 So I just saw him is from a friend that Amazon is having some problems Some what they're having some problems apparently us central one is down That's not good When Amazon catches a cold the whole world it gets comes down with ammonia. Oh, yeah So I've put a link in the chat If folks could please go add themselves to the attendees list that would be great also feel free to edit the agenda We're just getting the agenda together. We typically start about five minutes after also Please be aware of the calls are recorded and posted to YouTube My Bluetooth device is having a little problem. So I'm gonna try a different one. I'll be back in a few moments Okay, and a new device. Can you hear me? Cool? Let's show we get started. Let's see is pulling the agenda up Is someone able to share the agenda as well while we're doing that We'll jump into the meeting. So welcome everyone to the next network service mesh meeting This is the last one before Kubecon so the Kubecon NSM meeting is canceled the one that is the one that was on Tuesday during the Dear over zoom we have Coming up the What so we we have So we have this particular meeting every every Tuesday at 8 a.m We have the CN the CNF telecom meeting Telecom user group meeting that we participate in which occurs On this screen's bouncing around Okay, that should be better Okay, which occurs every first Monday at 8 a.m. Pacific time and every third Monday at 4 a.m. Pacific time We have as well So we have NSM come NSM come coming up every or this this Monday, so I believe the session starts at 9 a.m. Yeah, I believe it runs 9 a.m. To 4 30 I think that there will be Typically, you know, you know, there's some sort of breakfasty thing that happens before the actual actual commencement So show up early munch on donuts and coffee and you know visit with folks Yeah, so did definitely make sure to Yeah, feel free to feel free to pop in earlier on is as I described we have we also we also have a Session on Thursday at 2 25 p.m. They usually shuffle the rooms around so we don't We don't we don't put the room on here until after they've properly announced it Yeah, that said please do the way they decide whether to shift the rooms around is based on how many people have indicated in sked That they're going to attend so if you intend to attend, please indicate that that will help us get a larger room That seems to have worked very well in North America We we got bumped into really large rooms and then in Europe We wound up with a crowd of people in the hallway who couldn't get it at all because there wasn't sufficient space so Yeah, we do tend to fill up the room. So Definitely definitely help us by adding yourself to the schedule So if you go to the cute pod website to the schedule section, there's a place where you can log in Another place you can do it is on their scheduler app which is ran by sked sehpd and You can sign up for an account and Follow follow which sessions you're going to you enjoy the most and please add an SNL to it Especially if you intend to to come see us we We also have coming up Dev conf With two talks that have been submitted the status on those is currently pending that one is in January 24th to 26th in Brno We I'm a bit surprised they haven't announced that that yet considering how close it is We have FOSDM 20 20 coming up with in Brussels The call for proposals ends in the first of December there is an SDN room which Which they reached out to ask for for talks We have to con on native calm Europe And just a reminder December 4th is when the call for a proposal closes when the CFP closes So Please make sure you know even if it's not an NSM talk getting Getting people in the community to to show to show up and give talks is still is still useful So December and December 4th is the last day and it is on a Wednesday Notifications will be in January and a schedule will be announced late January. We have Oh, and this has been renamed to oh any yes So let's we get to see how people will pronounce where pronounce that will be on S or ones or or Oness or whatever so else, but that will be in Los Angeles the Event URL is to be determined and Same with the CFP So a couple announcements I know we've announced this before but NSM calm is sold out. There is a waitlist available And again this meeting is canceled on Tuesday, November 19th This is Lucina on the call. Hi there. Yes Well, you have the floor. Thank you Great, so let's see 10 more Twitter followers Followed nine more accounts tweeted retweeted 22 things including the network service mesh con waiting list The Cisco keynote will feature network service mesh Five cool things you can do with network service mesh the Open networking summit Europe highlights reel included a little an interview with Frederick. So congrats on that and A reminder of today's to working group meetings and then scheduled posts for each one of the network service mesh con sessions So those have all gone out plan for this week Play one one or two more promotions of network service on the session at cube con Plan to attend network service mesh con on Monday and We've reached 300 stars for github so I can post about that as well So when available, I can share the contributors podcast The next release announcement or any news about the next release as you'd like and If there are any recaps on open source summit Europe excellent yeah, thanks for Thanks for mentioning about the highlight reel and Also for for the for the keynote. So I believe that keynote is gonna be done by by VJP andy Is that is that right Ed? Oh And the same con or at there is a keynote being done Assistive keynote yes by being done by VJP and a at cube con proper Yeah, and I believe that he's going to talk about in the same one that based upon the synopsis so I have not looked at the synopsis, but believe it. I honestly don't know what I'm allowed to say or not say at this point apologies Fair enough So so we will make that assumption or at least I will make that assumption Based on the synopsis cool, we have and with that We have no listed a no listed agenda that was added in so What do we want to talk about? We I know if we I know we Nikolai has been cutting some branches Nikolai. Are you on right now? Yes? Well, I mean the Manism the main netflix always mesh as a repo branch was good by Actually, but Yeah, I learned that okay, so we have the branch we Ed has experimented a little bit with Having a proper helm charts which seem to Work from what I can tell I know if there's going to be a separate announcement about that it but yeah, just wanted to mention this and I Have cut the examples branch Which is going to yeah follow the 0.2 release. This is the first branch for for examples so I hope that it will go good in a nice direction and I did not with this release as we said last time this is going to help Probably everyone on the first Project that wanted to use this was a CNF testbed. We're actually we'll probably just be able to download the Job charts and then just use the Images that that will be compounded against the proper Release from the examples So this is mostly the universal CNF image that is used You know So really quickly I've also started doing a bunch of cleanup of the docks that I'm hoping to get pushed shortly The state of our docks is kind of all over the place And we've been having a bunch of people from the community who have been basically pushing little updates here and there as they find issues which is really awesome, but The whole reason I pushed out the new helm repos is Because I was trying to write that documentation and we wanted something where people could try it without actually checking out our repo GitHub repo and that seemed the obvious way to go about it So that's a thing and then The other thing I wanted to suggest is I think we really before we push the release out We should probably switch from our default behavior being vagrant to our default behavior being kind Kind is enormously simpler and we've had a bunch of people who tried to go wander into vagrant and Gotten stuck in various ways because there are more pitfalls available And but I wanted to see what people thought about switching to kind just the defaults instead of vagrant for our providers But this from our experience, so I have been using kind for a long time and Kind is the default method to test the NSM with within the NSM within the examples Repo just being stable for a while We have a release. I think it's a 0.5 0.5 1 or something The release is there Yeah, so effectively if we point them to release version Maybe I know one should be should be better so that struck me is probably being a A better choice because I've just run into people who've had all kinds of complications on the vagrant front I mean just starting with all the many vagrant providers and vagrant is a kick-ass tool of what you need is a VM But if what you need is a Kubernetes cluster kind has just been kinder Let's see the point. That's that's kind of true I've kind of lost control of this But And actually actually I've experienced while using both vagrant and kind So one question is that while using kind kind is definitely much easier for To for deploying and try to do but one thing is that whenever I try to want to debug something because actually notice actually is Sealed in a docker and actually every time I need to do some inner operations I need to do I need to go into the docker in a container and do execute but sometimes I there might be a problem with the command the lack of commands and the lack of tools so that We're talking about the vagrant will tell you actually is virtual machine then Definitely vagrant is out way the way more like say our way in the advantages compared with kind so I think this like a trade-off so kind is much easier to deploy and vagrant has more power for Debugged I completely agree with you if you've got something you need to debug down to the node level You definitely want to fall back to something like vagrant And then I'm absolutely not suggesting we do away with the vagrant support or the vagrant documentation I'm just you know basically trying to point what my suggestion is though that the Sort of out of the box thing that we point people to be kind and then we can sort of list the other options So like and the stuff I've been putting together You know literally we've got an embarrassment of guides for getting this stuff together and literally it's like step one You need a kubernetes cluster here are your options you know We recommend you start with kind and the options that literally are because we've got documentation for kind vagrant GKE aka AWS packet, you know, we've got a crazy amount of documentation for how to get a kubernetes cluster going that you can run this on Sure, I complete I completely understand those people who who got stuck while using Vagrant because it can be a handy ask I saw especially for kind of users like me to work behind a proxy if things get much even even even worse Oh, I'm so sorry. Yeah, proxies are miserable Yep No, I totally get it. But no, that's a very good point Now we probably actually do want to take a bit of a note in the documentation as we're sort of listing that amount of Making recommendations that if you need to debug node level things vagrant is probably a good place to go Yeah, perhaps perhaps what we could do is Build up our documentation around that like a developer's frequently asked questions where how do I have trouble with what with my notes? How do I how do I debug that and then we can describe the describe a path Just to just to get people into that for for common things That sounds great So I think we still want to try and get the v.2 out this week Um But you folks will be patient for another day or so on things like doc updates. I think we'll get a much better result Does that sound okay to you Nikolai? Oh? Okay Yep, yep, I I count on you Nikolai to be the realistic one of the room Because clearly that is not me That's not true. Okay, okay, um Um, cool So are we done talking about the release stuff? Do other folks have things that like to bring up about the release? Where do we put the name because it gives such a nice name and never mention it. Oh That is a good idea. Good question. I don't know that may take a little bit of thought I Mean what one thing we can do with the new home repos is the way the current home repo is set up is Actually, I need to fix this in the notes because otherwise people will get the the older version The way the helm repo is being set up. Um, we essentially wind up with a um A per We wind up with a per branch or label Helm repo and if you go to the root one, then we have a redirect for all that will redirect you to the latest release And so right now The latest release is v zero dot one dot zero Which is old so old But we could also put redirects that would redirect from like borealis To the v zero dot two. That's a totally doable thing Okay, right in that way Yep, good. So a couple things we should probably look at is is there a helm repository that we can Submit a helm chart too so that people could just do like Helm install some some global repository name Uh, I do not I do not know the answer to this, but if there is a repo we should consider Uh upstream There is actually one we should submit upstream. I agree We should absolutely submit upstream and um So did we manage to lose by the way somewhere along the line? There was an item from the misery guys on the agenda that we may have lost somewhere I think Hey, thanks a lot. I'm sorry. So I actually created, uh, you know an entry Of a new agenda section, uh, you know based on a snapshot I created last week But you guys kind of started like you know a new section And I tried to put it at the end and I'm sorry like and I turned out to be the question answer section I'm I'm really sorry No, no This is this is all fine. It gets a little confusing when we have two meetings in the same day Um, yeah, and I I all I want to do is make sure that we didn't have it dropped silently off the agenda I think we got a couple more items that that have crept in um But yeah, I I I did want to make sure that we um I did want to make sure that we we we didn't lose you guys Awesome. Thank you. Cool. Uh When we're whenever you want me to jump in like let me know I'll jump in with whatever I have I actually wanted to show a functional demo, but unfortunately the demo is broken But I can at least like you know tell you like you know what we are up to and you know We would definitely you know want feedback from you guys because You know, I mean partly like I'm I'm still like learning nsm So you know, I'm not I'm not too sure like you know what I'm going to talk about is it's going to be That's Fantastic and do reach out if you need help debugging what may be broken with your demo as well after the meeting Awesome. Yeah, actually I have already reached out to the nsm dev channel and I think Alexander is already like, you know Probably looking at my logs And thanks a lot for looking at that Alexander if you're on the call Excellent do you You want to go sooner than later? That way you can you just you won't feel pressure for time and you may also have influence on what we want to add to the releases as well Absolutely, absolutely. I mean like you guys want me to do it right now like I can just take them on I want to find with that. Are you comfortable with that run of slob? I know you've got a agenda item before as well I was trying to find my own mute button. Yeah, uh, no, it's fine. No problem. Perfect Fantastic. Well, uh, what was your what was your name again? Uh, my name is girish Here you can find it under the measuring section there. Yes So the first one is my name there girish And we have you on the call as well Yeah, we've met uh, we've met li uh before and uh, good to meet you and so yeah, you have uh, you have the four Uh, stop sharing so that uh, either giri or leak and share Awesome. I'm gonna grab the ball for like a very quick second Shouldn't take me too much time. Um, I hope you guys can see my screen So here's my Kubernetes environment And you can see I mean, so I'm I'm just going to touch on the problem first. So the problem is that I try to deploy the nsm, uh, you know with uh with the master changes you guys have and you know, I think I just noticed it yesterday probably so but But you can see the admission webhook is is the one like you know, which isn't the scratch loop But but that's not the demo though. I mean like not the demo is that like you know So so we have this, um, uh, we have created an adapter for uh nsm So so this is how measuring is right now um, I'm already connected to a coronetis cluster from measuring um, and uh We have this way where we can actually connect to some adapters So and I have the nsm adapter that you know, we put together with harsheny Um here running on port 1004 We tried to make it nicer like you know with your logos here and there And then once you add an adapter like you know, there is a section which pops up in the navigation menu Like you know, which actually provides a place where you know, um, you can play with some stuff Or work with nsm from within measuring. So so the first thing is, um, you know, so I'm already connected to coronetis cluster, which I already spoke about I'm first going to try to remove the nsm. So you know, so We try to like, you know, uh help provision nsm and some sample applications very quickly So you can see, uh, I'm actually doing a watch Just to let you guys know So you can see there is nothing in my namespace now So I'm going to go back and then I'm going to install So what this does is it actually essentially I mean at the moment You know, you pretty much were touching on some of the points like you know, which you know We wanted to bring up which is uh Helm repo because what we're doing right now is we're actually doing a git clone and then trying to you know I use the uh helm content there to Generate the ammo and then deploy it on the cluster. So you can see I just click appreciate your persistence But that sounds painful Yeah, but but that's okay. You know, you know, you can you can see like, you know Harshini has actually done a great job Like, you know of automating this so you can see that it it only took like pretty much like, you know One click to actually get the whole thing working With some of course, like you know with some, you know, predefined defaults. We have set But yeah, so this is the problem now. Unfortunately. I actually also wanted to show a demo application Which we called Hello nsm app Of course, like you know, we are trying to bring in all your other sample applications, which you know We could find in your repo the icmp vpn vpn icmp, etc But of course, like and since that nsm is not working and I'm not able to show any of those And then we created this what's called the hello nsm application I'll tell you the reason why we created that is that like, you know, mesh We also has this section where we can actually do No kind of a load test, but unfortunately today, like, you know, we can only do a, you know, a http load test We recently integrated with wrk2. So it gives you the option of like a ford.io or wrk2 for load generation Um And and of course, you know, so we have some nicer integration with prometheus and grafana So so this this was actually the reason why we we thought it'll be nicer to create a sample application Which would talk http Uh, and then now unfortunately since that I cannot show the application I I try to put put a diagram together in the last five minutes Which you know, essentially tries to show it I I you know, I try to learn some stuff from pream like, you know, who has been very helpful You know, I'm patient with you know educating me with nsm So so the first thing we did was, you know, a very simple microservice Uh deployment, which is you know, I'm just calling it app a and b, you know, so Essentially incorporated terminology a deployment and a service definition for each of them Um, and then now, uh, one thing which pream told is that like, you know, when I mean appropriately like We create the annotations in place so that, you know, when they are deployed like, you know, they are getting deployed on the networks of the mesh Essentially they have an extra, you know, like a network interface like, you know, we just connected to the nsm network, you know Which is established by or created by nsm So earlier I was just using the kubernetes service to talk to app b from a Then I later learned that I mean, that's probably not right. So so what I'm doing right now is my app a when it brings up Uh, it tries to actually talk to kubernetes apis and figure out the, um, you know, uh, the, uh Uh, sorry, it tries to exact into the app b pod and then it tries to find the ip address Of the, you know, uh, ip address of the port, um, which is on the nsm network And then now app a uses that ip address to talk to app b So now when a history p request comes to app a, uh, what app a tries is it tries to use the not not the Kubernetes service name, but rather the nsm ip address of app b and tries to talk to it So this way we can actually do a load test. This is kind of like, you know, what what I have now So this is what we call the hello nsm app And and we have it functional, but unfortunately i'm not able to show it today. Uh, you know We'll get you fixed up right after the call Awesome awesome. Yeah, and and and you know what? I mean, I was very excited to actually hear you guys talk about You know the helm releases and you know getting with uh, you know the 0.2 release We're just fantastic because like, you know, I mean a better strategy for us would be to actually rely on your releases Rather than your daily bull because I mean, I know, I mean we have an open source project and we see It has issues once in a while like on the master branch. So, you know, it's totally understandable But then you know, it's a bit scary that you know, I'm trying to present this on uh, a bigger stage next week at nsm con so, you know, I Totally. So we will we will totally get you started. Um, like I said, the release should be coming out this week So you can finalize to pointing at the actual release Fantastic for your stuff and and that you know that that'll be great Yeah, and nsm epitomizes edge networking in more ways than one Why at the end sometimes But that but thank you very much for we're going through like this this trouble and for and for persisting So we're well by the way, one of the things that sort of came as a comment is You might have stale docker images. Um, it's someone is saying that that are causing your problem. Denise is actively piping up on the chat on the meeting. I'll add into the thread you got Gotcha. I mean, you know what, you know, that's a fantastic suggestion I I'll I'll try to clear the images and then I'll try to give it a try or maybe just change the policy You know the image pool policy to you know, all the way it's full So that way, like, you know, it'll be always consistent. That's a great suggestion. I'll give it a trial again right after the school Yeah, so Denise put a couple commands. So make sure you copy those before Before the writing ends, I've also attached Denise the slack thread so that we can get that capture there as well So should be all right awesome awesome Because the the issue that you're seeing there with the webhook that I think the last time we saw that it was someone had stale Docker containers in their In their their Your Kubernetes cluster local repo for some reason doesn't mean that's the situation here But we've seen that error before and that that was the situation that Gotcha. Actually, I mean like now that's very possible again because I I did not clear the image and yes It just occurs to me now that I didn't do it Uh, awesome. Like I mean, uh, that would be great. I'll give it a try like you know, maybe it'll just start to work Uh You guys don't have to wait for me. I mean like, you know, please please continue with your next agenda item If you know what I'm going to take two minutes if I if I'm going to be successful I'll try to you know, let you guys know and grab the ball and see if I can show you the application You know, uh, in a fully functional state Nothing beats live debugging on a recording call Um Sure, um, so how about we do this, uh, let's let's hand it over to to Radislav And then we can come back and uh and help you Uh and see the rest of your of your demo that way we we pack in the agenda Awesome. That's fantastic. Thanks a lot of this Sure, pleasure. So, uh, Radislav you have the floor awesome. Yeah, so, um Ed was in the loop actually he he put me there as well I was not aware of the of this discussion, but so what I want to bring up is um the topic of monitoring and perhaps if Ed wants to share more details about that because, uh Yeah, I So far, I don't have uh, for example in the kernel forwarder in the kernel forwarder We don't have uh monitoring In the in the sense of Monitoring the if the interface is available or not. So I was kind of expecting that to be handled by the manager Uh, May I get I can explain to you why it's not currently handled by the manager Please note. This doesn't make it a good reason or a bad reason It just makes it the reason the choice was made at the time, right? Yeah, and the reason is because So if a if a pod simply goes away Right, you need some way of detecting that the pod has gone away um If the pod And the the only good way that we have to absolutely in all circumstances know that the pod has gone away Is to know that its network namespace is no longer associated with any process on the system Um In order to determine this you need to be running in the host pid space Now the forwarders are already running the host pid space so they can they can actually find the correct uh network namespace to begin with um, so rather than Having a second component that was running in the pod network namespace From sort of a principles of least privilege We wanted the smallest shell of thing to be running with any kind of privilege as possible right now The network service manager has no privilege of any kind The forwarding plane has to have some kind of privilege because it has to be able to insert kernel interfaces into pods And so that seemed like a logical place to put that Does that make sense? Yeah, yeah And so That's why it's there now doesn't make it the right place to put it that just sort of explains to you the thought process okay, so Just to be clear so so far the The monitoring of of the end points life or the service life Should be handled by the forwarding plane right? So the case where or no So a a very polite end point would tell you that it has closed Right, but we know that we can't do that right so in the event that the we are not told by The client politely that it is closing We need some way to clean things up Does that make sense? Yeah, sure. Yeah, I'm just figuring out Who is responsible for that? I mean Should it be the the forwarding plane or should it be the manger? Yeah, because I would yeah, I mean, yeah, totally. I mean as I said If we were to transition that responsibility from the forwarding plane to the manager The manager would have to acquire privilege in order to carry out that duty And so this means that this whole Healing process out of healing process is handled by the forwarding plane and from the from the from the manger No, there's lots of things There are lots of things that can kick off healing lots of things, right? So There are sort of two things the thing that you hit this morning out of the bat was How do I detect that a client or an endpoint has disappeared without bothering to tell anyone that it's gone? Right, um, which is one of the things that can go wrong You can also get situations where the forwarder itself Has gone down and restarted in which case the manager needs to discover that it doesn't have the cross connects that it should have You can also have the case that the network service manager has gone away In which case the network service manager when it comes up can discover via monitor cross connect That there's lots of things it doesn't know about the world that it should know Um, you can have cases where end points go away in which case You know, basically that has to be discovered and propagated to whoever the peers were um, you know, effectively the the the mechanisms behind healing are Someone detects that something you know, either someone has restarted or Someone detects that something is wrong and because state information is shared between The network service manager and its forwarding planes and between the network service manager and its associated network service end points You can always go through and restore the state if one particular element goes away Um, or if you for example lose a network service endpoint So the the the two things that are necessary for that to happen correctly are The forwarding plane has to implement the monitor cross connect api um Is the first one and then the second one is that someone has to detect The unusual case where a client or endpoint simply goes away and doesn't bother telling anyone that it went away Um, and we're currently doing that by having the forwarding plane because it has sufficient privilege already Detect the namespaces going missing Does that make sense? Yeah, yeah um, and in that context, do you think it makes sense that Because I think this feature will be extremely common between all forwarding planes Yep, right So the thing that it makes sense for this one to be, uh, I don't know extracted from the From the current forwarding plane for example vp and just put it in the common place I would I would expect yeah, I mean there's some common things That I would expect every forwarder to do independent of what it's doing And there's no point in having people have to rewrite the damn functionality every time Right, and one of the one of them definitely is detect that namespaces have gone away and take the appropriate action um, and the other one is It's probably the case already That you could that I think all the monitoring code is reusable So effectively all you would have to do is instantiate a monitor a monitor cross connect monitor server And just make sure to tell it when you create cross connects or delete cross connects or update cross connects So, you know, effectively, I know you commented the kernel forwarder with stateless And you know, there's a lot But um, but effectively The code that you're writing about the kernel forwarder that's different could probably remain stateless it would just go poke its data into the Cross connect monitor server and the cross connect monitor server is probably already fairly independent code Because I think it's just starting up a monitor service in general Um, but I'd have to go look at the code to be sure Because we we've already pulled monitor stuff a lot of the monitoring pieces have been pulled out into stk already So if I get this straight basically the the events that are sent to the cross connect monitor That's the forwarding plane sense Should be the result of of a thing that the forwarding plane has done, right? So for example, if the forwarding plane created some interface or something else it should sense that update event, right? Yeah, exactly exactly and In case that the forwarding plane the text that the the namespace is missing or the interface for that namespace is actually gone Uh, or it is down then Let's say the the event that happens after that is uh, it's a deletion. So then I will Basically the forwarding plane will delete that interface or something like that And then it will say I believe it currently marks it is down Which means the network service manager finds that it's down and can then take appropriate action Which is usually going to be updating the cross connected some way I'm trying to figure out I'm trying to figure out the the update event that That should be sent after this self-monitoring thing that will happen in the forwarding plane. So it should be a delete, right? Yeah, I think it's actually marking this down because you you want Manager is the network service manager when a particular link goes down so say it was a link to a network service endpoint Um, you probably don't want to just delete the entire cross connect You probably want the network service manager to negotiate a connection to a new network service endpoint And then update the cross connect so that it sends you the new network service endpoint That way the client is undisturbed That makes sense Yeah, yeah, yeah Okay, I will I will continue to look at the code and see how Feel free to feel free to reach out because I think we should be able to help you along And I think most of the code is either fairly trivially relocatable to to common Or could with a very small amount of work we were locating to common. So I think it's Uh, it should be quite doable Yeah, yeah, I agree. Okay. Thanks. Appreciate you bringing it up though. Yeah Um, so just yeah, so just as a follow-up, uh, any Any thoughts on the second? part of rudder's left Questions about the granularity level any ideas about this Uh, could you could you say a little bit more about the granularity level part of the question? I Apparently missed that going. Yeah Yeah, I think the question is basically, uh Just just as the self stated as uh, so I know it is Uh, so basically we have like a monitoring the name spaces of a pod So is it necessary or is there a use case? Uh to go further down to monitor, uh, the interfaces Yeah Okay, so that that that's an interesting question because essentially what you're saying is right now The forwarder is keeping track of the pod's namespace And if the pod's namespace goes away Then it figures the pod is gone and it should you know mark things as down so the appropriate, you know Action can cascade through the system So that that's the way it's currently working. Are you asking should it instead be monitoring the interfaces? Yeah, yeah, what I'm thinking is about so, uh, I don't know whether in the future is likely for a single Endpoints, I'm probably going to receive a multiple multiple requests or that means multiple connections It belongs to different, uh network service. So basically, uh, it is likely that only one of this or some of those connections are down or broken Well, the entire pod is still running So the entire pod is still running. We'll try and connect them to someone else who can give them what they asked for Without disturbing it already today. Oh, okay. Yeah, so if I've got a client Let's say I've got a client and the client you had basically So take the case where I've got an endpoint as you said that has multiple connections coming into it for multiple different network services Which is a totally valid use case. There's a lot of handy reasons to do it sometimes And say one of the clients somewhere else disappears the forwarder detects it It'll mark the client is down in in the monitor, uh event to update the network service manager The network service manager because it knows that the client is the original source of this Realizes that it now has to send a close message Towards the network service Okay, so service endpoints point of view all it is going to see is an orderly close for that particular connection And as soon as it has, you know, received that close Um, we're gonna go ahead and in the normal way as if the clothes that come from the client will tear down the interface Okay, cool. That makes sense Yeah, yeah, but it's a it's a super good question now that we dug into it cool cool so, um Back to uh girish. Did you uh, did you manage to get the um The demo up and working Yeah, it worked. Yeah, it was just a mergle Fantastic. Well, let's uh, let's go and jump back to that and uh, and see what you have. We still have time, right? I mean, are you on? I'm sorry. Is that question to me or yeah? Yeah, the question's to you based on time We're on 12 minutes. I'm absolutely here. Yep. Okay, so that fixed the issue in garage Absolutely. Yes, it did Yeah I think the trick would have been like, you know to change my policy like, you know the image full policy Which I I actually just updated anyway, so I hope like and I'll not see this issue anymore Plus once you guys have the releases in place like and I'll probably switch to it like and then we'll probably not see this issue ever Yep All right. So so I'm back in the uh, measure ui So I'm just going to reload the page, you know, just for sanity and and here we are Um, so sure I'm just going to choose the nsm namespace and and if you guys can see my uh, Show here. I'm again watching the nsm pod namespace. So um with with some predefined defaults The operation is like, you know An installation is actually requested and you can see the pods, you know come up like, you know as we speak Um, it takes like a few seconds, but then like, you know in in about like a minute like, you know, it's it's all up and running Once this is all up and running I can actually show the demo application. Hopefully like, you know working Then go so now you can see like all the nsm pods are up and running. Thanks to you guys like, you know, that was actually a great help We're like, you know, right on time so So now, uh, I'm going to show you the the hello nsm application, which I you know, tried to showcase in a diagrammatic way a few minutes back so, um Once I click on that, um, you can see the uh, two pods come up like and and you can see I've annotated them the right way Hopefully, uh, I kind of like, you know leveraged the examples that you already had in place so you can see the app a Um is the client and ab b is the end point The nsc and nse if I can probably say them the right way So so you can see Okay, awesome. So two pods came up And uh, one other thing is that like, you know in the ui one one thing like, you know You can see here that like it shows you some markers I'll also add the easier way to actually work with this like, you know What what I've done is that if you click on the second one here, you'll also say that hey, the service is available And it's also like, you know, so it's available on a particular port So now, you know, this is the ip address of my note and if I go to that port like, you know, you can see now Uh, the you know the services up and running and and this one is actually that so the client Which is a browser actually spoke to the service a and and and this the app a is actually talking to ab b over The nsm network and in order to prove that it is like I'm just printing the ip address as well Like, you know, just for sanity So you can see like, you know, the whole thing actually just work like and this is We like things that just work just working Absolutely, yeah Awesome, yeah, that's that's that's pretty much of a demo Like I mean we can make it much more colorful But you know, I just wanted to you know ensure that hey, you know what like I'm trying to use a demo Which in fact is a legit legitimate like you know nsm demo But not something which is you know, kind of not using the nsm capabilities So, you know, which is actually the reason why I wanted to run it for you guys today So this is this is definitely good simple demos are a lovely place to start Um, and you guys have done a great job on this and I'm quite impressed with the integration New York you've done with the user interface as well. Um, that's actually very exciting awesome, awesome Yeah, this if you um if you have any anything written up on this We we can ask Lucina if she can post it posted on twitter And simultaneously I I personally think it would be great to get Uh to get a little bit of instructions on this or at least point to your instructions on uh on meshry If you have any documentation Uh written up and we can we can we can point from from our website and repo has an as That would be the other one that was mentioned. Um The other one that that nicole mentioned is we have a bunch of other examples That we could add here as well um, and We probably want to chat with you a bit a little bit on slack about getting you guys on the Helm repos because that will probably make your life much easier awesome That'll be wonderful. Yeah. Yeah. Yeah. I mean, it's No, you should not be having to check out the repo to make all of this go Yeah, I know I mean feel feel free to check out our repo whenever you want, but yeah, it's not necessary No, it's a nice repo. I recommend it Awesome guys, thanks a lot like that was very helpful. