Antrea Community, 26 Feb 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Antrea Community Meeting 02/26/2020

Description

Antrea Community Meeting, February 26th 2020

A

Of our low-income.

A

Ok, Jin Jin. Do we have a ticket for a issue for that? Not yet not yet, but I know rahu in the SU are working on it. Ok, I'm gonna go ahead and just create a placeholder and then have them update the details of it. Yeah.

B

All right perfect, so the the other item is Prometheus. I. Think that yo will we work with Kobe to make sure that we get it done for 0.50. In my understanding and correct me if I'm wrong, the patch is at the moment is to be still pretty much a work in progress, meaning that, from a functional perspective, we have the basic functionality implemented, but we need at least the unit tests and the patch as it is.

B

We implement a basic set of matrix, meaning that it's not going to be like extremely trivial, like you can only get how many agents and running, but it's not like it's not offering a great deal of details. Is that correct copy, Oreo and.

B

So this is a the target for that 0.50 release. Ups I believe that we will expand on these in the next releases by adding more tricks and I think that as a community, we should work together on defining the witch matrix we want to expose because it would be nice if we use pom prometheus, not only for color, guided gathering data about the status of the system.

B

Like you know how many agents are running, how many network policies are implemented which are at the end of the day, informations that are useful, but are also pretty much static in nature. It would be nice if we start to collect in real time serious time series data, like you, know, throughput throughput data party data flow processing information, so it would be nice to define which are the metrics that we believe are more important, define a way to calculate them so that we can.

B

We can expose them into parameters and another topic that I believe might be interesting, and it will be then to not only collect these metrics in Prometheus, but then also consume them in the octant UI plug-in I, don't know. What's your feedback, if you feel like this will be something good to ever, or maybe you don't want to have an interaction between octant and Prometheus.

C

Money marketing that is very simple, UI I'm, not sure how much is there that.

B

Is that is he did, but what would my question as well? This is something that we think that what sets the visibility, if we can like connect octan to it, something like a dashboards or stuff like that, so that we can visualize the data.

B

That is a good point is an open question about feasibility. But in terms of in principle, do you think that octant could be the right way to press the right place to present this data, or maybe we should keep a minimal amount of information in the octant UI plug-in, and then you know, then: defer defer, Prometheus data collection data visualization to some other tool which can be built using graphing dashboards or whatever I.

C

Say all things all fully cannot give you much visualization functionality, but I'm, not sure if you can impair any other I mean like the Sun oxygen. You have even refer to some other UI for the icing yourself can Pollock support for additional folks.

A

That's probably something I need to followup with Patrick on and ask him what the intent you know on octant is in regards to know. Is it only gonna, so static information or is it gonna show dynamic information.

C

If you want, maybe you can show some simple stats, but what I mean I think all you cannot have waste instance.

B

Definitely not the front end person so I'm totally. Not the right person to ask this, but Cody I was a I was thinking that maybe you can involve more so yo in this conversation, because she has some front-end experience, and this is something that might be interesting for you. Perhaps you can collaborate with Cody with the octan team to find if there is a way for doing some sort of advanced data, visualization I.

D

Think for the basic matrix we can show it in the table, but it involves with time series. Maybe we need some more complication.

A

And we also need to know where octane is querying, that from right. So if it's operating in a federated manner, is it having to reach across to get dynamic data? There's a lot of variables there that I think we definitely need see what we're doing from a TMC perspective as well. No.

B

I mean octant will be gathering data. The data source will be Prometheus, you know, octant will be consuming Prometheus. Basically at the moment for reasons, I think that octant is consuming data from some theories that we did--we.

B

The idea will be either add the parameters or consume only from parameters or the data, which is typically what happens in many dashboards that you see typically see the ones built with the graph Anna they connect to parameters and they get all the sort of information about your system and, for instance, you open shift dashboard does something very similar. It gets information about data, health, state information and dynamic data about your class stern, connecting to a primitive service that is running inside the cluster itself.

B

So basically they have they ever the overshift dashboard that data from Prometheus that collects data from every node. Let.

