Cloud Native Computing Foundation Service Mesh Interface, 24 Jun 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: CNCF Service Mesh Interface 2020-06-24

Description

CNCF Service Mesh Interface 2020-06-24

A

Excellent hello and welcome to the SMI community meeting today is Wednesday June, 24 2020 and welcome to everybody. That's been here, I see a lot of new names today, which is fantastic. My name is Lachlan Evensen I'll be moderating and bridget is with me taking notes. Bridget has so kindly put in the link to the agenda and notes in the chat so welcome everybody we'll get into it. We have an agenda.

A

We have 30 minutes, so we like to drive a very quick productive meeting, but if there's anything, that's coming up feel free to add it to the agenda as we go. If we don't get it today, don't don't get to it today we will endeavor to get to it next time. Okay, so right into it, stand up discussion, items website, blog post guidelines, Bridget yeah,.

B

I was gonna mention that, because we talked about it last week and I said, I would do it and I have not done it, but I also have not forgotten and I will actually do it. So, hopefully we'll get our blog post blog posts going. The reason I want to talk about it is because people might have felt discouraged about writing a blog post about the things they're doing an SMI, because we haven't published the one that we talked about from solo. That's entirely on me.

B

We will do that and please think about what blog posts you might want to put on the SMI website that.

A

Is fantastic? Thank you very much. So yes, if you are working on SMI and you have a great story to share, there is a premesis that we just added SMI back dot IO for blogs, which is a neutral place under the CNCs, where you can put your series about how you're solving problems with SMI or anything so, and we actually have an entry from solo about some work that they done Bridget's. Adding that. Thank you very much. Next up we have top-level spec update almost ready to go needs one more reviewer, Bridget's name is there.

A

B

Not really about me, it's more I, think Stephane and Thomas might have been talking about it and they are. They don't look to be on the call right now, but if there is somebody who is a spec reviewer who reads these notes later or whatever, yes, by the way, there's a question in the chat about, can there be a link for the blog? Yes, we'll add all of that information to these notes when I stop talking, which I am going to do now, I.

A

Quickly, just grab them grabbed it. There I actually had it up easy to find okay. So if you are on PR.

A

169, it looks like there were some changes that Stefan has proposed. Thomas is actually lgt and then we need another round. Michelle had lgt and them, but Michelle is out at the moment. So if anybody else is a reviewer on the specification or wants to be a reviewer on the specification, we don't mind if you don't have a binding, lg TM, please feel free to review it anyway and add your comments about this. So this is looks like changes to some top-level fields in traffic access and splits.

A

So I won't go over that. It looks like it's good to go, but if you want to take a look, please do so. Okay, we have a question from slack time frame for 0.40 release of the SMI SDK go Thomas I thought it was possible to do it without further delay. What does the fun think I? Don't know if we have stuff on here, quick inspection? No, and we do not have Thomas either. Does anybody have anything to say about getting out the latest release of the SDK?

A

Would that help them do they need to voice any concerns.

A

Nope I will I will go and pull the thread on that with Thomas and Stephane in in the CNC a slack channel for SMI, and if this is your first time as well, there is a SMI channel in the CNC up slack where all these discussions happen. So you're welcome to join that too.

A

Ok, next one is a question from slack: do traffic specs apply to all backends? Why a matches the traffic split in traffic split back-end and not in traffic's split spec? Why is the just type, the traffic's that loaded type 2 local object reference instead of string like it is for traffic target, we have Kalia their Kalia on the line. Did you want to speak to that Kavya.

C

Yeah I think regarding the first question, for why its associated with the back ends just looking at the example for the mo for that I guess it didn't make it obvious that the matches would apply per back end. I thought I would apply to both. So I was expecting the matches to be in the traffic splits back itself and not inside each back, end object and then, on top of that yeah I, just wondering why there's this type discrepancy?

C

A

Can you can you provide some links on exactly where the documentation is not concise, because I would like to raise an issue there or you can raise an issue?

A

B

You so much about Kalia for putting that detailed info in slack and if it isn't already under an issue, it would be awesome if it could be just so that that way we don't lose them slack. Yeah.

C

Okay, there's the traffic split like the API yep, and then we can compare it to.

C

Traffic target.

C

A

And was there documentation as well yeah.

