DASH Community Weekly, 23 Aug 2023

Previous Meeting

Next Meeting

⏯

youtube image

►

From YouTube: SONiC DASH Workgroup Community Meeting Aug 23 2023

Description

PR360 - SAI attributes for ACL rules should include list of IP prefixes for the Bmv2 compile (follow up)
1. The current P4 code has a mismatch between the BMv2 and the SAI generation. To SAI, it’s list of prefixes, But to BMv2, it’s just a single prefix. This reason of difference is due to the limitation of BMv2 that cannot support the list of prefix as the key.
2. We totally agree with using DASH_MATCH in dash_acl.p4 as default so that other vendors don’t need to declare DASH_MATCH in their pipeline.
3. So, here are two solutions to achieve our goal.
1. To set DASH_MATCH as default, but to remove this MACRO in BMv2 pipeline. This solution is simple enough but isn’t perfect because the gap between BMv2 and SAI still exists.
2. And another solution, (possibly better) comes from @lguohan . We should keep the same declaration of list prefixes in P4 code. Although BMv2 doesn’t support list prefixes, we can expand the list of prefixes in saiapi level and input the single prefix to BMv2 one by one. This solution will be a little bit complicated but more reasonable.

Guohan's SmartSwitch Linux Foundation Meeting
https://lists.sonicfoundation.dev/g/sonic-smartswitch

A

Okay, so I just want to say hi everyone thanks for coming to the dash Community call. This is our 81st meeting together and so in September. We'll have been meeting for about two years.

A

um So that's interesting, fun fact and so I I didn't get a chance to type up the notes last week, but I will um I'm kind of like a one-woman show sometimes and I get um I've been getting a lot a lot of Hardware work done, so uh the administriva has taken a little bit of a back seat, so I'm pretty Hands-On. You know in the lab and um and and with the different pieces of Hardware. So sometimes, if you see me fall down in one area, I'm excelling in a different area just kidding.

A

um So here are the notes from last week um and I'll send these out.

A

um So, let's see if we have anyone new this week, so I saw last week um hanif that you kind of came in later and so I meant to ping you, but I should probably go over that we um we talked about um whether we really needed those Ackle tags and could we remove the requirement and we kind of agreed that we probably don't need them, and so we put in a work item to go ahead and remove that um information from P4 as well. So now the technology providers kind of got together and decided.

A

You know, given the memory optimization or lack thereof, we we don't need them so just catch. You up on that and.

B

Thanks a lot Christina yeah thanks for the recap. Oh.

A

B

A

Problem then we had a guy from Microsoft. His name is Camille kudnick he's over in Europe and he um he's been kind of working on the back end of the repo doing some cleanup, and so he joined for about 20 minutes and um told us what he's been doing um just um again: I'm, not a developer or coder, but he's been doing some reconciliation to make little changes and big changes to make sure things work. Well.

A

So if you needed, you know, reviews or anything help them out, and uh so things like that, and so um let's see then I did Vincent I followed up and I found I believe I found the um the pr where petrosaur, also known as z, um had the pr that tied in with the the ones that you were talking about um and I can find it real quick here. Let me let me find it um I think I put the wrong comment. Maybe I fixed it in the uh no you pull this up here.

A

Did you see my comment in the I in the issue? um Vincent.

C

A

No I'll check it for you. It's okay.

A

A

Which one was it again? It was.

A

C

A

D

A

We go there, we go the ones at the bottom.

A

A

Okay, I have it. Let me check my email, real, quick or actually, while I'm doing that, did you guys have uh um PR or an issue. You wanted to bring up today or q a how about that.

A

A

Okay I found it I'm. Sorry I went back I found it. um Is it okay if I bring it up? Chris? Oh yeah, yeah, okay, all right so um Vincent uh Z said so it was. um He said he discussed the issue with uh his upper level management and with Marian um offline and just pull this over here.