A

Me refine my question: do we do we intend Arkin to be a a one-to-one mapping per cluster or you looking at it as a summary across several clusters? I.

B

Don't know yet I think that it doesn't be one-to-one right in the way that we are designing it at the moment. Okay,.

C

B

I think that we don't have the ability of adding the multiple on three resistances to a given octant instantly.

A

Then that makes our options.

B

Anyway, so I think that it, that is all for Prometheus, we all had a few issues that were deferred the from from sorry that were deferred from zero point four. So let me look at the current at 0.5 payload and perhaps also let me share my screen so that.

C

Reasoning and I found their services with IPSec.

C

Especially the most serious one, I just do the cost. Yesterday, on likely its recruiting the adult photo one that I try to change the IP side tunnel so to use flow based tunneling for the sending pass. That means we we send the package to the D for the tunnel port using flow based econo for the not using the IPSec autonomous. We clear. Another switch for the stems. I found that unfortunately received the package were still go through the IPSec itano's, because we we create one tunnel pod for every remote node.

C

Even we don't send the packets using these tunnels. What do we use tunnel for easy packing seems OSU, including from the from the Lisa IPSec internal polls, I, think these banks will be chopped.

C

Unfortunately, it's not captured by C III now suspense because you can say I we just created the country are differentially, always an IPSec disable. The first. Then we enable on TR and redeploy sorry. We branch here starts since we don't reach him. The Cano type from the default one will stand to gie, it's a little complex for the water piping, that I'm seeing say: I only enable IPSec, actually, the the person a little Susan hello, zero tunnel, zero using wisten wisten tunnel tab, but I'm using a common question.

C

You GRE, so even we have this IPSec policy created so for GRE traffic, so you know say advanced after waiting lay by beside. Actually the the travel is not being encrypted IPSec for this new to service tunnel and zero loss in brick, but I believe if you create a entire castle from scratch and with IPSec being able and the traffic will be, this receive the travel be told it's it's like.

B

A serious yes I see, but let me see if I understood so. Basically, the issue is that it's it got not detected because in the CI after you enable IPSec, the traffic is actually still going on the GRE tunnel and not being encrypted.

C

Where Stanton.

B

Cause ready for.

C

34 were using worse than to be the tunnel.

B

Okay, be excellent. Sorry, we excellent tunnel, but it's not being encrypted and if you start from scratch with the IP sex, the turn the return traffic is not being received.

C

Its received an host is included from the the IP psychic on important not to the tunnel zero. Okay,.

B

C

After changing the other, for that one will chop the packets.

C

B

Like really broken, yes, it's.

C

Just one we used to do some other smaller issue.

E

If we, if we update the test, so that I mean, if you update the engine code so that we did the port and recreated on the top I just changed that solvation yep.

C

So what we can do sauce the fix for the micron Egon Schiele a quitter press release. We should wait for the other. Finally, what do you guys think it.

B

All depends, in my opinion, whether you think this can be fix it in two days or whether it can be fixed. It is needed at least a week to get physical fix it. If it can be fix it in two days, then yeah we can make a patch release. I guess in current today, I mean like a few days, including also addressing the CI tests so that they are properly covered. But if it takes more time and we will get closer anyway 20.50 release, then perhaps we might just create the issues who people are aware.

B

That IPSec is not really working as expected and then fix it straight in zero. Five, zero I, don't know called you. What do you reckon.

C

First, mass meeting, like poly, led like three or four days how today's weather, for not one week for the only.

A

The we've already announced support for IPSec, so I definitely think that since we've already announced the support, fixing it and and doing a patch release makes a lot of sense.

A

The other question I have is: what are you gonna have to delay to focus on that I? Don't I, don't think that we'll have a lot of customers using IPSec, probably before the next release is the only other issue, so two-way, so with that I would say, go ahead and and and prioritize it get it done. We can issue a patch once you once you get to the end of it.

A

If we think we're you know still far enough out from our next release, because it's we're going on a four week iteration here, it's gonna, be quite a while before we have another release. Okay,.

C

Let me work on the fix and we can come back again: okay,.