C

And then is the documentation looking at multiple windows, okay, yeah.

A

Excellent yeah and I.

D

Know Andy's.

A

List excellent, thank you. So much I think you know. Does anybody have any did they want to comment on this now? I know, Stefan probably will want to comment so I think capturing this in an issue with the detail you provided and then in the SMI channel in the CNC f slack, let's go and ask Thomas and and Stefan and then I could probably do some work to make the documentation a little clearer.

A

So are you hoping for making the documentation clear and then or just some background as to why this has been done? I think Michelle would probably offer this up as well. Yeah.

C

Probably Oh Michelle was also kind well, I, don't think Michelle knew either okay but yeah looking for background and then perhaps updating the documentation. Okay find that the types are different. I think it just makes it harder, because then we have to reuse code, catspaw, probably or like I, would expect consistency. So I don't know if they're trying to update traffic target. Eventually there are lots of questions. So yes,.

A

Well, thank you for taking a look and keep bringing those questions, because we want to make sure they're answered, because, if you're coming across this I'm sure other people will find it too. So let's make it clear.

C

Issue in tag Stefan, yep.

A

Create an issue you can send it to me in the slack and then I'll kind of fan it out to whoever or you can just tag, Stefan, Michelle and Thomas, or anybody else who would like to pick. It up feel free to comment on that. Okay,.

C

A

I know: we've done some strange type changes across API versions which had caught us up in the past, but I think if the api's haven't changed, it may just be an oversight and we might need to rev them. So we don't have typing clashes, because that does cause some pain when you're implementing it. Okay extremely valid. Thank you so much for raising that Kavya.

A

Anybody else. Wanna weigh in here.

A

Okay, that is it for written agenda, I'll, open the floor up to any comments, questions or other discussion.

E

Do we have anything planned for the upcoming cube con mid-august there any I know it's obviously virtual, but.

E

There anything community, wise I, know.

A

There was I thought Michelle might be submitting a proposal, but I'm not sure so I'd ask her on that. I know in previous. We were we had plan to do a face-to-face, I, don't know if we pegged out because I think anything in sandbox needs to apply for a session. Now it's not yeah I think much.

D

Doesn't get a session right off the bat right.

A

You have to ask so I, would you know, put put an item here? I'll ask Michelle. If there was anything submitted, I did not submit anything. Do you have any other ideas on top of that Michael, because I know that you've come up with a bunch in the past.

E

Well would be good one hand to kind of like give a little bit of a stat. I was like where we are, and you know, what's the current thinking direction or whatever and potentially you know might be others out there who you know might want to start contribute or have something premature awareness and kind of like okay, showing you know where we are. What what we plan so I mean I said it's giving it's not physically, but it's a little bit hard you're.

E

Otherwise you could just say you know, let's have lunch together or whatever and hang out, but are.

B

They doing a project pavilions sort of thing at all, for virtual for, like projects.

B

A

I know Karen is looking at that to helm, so I'll find out about the virtual pavilion thing, which is some kind of premises where you can put things I. Think the other Avenue outside of that event is. We did do an SMI, CN CF webinar, probably about two months ago that Bridgette moderated with Thomas, myself and stephane, and then even at the camp cloud native yesterday, there was a higher side chat with Bridget and Lee Cal coat with the new stack, so I encourage you to also think we're trying to do those kind of outreach.

A

Things in-between cube con events as well. So if you have other events or you think we should do a webinar and you want to be part.

C

A

Webinar Michael we there are other outlets as well, but I'll find out about the project. Provision I can also ping the CN CF and say: is there any room to do an update because we could see if their schedule has room I'm, not sure? So? How about we ask that question too, because I think it's good whether we do a panel or we do an update or we do a fixed session. I think they all would be good. Just for awareness.

B

Yeah, we might have an answer on that by next time, just because they say might because I've been trying to even just schedule the helm session with the CN CF and they are maybe still a week or two out from actually telling me when my session is so. It's like. There's a lot being rearranged and.

A

B

A

B

So once they get a lot more of that settled, they might be able to answer questions like. Is there room for a sandbox project to talk about things? Okay,.

A

Thanks for bringing that up, Michael.

C

A

On hi Lee is there any any other questions? Hey.

F

This is Dominic from Cisco I have a question: hey dumb Ann go ahead, hey thanks and I was wondering about the scope of SMI and just for a little bit of background. We see a lot of interest lately in mighty mesh, mighty cluster or single mesh mighty cluster and I understand that that is currently not in scope of SMI, but is a preparation.

F

I was wondering if SMI may take some improv scope, some more self-awareness or clearer ability, so, for example, to be able to ask the kubernetes cluster. What service mesh instances are currently installed could be multiple instances on on one cluster as well. So what instances are currently installed? What services to these instances of service mesh know of and what workloads back these services so that we have more of a clear ability? The reason why I am asking is I was checking out. Solos service mesh hub and the service mesh hub actually includes code. That.

F

Parses, the the available objects right and does some quote unquote, like reverse engineering, what services are there and I figured SMI as the classy standard, or the only standard that tells us currently what the proper service mesh is? Maybe, and maybe the right place for that? Why.

B

I know solo is working out a blog post for and they asked my website about service mashup, so I don't know if we'll get more detailed there, but there will be a blog post about that soon. Laki yeah.

A

I think I think the scope of SMI can change. So you have that flexibility. I, think you know when when we initially opened it up, it was to create a common abstraction that provided the value of all service meshes so that you didn't need to be fixated on the implementation and create and and hence have an ecosystem of implementations. I would we've had conversations eat it eat it. Levine who's from the CEO of solo has been part of the conversations she had a conversation about potentially with me.

A

You know talking bringing those api's into SMI, so there hasn't been a formal proposal. I've also seen similar api's in link addy. They have a multi multi cluster set of api's, which can stat at the moment which clusters are running, Lincoln, D and then wire them together. I think the most pertinent things that I'm seeing- and this is just my opinion. Obviously this is a community we can. We can bring this up together. Multi cluster circuit, breaker and.

A

Rate limiting are the biggest glaring things that are missing from SMI, in my opinion, that almost also could almost all service measures implement in some form or another.

A

So if you have something Dominic that you want to propose, you mentioned solos, we can go back and talk to solo we can invite eat adhere to have that conversation I think she would be willing to see. You know if there is value actually having that spec go upstream, I! Think for the first you know first initial specs that landed in SMI.

A

There wasn't anything now we're seeing a lot more tooling in the ecosystem, and if we want to extract those api's from tools and have them in the spec I'd be willing to do that and I don't mind if it's at mesh. That brings us that from AWS I, don't mind, I'd be willing to consider those things, and we should be making sure that SMI covers the largest amount of use cases that we think are important for service mesh users.

A

So that's kind of the the high-level view as as for action items, I could go back and ping eat it. On the CNCs and say hey there's been questions about whether we could look at multi cluster interfaces and somebody suggested service solos service mesh hub one looks like a good place to start. Would you be interested in having that conversation of either donation or understanding it more, and then we can talk about that. So if you don't, if you don't mind.

F

For me to chime in real quick I, just sighs.

A

F

Think that the mighty mighty cluster mighty cluster mighty mesh I would think is already stepped to step, one. That is what I was Curran interested in. It is the more the self-awareness of SMI when it comes to only one mesh instance, so, for example, service mesh hub currently parses the deployments and looks for a deployment that has a container and is there the string link at D in that container or is still D in that container and then concludes that this is an instance of an e steel mesh or an instance of a link.

F

Edema SH and I was curious if we should extend SMI with objects that tell us that here is an instance of a linker D mesh on this cluster, and here is an instance of an East EOD mesh on this cluster, basically as a prerequisite or step one to multi clustering, but not jumping to multi cluster right away, because it opens up many many additional questions.

A

And can I can I tag onto the back of that? What would you do by knowing that or what? What is your intent to do with that information, then.

F

That comes then to step number two that enable step number two for multi mesh, because if you have, if you have uber control plane similar to service mesh up, that control, plane needs that information and currently service mesh up, reaches into the implementation of sto and reaches into the implementation of linker D. But I was thinking. If there is a. If there is a common abstraction on an SMI level, then it is of course easier to to reach into the spec. Then, instead of into the implementation I.

A

Mean this is this is a great amount of detail. Dominick I think you know what I would propose is. Would you be interested in in flagging a proposal issue so at least I? We can shop that around to the linker D folks, the Consul folks. You know we have AWS folks here they might be looking to solve the same thing. At least then we can have it to start a discussion and bring that to bear I, don't think it's like a high level. I see value in that I.

A

Don't see why SMI couldn't take that it's not my complete decision. Everybody needs to decide that but I think it's not it's not a bad idea and I'd be willing to create the process, or at least have people propose new api's that provide value because at the moment, we're iterating on the ones that are already there. But what we're losing is context of new awareness or new features that people actually need and service measures, and we want to make sure that they're represented in SMI. So would you be interested in jotting down?

A

Just you know narrative form like you've, given us what you want.

F

To have like this is important or.

A

F

Absolutely I have one I have one issue on the semi spec github repo, it's number 170 and a little bit back and forth with Thomas Thomas Chee, okay yeah, if, if you wanted to chime in, would be great I'd be interested if the the community as a whole is interested to make this service meshes more self-aware, so to say.

G

For my part, I think the this as a spec like having the ability to identify which type of mesh it is it that your that the that that you're interfacing with um can make a lot of sense the implementation of that discovery and that synchronization, that's probably you know, beyond the scope of a spec itself, like the implementation that you're speaking of dominic and searching for the container image name inside of service mesh hub is one way to go about that.

G

It is helpful to be I, think a bit more self aware being being self aware, then enables some of the other use cases you're, describing some of which are covered in some related projects that we've kind of spoken about on this call in the past about you know: homeo genius, Multi, cluster and then heterogeneous multi-class sure, as well as a on a related topic, since we're talking about surface mesh hub as much as we are I'll mention that well, there's a similar there's, a another management plane project called massery it.

G

The community is actively working on enhancing, what's referred to as mesh sink, which Dominic is precisely what you're. Referring to in terms of being continually in sync with the notion that there are one or any number of the same type or different type of service meshes in the environment, and so it might be a point of interests for you. I mean yeah well,.

A

I agree with what Lea said: you know getting this back up there and leaving the implementations to decide how they want to fulfill the contract point of that spec, but having a spec that allows interoperability or a toolchain ecosystem based tool chain to hook into that. You know much as we've seen with flagger hook into all the service message via SMI to do its rollouts if I'm glad that there are that there are some several instances of this converging out there.

A

So it might be time to have that conversation because I've seen a simple, similar thing from Linc addy, so yeah and I'm not sure if sto has something similar to do. You know mesh mesh discovery, but I think having people from all those communities actually come and say here is what the spec should look like. We probably have enough to go off already in what measure he's done and what solos done and what link ADIZ don't know what converge on something I.

A

Think at this point you lead implementation and then extract specification out of implementation, rather than deliberate on something that's never been used before we're at that point of it now. So, if it's valuable everyone, let's just extract where it's implemented and figure out what the spec needs to look like from that two minutes left anything else.

G

Chickens left there's something something from me, and that is that, just you know, we've I've mentioned: we've talked about the conform, SMI conformance tooling, for a long time. Actually, the project is finally in flight. The design specs of I think we mentioned, maybe a month or more ago, kind of the call for review of the specs I'm trying to find where we have the meeting minutes. Yeah at the bottom of the meeting. Minutes are right next to where Prejudice typing is a link to this back.

G

If you're interested in reviewing kind of the approach to being taken in that project and what am I trying to say that we're probably on on schedule for about um about a month out, maybe from.

G

Being able to have repeatable tooling that, where, where you can define assertions on what conformance of each of the four respects is and then unrepeatable kind of tooling, to run those tests Tess.

G

E

I called for feedback, so you're asking us.

G

To yeah there has been previously, but but there's your feedback is most welcome now as well. Are you just.

G

A

Great to hear that I will take a look, Thank, You, Lee yeah, all right. We are at time anything we made it through the agenda. Please add any items for the next call in two weeks time from today and feel free to add them, we'll carve out some space in the note stock to add them. Thank you very much to Bridget for taking notes. I would like to ask if anybody would like to moderate the next one before you all leave any hands thanks Maria thanks for joining.

E

It's usually if there are no other volunteers happy you.

B

Know I was coming for you Michael. If.

A

B

Me into it thanks.

A

Thank you. Thanks for joining.