DASH Community Weekly, 7 Jun 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: SONiC DASH Workgroup Community Meeting June 7 2023

Description

Issue389 created Scale clarification
Issue382 for bug fix
Issue387 Meter Scale
PR289 still needs volunteer
Issue388 for OCP SAI change

Request Prince to present changes to DASH container to Community.

No online Community Call next week!

A

um uh Actually, it's uh separated out already I think for inbound and outbound. um The outbound route scale is 6.4 million.

A

um We were asking for V6 V for breakdown there earlier as well, um and um there are uh a few hundred thousand routes specified for inbound um in in icel as well. We don't know the difference between V4 and V6 Accel scale yeah. It will help to yeah.

B

It and I think that the last time we talked about that we were still analyzing and we weren't quite sure what was going to happen and when I talked to Michael zigmund on the S10 team, he had said uh we just can't really control what customers are going to do, V4 versus V6 and then the amount, and so his ideal was, of course, to make it configurable, um and so that was the last I had heard on that and okay. So he is going to be down at the Smart Knicks Summit.

B

So maybe we can um pick his brain a little bit there. When he's there sure.

A

That sounds good um and uh yeah, so these two things we can ask a little bit. Maybe you know something like the customer accounts. Would they be more V6 oriented the bay circles? The first two stages, where there be more b6b for that kind of general idea, would be good there yeah another thing that you know the number of uh flow scale entries today is: we have specified in the uh change in the scale document. We have reduced it from 64 million to 32 million right now.

A

We want to know whether, uh for 16 million are the ones that are active and the second 16 million are the ones that are passive, actually synced and not really uh forwarded by the current any given dpu card.

A

Half of them are active and half of them are yeah.

B

So I I don't have internet, so I can't get to my notes. But um do you remember um so Gerald had had us take down that.

C

B

Those were active, I thought go ahead. Sorry.

C

All of them are synced to all of them. They can be active, especially they can be active um because yeah, you have 16 active, 16 passive, but if one device goes down and one is to take over, it should be able to have all 32 million flow active and the hero test test for all 32 to be active.

C

So I think we should go with the assumptions on all 32 are active and make them work. That way, and in the real use case, probably it will be less but um yeah the device would work with all of them active.

A

Right definitely agree that you know when there is uh uh fail. Failure situation where there is the ha uh partner is down, uh then the active will have to you know, take over the passive entries also and do the forwarding there for sure yeah but uh yeah. So it looks like it's uh quite kind of clear that half of them are passive. Half of them are active, but we should go with the uh notion that at some point it will be all active.

D

Do we need some clarifying language in a spec somewhere.

E

C

It mentions, and it says that 32 with both it can open the Hero Test table. It has I, don't know six seven words depicting this idea.

C

That includes both active and possible ha or whatever it's a different term, I think user, but same thing.

A

Okay, we'll take a look at that and if there is anything we'll add a comment there, maybe.

D

I can I'll just display the file here, and we should be able to see it right. Okay,.

C

Yes, it's right in the table mentioned.

D

Yeah tell me in the contents: you know where it is in the context.

B

C

The middle of the page, I, don't know just.

B

C

You'll see a table, it's a one and only table uh too fat yeah there perfect so see it. So now it says: bidirectional reconnection tool capable of over subscription.

C

A

Right over subscription sounds good, um but yeah. For the most part, it will be half enough active and passive. Unless there is a yeah failure case, then definitely everything has to pay forward. Okay,.

D

Yeah sounds does sound like there's still a little bit of subtext that could be used to improve this, so otherwise we wouldn't have had this conversation yeah. If there's anything, anyone can think of improving a couple of sentences somewhere feel free to propose it yeah sure foreign.

C

Routes will be better also I.

A

Saw somewhere there was a breakdown of inbound and outbound, but I don't see it in this page for sure.

E

B

Other document that original um I can take a peek yeah.