B

Good all right so say that I will proceed to the usual boring analysis of the current release, pelada to check on status on and what may be needs to be already pushed out. So yes, I want to annoy you to that. So I will start with the first of the 11 issues that we have run the 0.50 release. First, one is issue number 181, the cleanup operation and antonin any update here.

E

No dates, but you know I think I mean I said that last time, but I think it's doable for the 5.7 is but I didn't work on this in the last couple weeks. But.

B

I I, don't think that's a huge task. It's just a matter for your finding some time and only try.

E

Yeah, but it depends on on Windows support, also because it's likely to be different on the box and and make sense.

B

Okay, then we secondly shoots Prometheus and we already discussed it so I guess we can skip it as anyone. An update on the status of the ant CTL issue, which will be issue number 312, I.

E

Think this issue depends on a progress on NCT l. If there are some interesting comments that can be run out of clusters and we will consider publishing the binaries if not much progress is made, there's and I think we should just postpone.

E

B

B

The number 244, unfortunately I, don't remember, I, never remember this. One and I know that we discuss it every we discuss at every meeting and every meeting. Every idea may I forget what this one is, that this one is about it.

E

Is it's just a test which is like which was flaky unkind because we were because of our winsome test to test our clean up works for Andrea? Would you need the data plane so oftentimes? It creates issues with core DNS and connectivity, and and that's why and unkind it's more of an issue because of our I mean the pinic, the NIC of the container is attached to the bridge.

E

So it's a short update is that Shannon and I are looking into this into making this better some patch I merged for the release, which is like preserving the obvious flows on we started helping with it, but I don't think it's sufficient, but we have some ideas.

A

So before we go on any further as we're going through, this list can I mark which ones are active and which ones are not so that we at least have an idea which ones that were actually actively working. Oh.

B

Yeah, sorry, that's the lifecycle, Activa label. Yes,.

A

It CX yes, and some of them are labeled. Some of them are not so I just want to make sure that those are correct. So the first one we looked at deleting an tria was not is not currently being or is that active, Anton.

E

Well, it'll be active next week: okay,.

A

That's fine, there was something put, it is, don't mark it as active the the metrics one was that active or not I. Have it currently marked as active the Prometheus metrics.

B

A

Great great and cuddle binaries, nobody is working on releasing it for different platforms right now, currently so not active, and then this was active. Okay, let's continue on I, just wanna make sure I've got the right status. Oh nice.

A

B

Flicking the free connector policies can I go, can I go ahead, yeah.

A

B

Test it was opening the just a week ago and I, don't remember a lot of activity, but it seems that internal and wanna have been working it and then okay.

A

B

It stopped here.

B

In East, it's like an upstream ish or is that something that we need to fix in our.

E

So it's fixed upstream, but it's not available in a you, know, test image. Yet I literally can fix it like last week, so we can mark it as active where, basically it will definitely be done in done in time for the release, but we're just keeping this open, and this is actually fixed for CI. Otherwise, people may run into that issue and.

B

B

Sounds good, so, basically, we can close this one as soon as the fix is available in the image write, a test image or technically. This is already fix it. We can see because it's not even an ant ratio, but we can keep it open just for tracking. Then we have this one on documenting ant CTL and my question here is that do we ever to document at the moment.

A

Why don't we move this to backlog until we have that ready.

B

Do we have a backlog, yeah.

A

We have a priority backlog: okay,.

B

E

Yeah, we definitely have things to document because NCT is already working, but it's not a work in progress.

A

Let's, don't probably wear ties that until we actually have some features that we want to advertise on at CTL.

A

But we currently have what the table listing and and the state of the the OVS state.

E

You know that's part of the agent information command, yeah they'll, probably okay, okay, I'll.

A

Take a look at it and play with the and if I can do some documentation there, I'll I'll start on some of that, but I'm gonna put it back log right now.

B

B

Okay, so exposing the logs publicly for changing CI jobs, which now has become ever publicly available, Jenkins CI server, I, don't think that I didn't see any activity in the past two weeks.

B

Okay, good! So are we bringing up the server on AC wherewhere? Are we bringing it up yeah.