A

Current P4 code has a mismatched to sign it's a list of pre. Can you see my screen Vincent yeah.

C

Yeah reading it yeah okay, cool.

A

Two solutions to achieve our goal.

C

Amazon to me the consensus is yeah. We should not have a compile flag, um but that requires BMV to work.

A

Okay, should I write that into the the issue here or I.

C

Guess you could you could copy and paste his email into the issue and I.

A

C

And I don't know where we go from there, but at least it's it should be tracked. You know.

A

C

Yeah, it's not urgent; it just should be tracked at all.

A

Okay, I will do that. I had forgotten you that this week, I don't know. If you remember at the front of the call, I was saying that I was doing some Hardware work this week and so I forgot to keep up on some administrative stuff. So I'll put that in there. So sorry about, but I did follow up so I feel like I should get points for that.

C

A

Okay, yay, okay, all right uh Chris go ahead; I'm, sorry.

E

um um I just wanted to follow up on uh full request: PR 4 22., the um Dash underlay enhancements that Thomas Made yeah. It's been waiting review. No one else has has.

A

Chosen yeah, it's been, it's been like two or three weeks now: yeah yeah.

E

So if we could just merge that, if.

F

There's no actually I I looked at it and it's uh fine, probably I forgot to put my approval, but.

E

I did a review that would be nice. Would you mind doing that? Just for the record one.

F

Second, the number is 422.

E

F

Oh thank you for looking.

A

At it, that's cool thanks, Marion.

A

Can anyone new on the call have a question about this one? This was the first swing at doing the uh underlay routing it was it between two ports, Chris.

E

A

Yeah, so you know um there's what was it the hairpinning Chris? We have to make a decision at some point. You know when we want to hear a pin, or you know, how are we going to do that and all that information is here inside of the pr it's actually pretty interesting um and and what we wanted to use to do that. So it's a pretty good one. Thank you. Chris.

E

Yeah well uh um ahmad's implemented that he was actually an intern and he's actually already completed his term. So uh we really try hard to get this in last week before yeah last day and uh it'll be nice. Thank you for um approving Marianne and review yeah great. That's all I had to say, okay.

A

All right, um so that's great, um did anyone else on the call have a um PR issue. They wanted to pull up or talk about.

A

Let's see what so I didn't see, any new ones come through this week. It is August though, so um that's a lot of people are kind of taking time and rest over the summer time. So these are the PRS Echo tags and gee. We know about all of these um oopsie.

F

um There is I think there's still a test failing, but I will need to rerun it uh supposed to pass from the logs. Maybe there is some other issue, uh so it's yeah um yeah, we've basically reverted manually.

A

Yeah and um and that's pretty much tied, okay and then merchat is out I. Think oh no he's here! Merch are you in or out.

G

uh I'm in office today, so I'm working I'm.

A

Here- okay, all right and so I remember you opened this one, and this one was more um definition. Based um did you want to talk about this one at all.

G

And one from Microsoft own information uh yeah well at the point, that's the line on what the definition to be clear: okay,.

A

um Maybe this is for me to take back internally then to the sdn or the v-net team. At.

G

This moment, I don't think we need to do anything because, as the devices come with the dash API and like queries, the stat via Dash, if I see two devices reporting, you know one, let's say: 10 million 120. Although it's same thing um then I'll report it as a bug at that time. So, okay, people should take a note of it now and when okay I'll see it happening already, 40 000 bug.

E

Somewhere that lays it down clearly isn't that the goal to have it to find.

A

Limited by a start and stop metered, okay, all right and I'll have Gerald take a look at this too um okay, and so, unless there's nothing else, um I see someone new on the call, um David, Harris and so I, don't think we've met before so is there anything I can help you with or um introduce you to the project. Nope.

D

I, just um there's tabs on things um working with reshma and then.