D

I think that what I uh what we do in certain other working groups like The P4, runtime, a p48 guy. When we have this kind of discussion that comes out, we just make an issue right away and, just you know, say: hey clarify right, so I'm just going to follow that that uh lead and say.

B

You know, let's see, clarify.

D

F

It's specified in the Dash Sonic actually, but it will be good to copy to you.

C

D

Yeah and and that's in this.

D

In this one right here, right.

B

Documents General and then there's a Sonic Dash hld, a different document: yeah yeah.

D

We want to reconcile them right. Yes,.

D

And also routes, right, inbound versus outbound and active versus, or active connections, right, yeah.

A

Active connections- oops, okay, yeah inbound outbound routes and active connections, yeah.

F

Yeah the another thing which is missing in there is the scale of the meter policy table s.

D

F

This document there is no scale listed for the meter policy table, but initially there's a outbound inter policy table. Okay and there's no scale listed either here or I. Don't think I saw it initially either so.

D

Are we are we causing trouble for ourselves by having specs in two places? They have to constantly be reconciled.

B

Well, this was Prince's concern in the beginning for sure, but um and I guaranteed him that I would keep them reconciled. So follow the issue and I'll do it? Okay,.

D

Yeah I'm just raising the question yep.

B

Yeah no I hear what you're saying, though,.

G

B

Fact I could even I could even put a cross link into the hero test document too.

F

Policy table is not in either of the documents, so I think if you can add it, that would be great yeah.

A

F

Yeah the meter policy table that scale is not in any of the two documents.

A

E

A

Policy scale numbers.

E

D

B

Writing things down, even though I'm not connected yeah.

A

Thanks yeah, this will help a lot.

D

Okay, does this uh look like a decently worded issue that we can work with? Yes, we're trying to do it in real time?

D

B

So um I filed an issue last week as well number 382 for um a question that came in just FYI.

D

um Okay, so yeah Christine I saw you had filed 382 PM uh before BB2 pipeline source code question we've had a question from Todd was that was that right.

B

Working group- oh.

H

Okay, yeah I can I can speak to that. This is uh an old issue and this is a bug because uh features were added uh or the tables were ordered initially in one way, but then additional lookups were added, which now doesn't make sense. So there is a bug and uh I will fix that and I woke up. That needs to happen sooner. That's why we have this issue.

D

Okay, so this is just gonna, be a people or source code change and.

H

D

We need to change test cases. What's the blast radius of this.

H

uh Test cases, uh hopefully not I, intend to keep all the same names. It will be just the reordering of the tables.

D

I

I think it's very old issue. Isn't it yeah.

H

D

Okay, it's the same one brought up.

B

Yeah so I'll go ahead and make a note in there when I'm back online and let sad know and that kind of thing.

D

I'll I'll just um I'll, make a comment right now in there that says.

B

Okay, Marianne.

D

B

Thanks need to fix yeah yeah.

D

I'm I'm starting to emulate my friend, uh Andy Fingerhut, who, just in real time it's real good about updating issues, I've seen him in the P4 working group everything's very current, and that way you don't have all these extra stray note story, but I just goes for it right right, Andy,.

G

uh I I sometimes do that fluidly, sometimes less fluidly.

E

D

E

D

A

Another uh um request uh Christina as well right uh We've uh if it will be good to present the changes regarding the dash container in Sonic here, sometime from Princess side, uh the code changes or the features wise. Both you know a quick glance over all the changes would be good. A presentation here.

B

You want a presentation of that in the dash Community call or in the documents please in.

A

The dash Community call.

B

Okay, I understand.

E

B

Then um I know that Merchant in a couple of changes for formatting I, think it was line, breaks or something and um to get to get his formatting in here.

D

B

At 369 did 369 go through after um he made. The changes. Do I need to.

C

One is to go before the other. One is the CI to verify for it, and one is the fixes for all the files.

D

Well so which topics it has so Christina, you mentioned 369, which is update. The scale numbers that that was merged two weeks ago are.

B

You talking about I thought, Gohan turned it back, which I tweaked it.

