Kubernetes SIG Network, 26 Apr 2018

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Network Plumbing WG Meeting 2018-04-26

Description

Kubernetes Network Plumbing Working Group meeting 2018-04-26

A

All right recording is on this: is the network plumbing working group meeting for April, 26th 2018 and then see we have a fairly short agenda today there were some spec updates, so first item there is Doug Doug. Do you want to talk to you about your stuck items? Yeah.

B

B

Section 2 this back comment next to it, but it's like right under the comma, delimited and JSON sections, and this is just to formalize some type of convention for the names of the network, attachments and I I kind of ran into the issue where one I'm trying to support both the JSON format and the comma delimited format is I. Don't have like a good rule of thumb on how to basically say: hey. There was an error loading year, comma delimited stuff, so I'm like hey.

B

Maybe we should have a convention, so what I did is I based it basically on how kubernetes says how do I validate a namespace, because I needed to have a namespace in there and I'm? Like oh look at this, it just basically uses its DNS 1 1 2, 3 label format for those and I'm like hey, that's, not so bad, and then we can just say that it's basically that units of that eliminated by other things. So, for example, you know if you've got kind of like a namespace, slash attachment name.

B

You would just use both DNS 1 1 2 3 label on either side of that slash, I also included in there. The idea that so you would use units of that DNS one one. Two three label that and then you can put a slash in there and have one on either side. But then I also said that you could also use an axe symbol to pack metadata into the name, and that was to support Suresh. His idea of the interface name at at attachment name format.

B

So I said that you could allow that. So, basically, that's the gist of it and curious. If anyone has any insight, if it's too wordy and the Hat spec section or and I'm paying for any feedback there.

A

Yeah, the only thing I was thinking of was, do we have and I didn't think we did. But do we have a more of explanation of suresh's proposal.

A

B

Actually, on my list is to update the discussion points yeah yeah, it wasn't captured.

A

There is an open questions section where I know I talked about it. I wrote some stuff there. Basically I was just thinking, maybe until we get that section or until we have a good idea of how that multiple attachment with the app doesn't make sense to have that sentence to include the @ symbol. But if we think that or if you think that you have a proposal for cleaning up suresh's language and putting that in in the next like couple of weeks, then we could leave that I.

B

Just found down in the discussion so I'm, just gonna, cut and paste.

B

The symbol and then just say to be added: if this formalised, okay.

A

Sounds good yeah.

C

Doesn't this run into the same problem? The the key idea of suresh's was we're hacking, the interface name into the attachment object name, but the attachment object. It has to be legitimate object, name.

A

Without having thought about it too much I guess: I thought that the meta plugin was gonna, have to do a little parkour on the name and like split those up yeah.

D

A

You're, probably right because we're gonna pass that through as the CNI network name or something like that right.

B

That's gonna require even more ninjutsu to tear that apart. So yeah I'll leave that down in the discussion for now, because I think that it probably simplifies the case here. For the time being to say it's basically a namespace slash. Another namespace, like a theme, I think that that probably probably simplifies it and.

A

Also, there was a comment that somebody had added to that section: I, don't think that person is on right now that was asking. Let's see here was Levant well I'm, not sure to pronounce the last name, but it's cometh there and I was wondering if Mike you could reply to that person's comment with one of your use. Cases for multiple attachments to the same network help.

C

Me identify the.

A

Comment, it is in the multiple instances of the same network attachment name section. You should be able to get from the menu on the left-hand side, the table of contents. It's like the eighth or ninth item down man. There's the in the.

C

A

That's a great question, maybe in the view menu show document outline.

A

And you should just be able to click on multiple instances, the same attachment name and then just right below the example. There is a sense that this person had a comment I'm sorry anyway. If this person's comment was, is there any functional reason to it's much, the same pod, twice the same network and yes, we've discussed that a lot and I was just wondering if you could add one of your use cases to that comment, and then we could resolve it.

C

I might might be a best source for this, because my motivation for doing it was I'm just passing along a requirement. I got from other people, okay,.

A

Maybe it was that Suresh had I think.

C

It was people talking about an MV that we're saying they need that the most kind of originally right.

A

All right, then, perhaps an action name from this meaning is to dredge up those use cases yeah.

C

I think we're gonna need that if we want to actually get a change through C and I based motivated by this, we need to be clear on them. They actually use cases and I do still think. A change just seen is the best way to solve it. But if you really do believe what you have to solve, it.

A

Yep and I saw that if she come through CNI and commented on that already so bring that up again in the next maintenance meeting and try to push it forward a little bit and I think as I said in that comment on the CNI issue, I, don't think it's a huge problem for CNI itself, because there's only a couple of plugins that matter in the CNI reference set.

A

But it might be more of an issue for third party CNI plugins that currently only use that key for uniqueness, but as long as it's a spec bump right. So that makes more sense.

C

Bump should be in the minor or the dot well,.

A

We're contemplating or we will be doing, a zero for zero spec bump anyway for the get functionality which should be happening a couple of weeks. So I don't know if that proposal. The multiple attachment proposal will gel before then, but you know probably makes sense.

D

A

Do a major bump for that kind of thing, and some other stuff so but I mean either way just do the PR as you yeah you're like you should, and then you know once that gels a little bit more than somebody in the CNI maintains team suggests what the version numbering should actually be there very good. Thank you all right, so I'm adding an action item for a tracking down multiple attachment use cases Mike.

A

If you could do a little research on that one and see if you can dredge those up I'll, do the research for my side too. I know Doug and Tomo and Fang. If you have any thoughts on those if you've come across those yeah I mean I. Think.

E

Those are fairly common, if you, if you think about things like bonding or you know that kind of stuff, I think it's when you would need those yep I think right. So if you do two SRV, for example, I, don't.

C

Know if bonding is the kind of issue that would surface to a pod, though.

A

Yeah well, I'm, not sure it might I call use cases that were more around obtaining multiple IP addresses from the same network, even if it was over the same interface. You just have yeah keys.

C

Yeah multiply piece, don't require multiple attachments, technically no you're right correct, but we had another document right. We had it's not even fully use cases with with, like a couple dozen use cases, yeah.

A

I, don't know if any of those actually included the yeah.

C

We don't have research this online, that's.

A

C

Okay, I just see what we can make sense and.

A

Yeah, you know what we should probably have a link to that other document near the top of our agenda. I'll go back and find that document too and had a link to it so that we all have the use cases. There were.

C

Two of them, if I recall I, tried to digest some stuff, and so there was more lest I, just didn't more digested, so yeah we have I think at least two others. Oh sorry, no I was just the question of services, that's right! So there's just original and then there was a thing about services, so yeah one original compliation of use cases, and that was actually that was in another document right to just include the compilation of use cases yeah. Let's get a pointer to that fact. Yep.

A

Okay, I will go track that down all right, any other comments and Doug's valid attachments, name section for now. So what is the conclusion here? The conclusion is that I mean we obviously have to restrict network object name, there are object, names and kubernetes because of the actual cube objects and Cube validates those kinds of things.

A

Excuse me, and so we kind of have to validate those names in the annotations as well. So does that make sense, I think.

D

A

Know the the section at least I looked over the section, Doug added and now the only comment I had was the at symbol stuff, the rest of it looked fine to me so and again I mean, obviously no this document is set in stone. So if anybody has comments after the meeting okay, please.

E

Drop them in the doc. Okay, I just want to make sure that Doug has enough information too, with his implementation, I think so. Okay by.

C

The way I've stupid little procedural question, I, don't recall, being notified of change to that document. Did you know the fighters? Did you go looking for changes, mom.

A

I get emails, I think you can subscribe somehow to it. If you're logged in then you might be able to I am.

C

Logged in okay.

A

It's a great question.

A

Yeah honestly, I'm not sure I get comments in email, notifications; okay, if you click the the little speech bubble at the top right, Mike.

C

A

Should be a notifications button that shows up and there's oh.

C

A

That only you only get emails. When somebody replies to a comment you made or when you.

C

A

To a comment that they made and then they reply to you.

A

All right, surberg, that was the only outstanding staff you had.

B

I just wanted to make sure that looked okay with I'm happy to make tweaks to it, but I just wanted to make sure that was cool before I started, to do an implementation, so sounding good and awesome thanks guys.

A

All right so I think the only last bits we had there was some outstanding stuff about IP requests. Mac requests that we had talked about a couple of meetings ago. I had meant to go, find out what the deal was there and then you know think about that a little bit more and make sure that we could support that other than that. Are there any other thoughts. People have.

A

All right, well good, I, guess.

B

One thing that I guess I could bring up with somebody. Ed I had like a passing thought and somebody had replied to it. But the idea was: could we add a overrides section to the like detailed JSON annotation and basically the idea would be that the meta plugin checks that override section and then overrides what's ever stored in the CR D object that it references. So you could say tweak an existing.

B

Network attachment configuration as stored in the CR D so so, and somebody said that there was like a n Fe I mm-hmm consideration for I think it was the same guy who made that comment. You're referencing here.

A

Yeah I'm, looking at that thread right now,.

A

Yeah I mean it on the face of it. That sounds reasonable. My only thought there is that maybe we save that for like V 2 of the spec I'm.

B

Totally with you there, because it is a fair amount of added complexity, so I would say definitely be to material. Ok,.

A

Yeah I mean that my my thought there would be to add it to the previously discussed topics section and add a decision that the will not contain it. But v2 should be an open topic. All.

B

Right, sweet I'll go ahead and add that in there I'll kind of document the comments, so you don't get lost in the shuffle.

A

All right and then once you do that Doug. If you can resolve that comment in the doc, that would be great.

B

A

So I think next up we had a demo or a demo update. Excuse me so thing: if you're gonna go.

E

Yeah, so we group of people are working on implementing the the with multis and what we're thinking about you know implementing at least a subset of the features in the spec and as a phase one, and we will or we're hoping to do a demo sometime soon right now we're thinking the week of May 18th. That is it off week for us. So the first question is: is that okay is everybody okay with having such a meeting right to go over one? What has been implemented? How a you know!

E

The idea is to show how a reference implementation might work and also to look at you know where the future is right. What will what should we do next and what what's our thought going forward and also you know any feedback that we have on the implementation.

A

And just to be clear, when you say off week you mean that it would be a signet work meeting, not a plumbing working group, meeting right, correct, yeah, sorry.

C

You suggesting this is an agenda item for the Signet work meeting. No.

A

Maybe we should suggest it and then, if, if there we've run out of time, we could like do it the next one. The next meeting for the plumbing working group I mean I, think it wouldn't be out of character to have that kind of demo in Signet work, I.

C

E

So you think it would be reasonable to do that kind of demo in Signori meeting. Yes,.

B

E

Don't got you okay, so that's fine! The reason why we didn't do it. The next week is because that's open sack, summit, so I think a bunch of us will be out that week but yeah, but if you think it's reasonable to do it in you know also that would be May 17th. So that's the Signet work meeting yeah. We could do that. Yeah.

A

I mean it: if we can keep it to a reasonable amount of time, then it probably would fit into Signet work.

A

Okay, so it's worth a shot. Okay,.

E

I do think we might have to introduce a little more context if we do it in Signet work to to motivate a little bit but yeah. That's fine!.

B

Another thing, too, is I kind of like the idea of having a cig Network, and it's kind of good to have some more people have eyes on it. I think that's probably generally a good idea, and the thing is, we always come back to this meeting and have a deeper dive kind of session on it. Because really you know the like initial goals of this here is to you know, refine this specification. We've got that's you know really the number one thing that we want to do so, hopefully they'll contribute to that yeah.

A

All right sounds good. Next, up is ed with network services, but it doesn't look like ed is on, so perhaps we will skip that for this time.

A

And I guess just general notes on the spec. You know it sounds like most of.

A

Well, in general, it sounds like you know, most of the large comments and sections are sort of slowing down, so it seems like we've achieved consensus, at least on most of the points I think going forward. We should try to set a deadline on when we should come out with an actual formal proposal for the specification or essentially come out with v1 of the spec, so that doesn't drag on too much longer and I guess I would suggest that deadline like two months from now.

A

If that sounds reasonable and that I was thinking that the the language probably needs quite a bit of tightening up and we need to clarify it, it needs some. You know pretty intense, read-throughs and editing, because you know, there's probably a bunch of ambiguity defined. There's, probably some things they're a little bit too wordy.

A

I'll just need the way for now so yeah anyway, so that was the thought. I would thought that maybe that would take two months up around two months for like the back and forth. Just so that you know, when we clean something up, then people review it again to make sure that that cleanup makes sense and actually does what we want it to do. That sounds.

B

Good to me, too, and also it'll, give us a bit of time to kind of work on this reference implementation, so that some of the stuff that we've planned for is backed by some.

B

You know rubber, hitting the road to kind of test out, what's been specified, which feel like it's been getting tighter all the time, but yeah.

A

And that that was actually one other thought I had was that you know, as reference implementations happen, I'm sure we're going to turn up a few more things here and there so giving those a little bit of time to shake out some of the more minor points would be useful. Does that make sense? Mike? Yes,.

C

Thank you. Okay,.

A

Yeah I mean I'd, be awesome if we had something in a month but I'm trying to play a little conservative on that.

C

Let's go I guess the question that raises the excess, like you've teed up tonight, on a work item that we have I talked about, which is some significant revision.

A

Yeah I guess I, wouldn't call it significant revision of the included features or write.

C

A

Exactly yep rewriting heavy editing right.

C

So that's a work item, so that's correct.

A

C

Going to do it when.

A

I am volunteering to spend some time on it at the very least.

C

A

B

A

Great thanks and obviously anybody else who is interested in helping on that front either helping to write or helping to edit read make comments. Please do that sure I do.

D

A

Okay, awesome: all right are there any other.

A

Agenda items or outstanding issues or anything else that anybody wants to discuss.

E

Along those same lines, if anybody is interested in working on the reference implementation, I think that would be great too. So we are tracking all this in in the Trello board. So if people are interested, definitely I think that would be good for everybody to see at least what's going on. Okay, do you want to put a link to that Trello board in the agenda for dot.

B

E

I will do that I think right now: it's not public. Yet, okay I think you need to invite, but I out put a link. There I'll make sure it's a public, yeah and.

B

Also, some of the interesting bits do wind up as github issues on Maltese and there's a branch in there. That's labelled Network following working group, so you know, if that's how you roll that's another way to get some insight into it and patches. Welcome for sure.

A

All right sounds good and unless there's anything else, we will end early all right, see everybody either next week for Signet work or two weeks for this meeting again, all.

B

A

B