A

Oh so Intel, okay, okay, well I'm Chrisney at microsoft.com, k-r-I-s-n-e-y kind of like Disney Chrisney like Disney, and so um you can always reach out or we have this call. We have a behavioral model call um on Thursdays, so it's nice to meet you uh-huh yep and uh we have Rob Rob is my counterpart Cisco we've got okay. Well, we don't have anyone else, new um yaakov. Did you have anything you wanted to talk about today.

F

uh Nope thanks: okay, all.

A

Right and then uh we have a member from the Sonic team, so we've got riff riff. Is there anything you wanted to bring up today or share okay, okay, and so we do have I, don't know if y'all saw, but we do have um Gohan created the Smart Switch Linux Foundation meeting that I've been going to and it's it's every Thursday, but it bounces between morning sessions and um evening sessions, and is it today riff at uh 10 or tomorrow, at 10 I mean tomorrow's Thursday.

A

It might I think it's tomorrow at 10, A.M or Thursdays at 8, 30 p.m and I'm going to those I don't run that meeting but I I do attend just because I enjoy hardware and the smart switches, my project, so um you know.

H

If anyone tonight.

H

A

Yeah, so if anyone's interested in that um Gohan has that project out there or if you need the information I could probably find it for you so um yeah, that's all I have for today. Does anyone else have anything.

H

H

Hey uh just a question: um this is maybe directed toward Marion um on the service tunnel feature I started looking into that last week and um I was looking at the uh the the PSI API and the P4 code for it.

H

um So when service tunnel creates an IPv6 address from an ipv4 address, the hld said it does it by using the 96 upper bits of the IPv6 configured address and the 32 bits of the right packets address um in the P4 code and in this IPI there's another field which is like a mask to apply and I didn't see, I mean and I think. The intention of the Mask is to make that split between the two, not 32 96 fix, but variable um I didn't see any requirement for that in the hld.

H

I was wondering where that came from uh right.

F

There is no requirement, it's just I tried to make API a little bit more uh future proof like do. Do it more um more, in line with.

A

F

Yeah kind of like more in line with uh things like.

F

uh Basically, allowing to encode more data than uh then is defined by service tunnel, uh but you can assume that if it's a uh the service tunnel that is defined in the Sonic spec, uh there is a a fixed uh length that will be an effect split that will be done, but API it's a little bit. uh It allows for a bit more focusability.

H

Okay, all right thanks.

A

Awesome thanks and welcome back. Did you have a nice time off.

H

It was really good awesome. I had a nap almost every afternoon. Awesome.

A

I had one day off yesterday and it was awesome.

E

I had another thought we have issue 399 and um I.

D

E

If Andy had any comments on that.

E

I

E

There's a question to Andy: it's actually a curious thing, I finally tracked down what I think the problem is.

J

um So is this: is this usually, why there's an if def for calculating the packet lengths, that's different for dpdk versus B1 model,.

E

It's not Y is there it's that there seems to be a path that can't be taken in the code and the initial initial implementation had a long expression with conditionals in it which always got executed, but the way you broke it down to a series of statements resulted in one expression: that's never tested or it's one path. That's never yielded and exposed kind of a dead code path.

E

Only in that case, that's my that's my analysis, and so we can probably just refactor the code and get rid of it, but I really wanted to check with you.

J

Okay, sorry I hadn't uh I see you ping me on it, but I.

E

J

Probably got an email and it probably fell off the top of my inbox and and didn't get to look at it.

E

Yeah well, there was a day between Sunday and Monday I'm sure you would have got to it. Then.

J

This is issue 399 on the public, Dash replay, okay,.

E

I'll take a look at.

J

E

Okay, thank you just loose ends, little ones.

A

Okay, um anyone else for today.

I

Hi Kristen, this is yeah hi ameth yeah I mean there's a good good to merge the code. I think you forgot to merge the code, or is it still open for the reviews to look into it? The 422 PR.

E

But we we already talked about it earlier. Did you were you in the.

A

Discussion, you may have joined late I.

I

Think a little later I've joined.

A

Chris yeah, we um we did get a review and it looks like it's going to get shame. Oh okay, yay! Thank you, yeah! Thank you! Yeah. We really appreciate yeah. Thank you. Thank you. Yeah thanks, yep, okay, so um I do promise. I'll write notes this week. I'll do it and um unless anyone new on the call has q a or anything they want to ask I can give time back.

A

Can honey feel good.

B

I just had one quick question.

A

B

This this Sonic subgroup from Linux.

A

B

A

Switch: okay: what's.

B

The charter of that you know.

A

What's the charter Marion, do you have a quick summary while I look for the email or no.

F

I'm, sorry, what do you mean by that.

A

uh The charter of the Linux foundation, subgroup for.

F

Sure what is what do you mean by Charter? Let.

A

Me find a real quick here.

B

What do they, what do they plan to discuss and what what do they want to achieve? You know out of that particular you.

F

Know, oh yeah yeah, so the uh as you may uh know that software architecture for Smart Switch will be different from software architecture of dpu Sonic uh as a standalone right.

F

um So uh the war group is focused uh on solving all this issue. All these issues, starting from the from uh questions like how the Box management is supposed to work to the uh control, plane uh software protector to differences in high availability for a Smart Switch, uh which uh also uh is different from uh an sdn appliance. So that's the focus, although the feature set in terms of traffic scenarios is the same. The way that the software uh is defined for Smart Switch will be different. So that's what we're trying to solve.

B

Okay, so no thanks for um for the recap, so so Marianne! You know um some of this. We also discuss here as well. So probably it you know, some of it will interest the work that we are doing, but it's just that you know in Sonic uh um subgroup for Smart Switch. It will be more focused on the Sonic Dash point of view just for the Smart Switch and not for the appliance or perhaps the Standalone dpu correct.

F

Yes and here's the difference this, uh this community is focused on on dash and our goal is to uh just to have uh desk apis agreed so that everyone can do the implementation. There is this gray area of Sonic because we have Sonic design in here. There's a there will be something so the Sonic design here essentially provides you I think. The main goal uh here would be to Define and an application with the schema, uh but the goal in the Smart Switch Community is how do you properly consume that schema?

F

So there is this gray area, I, don't know how exactly it will be split because Sonic design is discussed partly here, partly there I don't know exactly and.

D

A

F

A

Features I think don't worry into the traffic patterns will be discussed here. Yeah.

F

Yeah, that's that yeah.

A

That's Dash and although some of the platform and resource and Telemetry and stuff started to be talked about here, maybe maybe he wanted to expand it into this to get more Minds behind it. I don't know, but can you see this honey Phil.

B

A

Yeah, here's the um there's only been two meetings so far and like last week they talked about how to assign an IP address using DHCP and then what happens when, like a card, is pulled or goes down or something and how to associate the mac and what API they're using to associate the Mac to the IP? If I'm, remembering correctly, what happened? um So you know that's as far as we got in the last meeting, um so just stuff like that.

A

B

Stuff, that's pretty helpful. No thank you. So much guys bring me up to speed on that. Yeah.

A

Yeah, that's no problem. It's pretty interesting um yeah anyway, thanks Marion, that was helpful.

A

um And so again, I I attend those um as I can I mean for the most part, I have no life, so I have nothing to do on Thursday nights, so yeah every other Thursday night. So um anyone else have anything for today.

A

Okay, then I'll go ahead and oh um Rob did you have anything Mr Murphy.

F

No I'm all good Christina.

A

Good, okay, all right, then I can give back one half hour, which is good. Let me check the chat, real, quick, just to make sure I'm um added a comment that uh okay got it: okay. Well, then, thanks everybody, um I'll, stop the recording and I'll go ahead and post meeting notes Maybe by Friday and I hope you have a good rest of the week.

A

Yeah I appreciate your time. Thank you. Okay, bye-bye.

H

Bye. Thank you. Bye.

A