E

There's definitely a puff of concept, working I think we're trying to figure out some cost issues, but basically everything is ready. Okay,.

B

B

But if we, it might be nice for people like me that don't don't go to this lock channel every day when, ideally, maybe put some money here, I don't know if this could be useful just to keep track of. What's going on what what's going on, but anyway, for this specific issue, keeping a targeted, targeted, 0.50, whatever it's not really important, as it's not part of the deliverable I mean it's just good- to emit a VT CI available upstream as soon as possible, then the next one is update, OVS to 213 enhancer, docker image.

B

Of course we have to do it. It should not be a big deal, doing it for zero-five-zero and apart from a better obvious, are we getting any east enabling any new particular feature that we want to leverage? Oh, it's.

E

Fixing the next issue as a 348 container, mq, sighs, okay,.

B

Good, okay, so that's good to know which means.

E

Yeah I would have done this already I'm actively working on this. It's just that just in particular the OBS team, and he was in charge of publishing the really star balls so that three star ball didn't get published and I opened PR for the OBS website, so that the released our goal is published.

E

B

They find someone to do this, so basically, this also means that the next issue- 3 4 8- will be fix it by 404 automatically. So we get 2 issues for the price of one.

B

Perfect so the the past didi, the next one, one three six one we already discussed. This is a good first issue. We keep it targeted to because it's a good thing to alpha, but it will be probably a good activity for and you started to the project. So someone claims.

E

B

B

Okay and this one about this is not 417 about ESP Hydra sides should be removed from the MTU is not the IPSec issue that eunjin was already was discussing right now. Yes, it's a different one. Yeah.

C

B

And it is also a critical one, like some an issue that could greatly partially this.

C

C

Because you can manage him.

B

Perfect and I believe that issue four-two-one instead, is the one that Jenna was talking before about. Regarding the case, the same use case that we have in the CI is that.

C

B

C

Just created yeah.

B

I know this is three minutes ago. So that's right.

C

B

Will bring the payload for this release to twelve issues you may ever notice? Is that get up as a bug it automatically refresh the page? If you add an issue, but it doesn't refresh the issue count. We need to report this bug to get up. They'll give us a $1 as a bug bounties anyway. So now this completes the discussion of the items so far. We are looking good for all the issues.

B

I mean there is no issue that we can say should be referred already because we can make it so I would like to keep all of them targeted, keeping in mind that for the IPSec issues we should be for 17 for 21 and for 22, which is going to be created soon. We will probably consider releasing a partial list. Five perfect. So is there any other item that we would like to add to this release.

C

Hopefully, we'll have the first customer opuntia from industry.

C

Ideally wish you to start in for some strict upgrade and policy like a wing upgrade, we cannot bring any can tell ability that caused traffic outage happening.

C

So I switch I change looking into that, and it will fully define some policy. We shooter follows policy or challenge now when we make an en for the coach, Eng or the change another kind of polling, we should start with some basic upgrade has in CI to make sure we're not brittle.

C

Traffic across versions.

B

So sounds like a good idea: do you think it's reasonable for 0.50 to expect to add appropriate s into the CI, or we may need a little bit more time for that.

C

I'm not sure if someone have.

D

C

Hope we can is say we tasked for alpha.

C

B

Start creating an issue and target in Inked target, indict the targeting it and then we'll see if we can make it with. We can meet again in two weeks time and.

C

B

C

Have nothing C is something we can manually wrong before we really self I.

A

Do we have issue three three one as part of 0.5 on the windows support.

B

I, don't know because we are just starting the we are. We will create a feature branch we said for Windows, so alright not expecting that this is a task that it can be concluded in the three or four weeks. I was expecting just to be more than 0, 6 or 0-7 target. Unless we think that we can have something that we can deliver for zero-five-zero. But maybe we should listen from waving when you need to see if she says so, I.

A

Think this is more of a tracking of mechanics thing. So do we want do we have individual? We've got the individual paths that are being worked on as part of that do we know which tasks that are gonna be worked on during the period it won't be targeted for the milestone release, but at least I can.

B

Go ahead and say: I worked.

A

On during the milestone, I had.

B

I see a point honestly I, don't think I don't have enough visibility in what is what would the team is doing their to say exactly what they are going on dude they are working on, but we can surely ask.

A

Yeah, it's it's. My responsibility follow up on that, so I'll find out and see which those issues make.

B

B

A

We have because we originally wanted to see you know, get windows support as soon as we can into Q into Q I.

C

See winging where I had a few punches being regular I.

A

C

Feel the other five are it's a little risky? Maybe they don't say: seasonal failable release, windows, support, guys, yeah.

A

I think we just need to track. This is like a larger, like you said, a larger epic, so we can do that.

C

A

I'm, just gonna create a separate epic for that.

B

B

All right and I don't have personally anything else that I think it's worth adding to zero-five-zero honestly, considering at the timeframe of the release, there is already quite a lot of stuff in it and I don't want to make it even more loaded, I think it's a good payload and should be good.

B

So the next step should be to evaluate in the next few days where whether we are able to do a patch release for the IP sex issues or not and then, let's say longer, slightly longer term items we we started discussion on visualisation of Prometheus data in the octants, UI plug-in and well I. Guess that's! That's! That's the summary of today's meeting is there anything else that we would like to discuss for today.

C

I have one item now: yes, the twin Molly's windows support. Actually um we need a few patches made by Colby's okay for this, but it's not in our dream, repo. Yet so I want to this one. What's our strategy here, should we add push for being optional born before that we just used cloud-based service, build yeah.

B

The the the first question is: is cloud cloud base OVS build a fully open source or to the air for parts of it to that is not completely open source.

C

Amano-San is open source, so Apache 2 license yeah.

B

Understanding as well I just want to figure out if these patches that we need are entirely from the cloud base are entirely from the open source meat. In that case, I think it should be fine, Pikachu the cloud base lluvias in the windows image, at least for the short term ago, so it makes sense to ask for these patches to be applied web stream of yes, but that could have timelines that are not predictable for us that we cannot control in the recoil that the change is available upstream.

B

I am fine with the package in the cloud base of years being the Apache 2 is not no is not going to create any problem in terms of our adoption of the license like no, it would have been a problem. It was GPL, for instance, and some users might have preferred not to use it, but I party I see no problem.

C

A

How differ like, how much does it differ from the upstream in terms of I mean we had this one patch that we need? Is there other differences that we're gonna have to accept, or is it? Is it just a very specific patch you're, looking for over on a window side mm.

C

I, don't is really looking to water scrub is made for icing. They are mostly you know, for the optional west and based on those specimens. So.

A

B

A

B

Another question.

A

I had is is when it does go upstream. What do you anticipate like switch back to ending that.

C

For Windows so.

B

In my understanding, the interfaces are absolutely the same. I, don't even remember if they have some custom open flow extension, but even if they have, we are just not going to day of one. We are just not going to use it, so it should be okay, so they are fully compatible with the open flow specification that we use. Yes,.

C

The potential I'm not sure I cast know the potential used to be not an open flow. It's more on the way you quit the port on, let's put on Windows without crawl change. If you windows.

C

The way you quit always positive in the.

B

East implementation different, which I understand, but even the interface that you like you know.

C

What I mean that, with all this change yeah? How do you create an internal hold on open? We switch without with the culpers change poly you don't quit. I forgot the details. Probably we need.

C

The the port type are different, I see.

B

I see yes, so so these means that perhaps we might not be able to do then an upgrade with zero downtime. No.

C

Depends on if we believe the cabbage and the right way to create quote on Windows, if it's right away, we should oppose that to being out doing the West. Ok,.

C

B

Ok make sense why.

C

Personally, a favor I mean called biscaia I, didn't look into alpha I think they must have a good reason to make the change and they are actually maintaining the windows code. So probably we should just push their changing.

C

Behavior change, okay,.

B

Perfect cool and if you have some windows.

A

One other question for ginger non-issue 421: is that also part of the the IPSec fix so.

C

A

It's the entry agent does not update tunnel type I.

C

Say I think we should fix that together because we saw that we can. We could have 30 second CI. Okay got it.

A

Make sense what.

E

Was the decision for a cloud base we're gonna push for the changes to be moved upstream, but we.

A

Haven't we haven't made it a decision, but I think if the license fits, what I don't want to do is stagnate our capability to push forward on Windows support and if we think the risk is relatively low in terms of interface changes, then I'd suggest that we we consider going to that.

C

Yeah, the sinking that actually we already try to commit with color based guy so far the response is not very fast, so I assume so.

A

So there haven't been response. I said they haven't been responsive, submitting that upstream, to on their side right. That's why we're having to look at this patch.

C

A

C

A

Being able to submit that change upstream for them.

C

That's why I'm not sure I feel it's not worry friendly to this guy's way. They made a change. The.

C

A

What I don't know is like how extensive the patches is that something that we could, if they're, not being responsive push that upstream ourself I mean it's not my peak, so I I think that would be the best. The best plan right: let's try to get a response from them, but if they're not responsive and frankly, we preempt that just push that upstream yeah.

B

Okay, make sense.

B

Okay, so I think I believe that, with this we can conclude the discussion of Windows unless anyone else as anything else to add the regarding Windows 23:32 going one and ideal windows. So we are done with Windows and is there anything else that we would like to discuss? I've.

A

Got one thing yeah so for coop, con EU we're gonna have an opportunity to demonstrate an tria in the VMware booth, and so, if there is anybody that wants to contribute to what we should demo and what that demo should look like from a story, perspective start setting me that information and I need to you know, get that decided and put together. You know within the next week or so good.

B

Will do, and hopefully there will be a cube con. We see well.

A

I've got away I'm taking bet right now, so if you want in on the wager.

B

Anyway, so Cody I had a question for you. If there is any progress on work that you were doing with abhisheka and the other people from the communities community about the testing framework for policy.

A

Yes, so we've made several modifications to not only the what we're testing and how we're testing it, but the testing framework itself I would like to get a review done. What I've asked J is to come speak at our next community meeting. If we could, if we could schedule one for next week, if that's possible, to present his work so that we can talk about there's a couple of things. For example, he's got some scripts in in his branch and hack and some other stay.

A

So he wants to know like where to put some of the go code that he's working on for the tests you know, etc. If these things are eventually gonna end up in upstream, do we want to host them temporarily or do we want to just because they are gonna test an tria, or do we want to just try to go straight to upstream, so I need some input from from the team on like where to put those tests and also to take a review at the strategy of the test as well.

E

I'm gonna try to take group today.

A

Okay and again like we can bring him into the community meeting and have him kind of present. You know what he's been been working on. The key thing here is he's really trying to test all the iterations, all the boundary conditions of all the tests and then, and do it more efficiently without having to like restart pods and things, so the test should execute faster. Even though it's going to be running more tests.

B

All right, okay, thanks for the update on this front. So do you think that and then we should schedule, keep the meeting scheduled for next week, and so that weakens.

A

Yes, yeah, especially with cube con coming up, there's other things that are happening. So, let's, let's keep the meeting on for next week. Okay,.

B

Does everyone is everyone, okay, with having a meeting the next week as well.

C

B

Good, so any other topic that you would like to bring up for today, you have 25 seconds I guess.

E

Yeah, the proposal to talk about Andrea was accepted for reject, so, as you mean, UConn is maintained, rejects will be maintained and I'll do a five minute talk on Andrea the weekend before click on great.

B

Congratulations.

B

Perfect, so that would be so. The reject is the weekend before cube con right, yeah.

E

That's correct, not the same location but very close. You.

B

Also get to spend the weekend in Amsterdam on.

E

B

You have to change your booking as well. No.

E

No I waited for confirmation.

B

Okay, perfect all right.

B

Is there anything else that you like to bring up any last topic? We still have 11 minutes. If you want to talk about anything, we have time. Otherwise we can get back 10 minutes and 40 seconds of our lives and go on with our day. So it seems that we are perhaps at the end of the meeting and therefore I would like to wish everyone a good morning, good afternoon or good evening, and talk to you next Wednesday sounds good thanks.

A

B

You very much bye-bye.