traffic ownership, controlling policy governance with this authorization. So and you like requested, we make a community decision basically, because there were like multiple proposals.

A

There was the the biker if a traffic ownership proposal by Brian from Aspen mesh and also there was Lehman presented as our Park or controlling policy governance with its authorization and Arabic rules right. So we we need to make a decision on that and I think we may need to separate the various aspects of this problem. So I think we abbreviate charted with you and you had like somebody on like how we should separate the concerns here so that we they're.

I

They're kind of three major API design things in flight right now and they kind of have a natural priority. The first one is Network scoping right. If the ability to down scope the amount of network configuration that we deliver under operator controls, so we don't have N squared networking. Config problems is where we are today, and we have customers suffering with those issues.

I

Then the second one is this one right, the B traffic ownership issue and then the third one is whether we want to support delegation or not as an explicit API feature.

I

Clearly, not all three of those things are going to get resolved in the one time in time frame and given that we've implemented merging at least a limited form of merging I suggest we put the delegation thing.

I

It's just not gonna get reviewed or done in that time right, because these are pretty large features and that we don't even have a proposal yet for scoping, that's formal, but you have to have one, and so that is the priority for people who would be doing reviewed.

I

So, coming back to traffic ownership and governance, I haven't had the time to go and do follow-ups with the people who provided comments so I actually don't have a firm sense that which was a better solution right now.

I

Shriram. Do you.

H

Know I mean like all I see, is that since we added that simple merging I mean we have not seen that many, the Gateway issues, like you know, I also you, author and the authorship issues but yeah it would be nice to actually validate with the same users who actually complained or like Andy, authorship and Lester, to see if the simple merging actually made sense or if they need something more sophisticated, where you know could actually drive the traffic ownership, propose those kind of things or the other traffic ownership is odd. Orthogonal to.

I

Delegation or merger, oh the.

H

Delegation, sorry yeah, um yeah, delegate, I think we.

I

Know, delegation and merging, given that we have merging delegation, is the lowest priority. The three things to resolve yeah.

H

Yeah I agree yeah and.

F

What I think we need to seriously consider the implication of our buck? I mean: can we here's our buck? How much our buck in use for for this kind of stuff? Because if we can leverage our buck for this stuff is probably better for for for delegation the ocean.

I

That's by that you mean really, we have merging and are about control, yeah right, so it is merging in our back control sufficient to meet 90% of the use cases of what delegation coming. Do we.

E

Need to find out right.

I

So now we need the issue we're trying to talk about here is we have two.

B

I

Our Mac specifications for network namespace ownership declaration right.

A

So one way is what likely mean proposed and the other one is I think it was originally Brian who proposed, but now andrew is taking over that a proposal. So so.

I

I don't know andrew, has the document updated much since Brian welded, because that's the last time that I looked at it? No.

B

Not really- and you know, I was kind of looking at this I- think almost that if we decide what kind of interface we want out of this for users, then I think that there may be even parts of either one like I.

B

Don't think that this has to be a a or b decision right now, I think it's okay for us to decide how much do we want to involve existing automatic systems or how much we want to make our own authorization or req system that was kind of a level that I think would be satisfying. You know for this kind of discussion right.

I

So the only other Arabic system- that's currently in place, a slide from the is geo. Our back system, which is a coming traffic level, is the kubernetes network policy.

B

Right so my perspective on the first traffic ownership, one is that it relies on somebody else's our back system.

B

It could rise as an example it's or it delegates that to the environment that is I different than the traffic governance proposal, which sort of brings along its own, our back and so fundamentally, I think that that's sort of the biggest question is this: the thing where you just try to declare objects and let somebody else is our back management or do we say SEO needs to say why, like our back rules, apply for defining these policy objects? Time are.

F

You talking about East yard, bugs wanna use today, 440c of policy or order you're talking about a certain matter: buck no Easter Bunny. The.

P

Ones that you have today yeah, we.

I

Know is, there's kubernetes are about, can is your back? Yes, it's to our back is similar to kubernetes or about kept in style in some respects, but it has some differences. Yes, but it talks about a different domain. It talks about traffic instead of about resources. Now we're talking about using these two are back basic, syntax and properties to control, something which is declarative, which is network, namespace yeah, but still.

K

The same implementation same interesting concept, same.

I

K

So is this the same syntax.

I

The same logical structures and compositional model I mean.

H

There's not one of like which I bet model, but let me want actually chorim's delegation parts, I mean I'm back or not. Did the traffic ownership thing? Has this delegation model which is like from one quantity you can chain it out, the other one and so on which.

I

Is like the traffic ownership, stuff effects, service entry and, like all the other aspects of name, declaration Shriram? It's not just about virtual service composition. Wasn't.

H

This, the that's! Why the one beneath the delegation stuff it.

B

Was spawned by that, but they're, separate and I really think that we don't have to worry about the delegation thing I'm not trying to sneak the knows the camel in on that. That could be a separate, totally separate thing. We should decide what traffic ownership alright.

I

So this would constrain. You know the ability of a namespace owner to be a liquidy virtual service or just a service entry for a particular host name or declare a gateway for that hostname or an IP one.

F

Slight change: I means that I want to bring up is that we are discussing isolating namespaces, so so one can, with with the other purpose on the graph. We are discussing that a mode where resources define your namespace are only visible inside that namespace and overexcite outside, unless explicitly exported, which would slightly change the semantics here, because is any in a namespace. You could define whatever you want, but.

I

I still don't think we want to allow within a namespace that to do whatever you want. Well, you're allowed to do whatever the admin, let's redo them. Okay, yeah.

F

But without affecting.

G

F

This so is right.

G

F

Another thing you can do whatever you are allowed to do, including remaining host names at other make space is defined, but you are, you are still not having other people busy right.

I

So there is a question something this introduce something of an ordering question, because if we prioritize doing the namespace segregation without say doing the name, the network name control within the namespace. It still gives a certain amount of protection at the global level or across namespace boundaries, because one namespace owner can't interfere, interfere with the other or at least it starts to go down that direction.

I

Although there are still things they could do, particularly when binding to gateways that we would want to make sure will be covered and I, don't think will be entirely covered by that laughter.

F

It can be stupid carved well except the host thing after that. So it's not completely somebody to solve some of the Highland pain. Point icing, so.

A

This is what you named like the first priority network scoping, because it's related to the graph running we're trying.

I

To do anything right blocking people from having seriously large deployments, okay.

I

So yeah the discussions I've had with people we've only just come to the point where we started to talk tonight. They are back control over the networking names. I haven't had a chance to go back over the proposal, so I, don't really feel like I can provide useful feedback right now in this meeting.

I

It's not that I. Don't I want this to drag on.

D

And there's being, how is this decision actually going to be made right.

I

That's a great question: well,.

A

Yeah exactly so, for now it's definitely not going to be hierarchical right. It's not like somebody will come and say we do that or we don't do that it also it matters vikon. Whoever wants to work on it, for instance, if and once work on me no matter what right, then, we can maybe decide on what recommendation.

I

Affects the API right cannot be done under that hoc way.

P

I

A

P

Think that was affecting.

A

The API I think.

B

So also that we do have users who need this to move to the next stage, with their limits. Yeah I think that one option here figure out if it can be something that's additive and on the side like, for instance, when I looked at this last time, I almost feel like you could compile traffic claims down to the operator roles and operator role bindings if you had to so. If what it is is that we provide one to our users and then the migration path to the other.

B

If we decide that that's what we want to do and if this is a prototype or an extension or in our own namespace, that's ok. I just wanted to see how far I could get to some resolution here before going off and doing something. That's gonna impact everybody else. So.

I

Let's take a survey within within the attendees now right, I mean I guess the first thing is: does everybody understand the two proposals? Do people have a preference for what the proposals are representing if there is a reasonable migration path between the two proposals moving from one layer of sophistication to another? Is that documented.

F

And can bring because you thinkin, combining one proposal. I think will have way the choice and.

I

And you know the brutal hard thing is like do. We have to go and kind of do a/b testing with users to get like, if not actually implemented the go in front of them right right up at doc and say you can have a or B which one would you prefer, which we're next a useful exercise and I mean if somebody could write that up, I can get the park managers here to go, run that in front of some customers and get some feedback.

I

Well, let's do the first part of the survey, which is do people understand what the proposals are are? Do they have a preference for which the proposals, and can they statewide before a over B? We have to do that right now, right now, but some some things should have that I.

A

Can send a Google Form to East your networking may induce them.

K

Yeah, that's right.

A

So I think I brought a very good point here is that we don't have framework for making decisions. So it's not quite clear how we make decisions for.

I

You in the morning.

A

One thing I did.

I

Want to cover orthogonal to this is we should at least delineate what are the been working proposals and what are the prior, our ranked order of those proposals within the working group so that we at least understand the agenda things we're trying to resolve and they're rather the priority to each other, and then we try and make a commitment to say: look, we're gonna, try and get through this many. So these were communicating what we're trying to do.

F

What comment here I think both implementation, but both proposal will share kind of the same kind of code, some codes that has some so I'm saying that this is a tight surface and the actual implementation can be slightly decoupled. Then we can have a prototype that implements most of the code, and we can, you know kind of, do a be testing really be testing.

Q

Young but implantation may all be in the other Mission Control, oh yeah,.

Q

But for charlie owner proposal you basically defined you basically define a ownership, a yeah, I think and put other networking objects. You don't associate all the network objects also with a namespace yeah they're, not basically assigned the permission to a namespace. Add me, but.

F

In the end, your hands are, there is a destruction in both cases, he that has what a functions that they using some data structure to underscore years. Don't know, are you allowed to defy the DC source or not? So it's not. In the end, you probably construct the same data structure, use kind of the same code to make the decision to.

Q

Call mojo, maybe you use I, mean user experiences kind of the event. Why is the kind of.

J

Q

Is the networking object, I think.

F

Q

F

Know in your application, your designer application has the chorus. Then you try different you ice and kind of to see which one, and so is the UI and the user. Experience may change quite dramatically, but still have the same course, so that one practice is that if we can have a prototype or two or three to make the decision much easier than then alright.

A

And it's also like whether we can get any of these done right like, for instance, I mean I'm, not sure what are your like plans to, let's say the liberties or have a prototype, because if this is gonna happen in two years from now, then it's not going to help that customers who'd rather go and approach just to have it right.

F

H

Point I mean looking at that my memory, so two questions point this is mean. The crux of this thing is whether pilot will do in place validation and detection of configuration or whether we will do this in an admission controller and then within pilot. We just treat everything as like. You know this is blood nation, we won't touch it and so, as a first start, we could actually do the other, the external one.

H

Where, as an admission controller, we can define one of both approaches which defines like which configuration is gonna, be accepted and whatever comes in to pilot test, we just treat it as play. Okay, this is all already allowed configuration and we continue doing this stuff. So.

B

I I: don't think that where the code runs is the most important part of either of the proposals right.

H

B

More, what are you asking the user to do.

H

Believe it, then what I'm saying is that the question right now is in terms of implementation complexity and whether this will actually be something that we can revert or not revert and how we can conduct the experiment right. And so, from that point of view, before it's really I.

I

Mean I, don't think we need to conduct an experiment in code to make a decision. I, don't think that's necessary. I also, don't think it's appropriate right, it's much easier to go! Get user feedback. Well, mostly we're talking about! Is user experience here, like the I'm, not worried about the implementation detail at all.

H

Experience we can actually get a much bigger, actual live audience if we just have the same admission, control.

I

It's honest experiments just as easy to give them a dog and say this is a this is B, which one do you understand right.

H

Yeah but understanding the defender max will deal light. You CH ER demand a or would actually definitely seem like a ton of emails and amazingly, say hey. This is not working. What the heck is this and at that one either.

I

H

I

H

Yes, but at that point only good use of it didn't have two different admission controllers and as part of our installation, we say hey, why don't you try this out and tell us which one works best for your needs, and then we can correct feedback, usage numbers or whatever.

H

It is then, by the next release and say: hey this one actually really worked, not just from product managers to one or two customers where I mind you, your product model, actually gonna go to some top-notch executive he's, not gonna, go so, no, not sure, that's not how we do it right. We actually go talk.

I

To those people during the deployments.

F

Document, any things are won by side by side this one sync using it in a real environment, we have our real our backwoods, you have real stuff and I mean it's a very different experience to have the reason.

I

Why is it no matter? What happens? Is it going to be alpha.

A

There is some effort involved in like making the prototype work right and I.

H

Mean like what like Aspen matchbox and lemon can actually work do both and the worst case.

I

Scenarios we have customers using one and customers using the other, not the river bends. When you tell one of them, you've got a term one of those features down. Mm-Hmm.

M

I

That's a very expensive thing to do to a customer.

F

I

So we would be much better off not committing to either alright having.

A

I

In another city.

A

Can you actually start by having some sort of survey based just based on documentation, like just gather some preliminary data and discuss again like in two weeks from now right.

H

Cool like what it doesn't have to be part of the release, our is to release it. Just like the C and I come sort of as a separate ecosystem.

H

Things can come as an ecosystem project, which is a voluntary atom and people know the rest. They actually opt into it and say I'm gonna try this thing if it works well for me, I'm gonna vote for if it doesn't work well for me, I'm gonna dump this thing in.

F

Practice in producer or visa, we tell people, hey, you see ICD with whatever you want to achieve the same result. So, technically, probably doesn't sleep surround setup. What she thinks is kind of right.

I

It's clearly that the logical featured self is additive. There's no argument that this is can be layered on top right, but you could go and hear your news over today to do this right now, with the old admission controller, more or less it's, whether it's an intrinsic documented feature, that's built into this T or not, and then there's a separate thing about whether we want to or one unit at all,.

I

Right these two community and what people outside this unit yeah.

K

I

I think we all agree that some control over this is an intrinsic feature of the product. Yes, is there any disagreement about that part.

H

Right now or whether can be intrinsic after the public judgement, is the question right. No.

I

No just the features we are not the implementation, not.

H

A feature yes yeah.

F

Use your own CD system and screen share schedule right.

I

We still want to give them an intrinsic feature that lets people control, who can declare what networks.

A

Okay, so you know what I will do for now. I will just send this survey with the link to both proposals right. It would be great Andre if you could actually separate the delegation. You.

M

A

Document right so that people don't get confused over the delegation.

M

Stream address just.

A

The traffic ownership mark all.

B

Traffic ownership: there was a link to the delegation because that was sort of what inspired it. Okay,.

A

B

It's just an example: all right, don't we need to have a distracting example in there, so I'll take that out. So why.

I

Don't you Andrew and me men work on the survey together. We get the right kind of questions asked okay,.

Q

I

Maybe it's useful today you create a document for.

F

Dislocate decision I.

F

I

Is a survey in a doctor's office and like compare and contrast the two proposals right and can contain some additional context inside here.

A

I

Then there's the survey which is.

A

O

Best way to making fun decisions, so that will be like Jason.

I

Targets a future hardest.

Q

I

A

So we'll do that that will be I emails about that all right and I I'm. Sorry Roberts! You are like you only have like six minutes left or like on this week over the meeting like usual to discuss the sidecar lifecycle management.

I

There was a question in here and I think I think it's relevant, like those sorry Robert I, just don't think we're going to get to your subject. It all.

J

Was about how did how do we actually.

I

Make progress right we're still learning how to make progress, nominally right, there's consensus within the community and the leads understand what the proposals mean and are comfortable with them, and if the feature is important enough, somebody from the TOC might get involved and need to feel comfortable too so that they can represent that feature to the TOC. Now three RAM and I can do that. In this context,.

I

The the leads for networking right now are you when a 3-run and Christopher right yeah.

I

Do you understand the merits of both proposals, not.

A

Entirely to make a decision, okay, but but so that's the question: do we make a decision as a group do we make? There are multiple ways to make decision you either have like consensus? Is a group or you have a small getting in but making the decision right.

I

Like ideally, we would have consensus right, so that's what we should always strive for. We cannot arrive at consensus and a decision still has to be made right. That's what leads are for and what that's what the TOC is for. Okay, so hopefully that clears us up, but let's keep working for consensus, but requires a lot of work in everybody's part. Right decisions.

A

I

Quickly, when consensus is the goal, so the other thing that might need to happen is, if we're not getting to consensus and it's not because it's because people don't have the time right, then we should have an escalation which is looked. Consensus is not converging not because of massive dissent, but because of lack of community commitment, then we go to a more formal escalation process. The leads and the TFC just step it.

I

Does that sound reasonable to everybody, because the one thing like I've seen happening kubernetes is they always try to get to consensus, and there were many cases where the community collectively couldn't put the effort in, and so it would just drag forever. I mean that's the one thing we don't want to happen right, that's the worst possible! Okay, wait either way me want more. These two solutions that happen better.

H

To have something than nothing.

B

If there was strong consensus on doing exactly women's proposal, I think that we would just go ahead with that right now. We're worried about it. There's no consensus on what's gonna happen here, so any complexity means we'll just be revisiting this later. Okay, so let's listen. Let's.

I

Set a timeline: let's do the survey if there's no obvious consensus, we'll run an escalation.

I

Occasion, processes with the leads and the TOC, so maybe the next thing is to bring the subject to the tea to pull up in the TOC schedule. If we can't come.

A

Back to these consensus, like ideally, they shouldn't get to this year's heosu, because it's very like networking.

I

Careful because it aligns with security features, so it's not just networking all right, so I actually do want to make sure the security people are paying attention to this tip.

I

So there's a large the main of alignment, so uh just being honest, I think well, if consensus means the leads of audio mini night, so I think that covers it. If we can get to consistence.

J

I

To consensus, then it's the leads in the TOC. It might be to the TOC anyway, because it's an impactful feature, and it affects eight right now, we're in a situation where anything that affects the API is right.

I

We were very ad hoc in terms of coordinating across api's across the different domains and we're suffering a little bit because of that there was a discussion in the TOC two weeks ago about this, and we want to get to a more formal, API review process. So, even if we have agreed on what the domain and the structure of the API is, it would still get reviewed to make sure that it was syntactically, aligned or semantics a-line with other features of this time makes API is hard. The API is a nervous thing. We did.

I

Okay, let's get the survey done and let's get the a/v comparison done. Okay,.

A

I

Like I will try and get some user feedback.

A

Okay, so for next week, then we will continue with, like generally the CNI and the cyber lifecycle management like Robert or your proposal, which it ties in very well with everything related to CNI. It seems that there is a general consensus about like going forward with the CNI approach anyway right, and we will also like discuss the survey results and make a decision so hopefully done, and there will be actually one more third item on the agenda. I don't even know if we'll have time about service entry, namespacing, which I keep postponing. Yes.

I

A

Like fairly important to solve yeah.

I

So I think yes, so, as I mentioned at the notes, we said the prioritized list of things to resolve and if we have to burn an entire community meeting just to resolve one of those things, because it's an iceberg and we do that well,.

A

I

We have, we hope, the frequency of the community meetings to meet that need yeah.

A

I

Originally me every two weeks we could be.

A

P

A

Much I already lost.

P

A

Weeks for now,.

M

M