D

Oh yeah yeah that was fixed, the formatting was fixed, okay and and subsequently because the main problem was Unix versus Windows line. Endings Merchant has now- and maybe you can speak to that he's- got two PR's one gone through and made a massive sweep through all the files to make sure they all have Linux line endings, and then he that's PR 379 on Deck and it's it's huge right. It's it's got.

D

136 files and uh I hope you wrote a script to do that and not manually uh and then there's a PR to automatically check that files have Linux line. That means before command like a pre-commit hook.

D

Should you want to talk about these a little bit and.

C

Yeah I mean you guys kind of talk, but basically my PR got rejected or fund done two weeks back, because I have Windows line index versus Linux line endings. So then I took an approach of okay. If this is up to the rule, let's make it for everybody.

C

um So I went through all the files and changed everywhere their mixed line, endings to Unix line, endings that one PR and updates about 200 files. uh Some of them updates the whole file because all of it had Windows learning some, it's just a line or two which have mixed um and then I created another Clause that are the CI and the pre-commit hook to verify. For this, an outlet to those kind of mixed line. Endings gets committed again.

C

So there is no code change. Nothing is just line. Endings.

D

Yeah and the issue is get will show a diff if someone checks in a file in their editor the line ending changes, because the editor is so kind as to silently do it for you, while you're editing, and then you push it to GitHub. Suddenly, the diff is gigantic. The whole file looks changed and it's really impossible to review. So does anyone have any objections to us standardizing on Linux line ends once and for all?

D

Any any objections to that?

D

Okay and there's is everyone: okay, with having a CI pre-commit hook. That will check that. So you can't check it in if it's wrong, were you? Okay, with.

F

That too yeah, that would be great okay,.

C

And it will automatically change the line endings for you to Linux and then you'll be able to commit. So you do a commit, it will fail and as part of the fail it will change it. And then you can click commit button again and it will be committed. The right way.

D

So this is not even this, this isn't even CI. Is it? Is it it's in your workspace.

C

It is in CI as well, because it will force it and you don't have the cookbook installed. Then the CI is going to run and it will tell you hey. It's.

D

C

D

But in your workspace, what what happens in your workspace, nothing, it doesn't and.

C

You know it looks basically pre-commit fully installed when you click commit button or you commit, it will run and it will fail after it fails. It updates the line endings automatically and if you commit one more time it will push it through, since it already changed it for you automatically. If you have the food, if you don't have the hook, it will just let it go through, but then the CI will catch it and fail it at the ca level.

D

I see so this uh pre-commit config.yaml, that's in the top level directory. That's that's equivalent to the pre-commit hook that GitHub will.

C

Know the python package and then do a PP install pre-commit, but it's a failure happens. You will see a message, an automated message printed into the.

C

Pr telling you how to install pre-commit tools and all this stuff.

D

Okay, we'll cross our fingers.

C

If you open that one yeah, um so one is a pre-commit hook and the other one is the workflow yeah. So if you click on the workflow, okay, I don't see the text. It's like. Oh I, see the text. Now it's a scroll down, scroll, scroll scroll. It will print you this message, but it will give you the steps.

C

If you don't have it and say install this, do that so it's little self document, if it catches you with listener and as part of this I have not enabled yet but I can enable multiple things like python flake and you pre-commit have support for many many plugins, but for now my purpose is to address. You know the things that I got burned for which was line engines.

D

All right, thank you.

D

um This will probably also any PRS that are in flight. This is gonna, possibly cause some. um You know merge requirements for people right, I,.

C

Mean I'm already six single, though so I don't know if it will create conflicts once it's, because it's modifying so many files. um This pre-commit, it's already used in Sony, so I kind of uh know. It's ripped off 80 of it from.

D

There so then, you less resistance right: okay, good, we'll, um we'll approve these.

E

D

um I'm I'm working on 372 uh updating, xcsc Snappy it I was hoping that uh Z gun would review it because he reported the original problem and I haven't gotten any review of this. So now it's behind other merges that we did in the last week and I'm trying to reconcile it. I'll get.

B

That thing thoroughly, it was Z that you were looking for yeah.

D

B

Yeah pterosaur.

D

The terrifying one yeah.

B

I will ping him internally. Thank.

D

B

Or see, if he's out of office or something okay, okay,.

D

That would be great.

E

B

Do that when I get back online, okay.

D

And okay, so, let's see what else there is there there's other PR's on Deck um Prince had fast path uh designs. It was reviewed that.

B

Yet yeah that one has an update at the bottom of it. I looked at it this morning before I went offline, can you click on it yeah?

B

No, not the metering. The other one I am that.

D

Was uh that was just clumsiness on my part or where are we.

B

I thought: maybe it needed a review or.

D

Account number yeah a review in these reviews.

D

Marion has been requested to review it yeah. So that's on Deck, ending reviews um and then BJ's been working on a fairly large PR for some time and he's responded to feedback and updated accordingly and so.

E

D

E

Problem was on Prince to approve what does it say at the bottom I'm trying to remember.

D

See, there's still some so.

J

We're waiting, yeah I, think that is one uh outstanding comment, which is the update to the agility, um so yeah I'll, sync up with prints on that one uh I think there are two places in the agility where uh the wording the verbiage is slightly uh different, so I think yeah. We just need to consolidate and make it uh consistent.

E

D

Okay and I had an afterthought. Also, do you intend to add any test cases new ones to test all these new features.

J

uh So currently I'm working on integrating Andrew's changes for the side, Challenger and then getting the skip test to pass. um Yeah and uh I think yeah right now, I don't have any plans to add new test cases.

D

Is that a concern of for the group- because this is this metering change- is pretty you know significant? Do we want to have any test cases that show that the metering works? It counters count Etc.

C

I think at least one test case.

B

Yeah, we'll probably need to do that. Do we just um attach an issue to the pr for follow-up later? Is that how that works?.

D

Well, that's one way to do it if we're eager to get this in if there's something like depending on it. Oh.

B

Okay, I can I can ask these guys. um I can ask Prince of PJ on Monday how we want to do that.

D

Okay, you know sometimes, if there's a.

B

Prince is on the call. I can't see very well yeah.

D

um If there's a PR like that's holding other things up, we want to sort of excite it. Sometimes we say well, let's get the pr in and we'll follow up the test cases. That's one possibility if, if there's nothing like urgently pending it, it's kind of nice to have at least a test case or two that confirmed it's working the way we expect you know if there's no there's no emergency and that's my preference, but I don't want to be sticking the mud either.

B

Okay, yeah I'll check on that um and.

D

B

D

Help in writing test cases, for this is, is the test case, writing from scratch kind of new to you, and you want some assistance and how to get going or if you want to collaborate on that.

J

Yeah so uh yeah I think uh uh uh I've added the v-net out test case uh before for some of the other features, uh but not the side. Thrift and the side Challenger test.

D

So, which test did you add before.

J

That was the uh no I think this was just the CEO driver program. I think it's called it out. Yeah.

D

Yeah yeah there were some some basic uh C plus plus side calls right, yeah, it's probably doing a test case with traffic. Well, you know with traffic and setting up everything.

J

Yeah uh yeah I might need some help with that, so I can follow up with that. I.

A

See uh I missed some of the discussion, but uh we from the dash PTF standpoint. um We have added some test cases for metering as well and we plan to Upstream them.

A

um If that will help- and you know um yeah, it will take a couple of weeks, but we can Upstream those test cases for metering as well. What.

D

What a restaurant thanks? What what do those? uh What features to those users, those using this new feature or are they using like standard Sonic features.

A

This is Dash PTF, only uh the overlay test cases that we have added all this while right, um new additions to that is the metering and that's related to the new metering that we have added in dash recently so yeah. It's.

D

Yeah, which metering this metering or a different metering that was added.

A

I, don't understand uh so you're talking about the metering, outbound metering that was recently added into Dash right.

J

Yeah I think probably the spear is not merged, so I guess Chris is probably uh asking. If did you like cherry pick this and then uh add the test.

A

um Yeah I think so yeah.

I

A

The latest metering, even though it is not merged yeah.

J

D

Yeah good for you, BJ.

A

I think so yeah I'll double check and get back to you. Yeah.

D

Okay, yeah: let's clarify this, that's why I was confused thanks for stepping in BJ's like how can you write tests for something we're still debating, um but.

A

I think a lot of this has already been discussed and there is already conclusion on that. Under your sharing screen.

D

Yeah Andy did you mean to take no.

E

A

G

Sorry that was an accident I, don't know how it happened. No.

A

Worries uh Voldemort: do you want to talk about it? We have some uh PTF test cases that we have added I. Believe Yuri is not here, but you know maybe voldemortnick can talk about it um yeah. Basically, uh we uh want to add the test cases for metering which is related to Dash, and today we have discussed um the outbound metering and our test cases are related to that and we can Upstream that at some point in a couple of weeks.

D

Okay, but maybe we can get clarification on that- just it'd be nice to know for certain that you know this. If this PR gets submitted without test cases, you know, then we know that there's another team working on the test cases and that way, BJ doesn't have to duplicate his effort, and we also feel confident that test cases are coming soon right right.

A

This is the only metering PR Vijay is this a new one. I can double check.

J

On that yeah, this is the only one.

A

There's only one right.

J

A

Already have reviewed this one yeah.

B

This PR has been covered a couple of times, yeah.

E

B

We went over, it ran out of time. We had to have a second we've covered it in a second meeting to finish it, along with some q a and then we had to brought it up for a couple of reviews. um Yes, I.

A

Didn't know this wasn't merged. Yes, as Christina mentioned, uh this is the same one. Then we are using this one.

D

I'm going to just make a note in here.

E

Yeah I'll find my friend feature.

D

Okay, yeah yeah.

D

Okay, so we can yeah.

E

Go ahead, go ahead.

D

I was gonna, say we can. We can uh get confirmation, then we'll, while BJ's working on merging in some changes from appeal Vision to get all the test cases unskipped.

A

Sorry, Chris I didn't understand your comment. uh Actually we will be merging sorry, yeah upstreaming, the um metering related Dash PTF, test cases from Intel.

A

D

Well, here's the confusion! If we, if Vijay, does nothing with this PR, how can you? How can you Upstream test cases against this new feature.

I

Just to clarify that yeah, the the PTF test cases we're talking about is is actually really went to this PR and um the as soon as this PR is merged. uh We can, you know, start reviewing those PTF test cases to Upstream.

D

Oh you've just confirmed that then, okay, so I'm going to change this comment because I want it to be making sense.

B

I'm so glad my phone is tethering, so I can be in this call.

H

A

Pending uh PR merge or something like that yeah, we are not sure that this beer needs to be matched and then you know are going to Upstream the test cases.

D

Okay, all right good.

E

D

It sounds to me like BJ, um you get a reprieve if you can get the pr um completed and- and we merge it in then someone else is. Writing some test cases. Hey. That's.

J

Nice yeah yeah, so.

A

Yeah Intel is writing the test cases and we will Upstream it.

D

B

Okay, so since we have the second, can we this old one, 289 Pier, 289 I, keep having it out there in email that it needs a volunteer? Is this still even an issue or a PR.

D

uh You're talking about the uh the one to do some dependency checking.

B

um Docker files, science, fair client, has a static tag in the from instruction. That's the title.

D

It's a nice to have nice someone when someone takes the time, maybe it'll get done.

B

Okay, so we still need a volunteer for that one: okay, I'll keep putting it up there then. Thank you. Yeah.

B

D

We've been from time to time we pick up some Doc, some make pile and Docker pile improvements, and- and this will probably get collected along the way at some point like Vincent made some recent changes that improve uh ability to build different ubuntu's um and actually some of the techniques he used in. There probably can be applied to this problem too, but take a look at it.

E

D

And we've got a number of old chestnuts in the bin here.

E

B

Yeah yeah I can um go through those too I've.

E

B

About going through those because a couple of my changes, I don't need anymore, so.

D

Yeah and also as part of um I think this was closed Andre if Andre's I'm, not sure Andre's on the call, but this one probably was closed by field vision.

D

J

E

Thank you. Let.

D

Github be the secretary right.

D

B

So did anyone else have anything since we've covered lots of PRS and issues today.

A

um Yeah nothing else from my side.

K

Great question uh um maybe I've talked about it last week. I don't know, maybe I missed it. um The update of the side to master you have an update.

B

And so that would probably be a prince or a Marian question. I wasn't here last week or do you guys have a recap of what you talked about.

I

Thank you so much.

E

K

Know Chris what I'm talking about.

D

Oh I'm, sorry so there's some conversations going on in the background. Yeah.

K

The size probably he's taken from a rich mental I believe and there's an effort to move it to pick up Master everybody.

A

Oh okay, okay got it um so yeah.

A

That's another thing that we are still working on with an Intel that you know uh there is a breakage in the dash PDF in the PSI uh ocp PSI, and uh once that is fixed, we will be able to move the pointer to uh the regular side, because we have a fix on top of um that uh Comet which is needed for dash, which related to vxlan and that's still uh we're still uh sort of working on fixing that uh in in ocp site and and then we'll be able to move the pointer there.

A

If that was your question, it said.

K

Yeah online, for that, because I know it's been complicated, so right.

A

um Yeah I think we'll have to prioritize that uh even uh more I I know that we have talked about it multiple times in multiple meetings for weeks now so yeah we we're looking into it. uh Hopefully it will be done in the next one or two weeks. Okay,.

E

B

Then just I guess I'll finish out: does anyone have anything else? I just have one last thing.

B

Okay, I waited four seconds, um so I thought about having the community call next week. You know in person live, you know doing it from the smartnext summit, but then Chris mentioned we're all gonna, be so busy um you're wondering it's the Keynotes number one and then we'll probably be busy doing other things. So I went ahead and canceled. It um I hope. That's! Okay with you guys unless you'd rather move the call to a different day.

A

That sounds good. Actually, it'll cancel next week to me, yeah, okay,.

B

Bye, everyone else and I see someone thumbs up. So that's okay. um Does that cool with everyone else.

G

Well, we get all our vendor bling on the trade show floor. If we we have to attend this. This meeting.

D

I, don't think this is a very bling heavy conference.

B

Yeah I might I, don't know, I, don't know the the coordinator.

G

That's it then Chris I'm not going then I'm. Sorry, I changed my mind.

B

The coordinators are some interesting cats. Let's put it that way.

B

D

That was a budding electro electric electrical engineer. I used to go to something called wescon, which was a gigantic trade show at the Cow Palace in San, Francisco I was full of trade booths and by the time, at the end of the show, I'd have a pocket full of LEDs toggle switches, you name it gave out all these parts.

B

D

B

Well, I was planning to do um two sessions one in the morning, one in the afternoon on Tuesday I'm just right now. The the draft name is Dash 101, just for not for us, but for new people who don't know where I could do an overview and if they um didn't come to the morning, they could come to the afternoon one.

B

That kind of thing and I have a few other things planned for Tuesday and then Wednesday are the Keynotes from various members of different companies in the industry and then Thursday I believe um I have two people from our software to find networking team, Michael, zigmund and you've talked to him before on this call and um aviji are going to come down and do that disaggregation paper that they presented at usnext and then Thursday I, don't have a ton going on so I'll just be wandering around so anyway.

B

That's all I have I'm gonna stop the recording, but um if you guys have anything else.

A

um Nothing else uh on Tuesday when you're doing the.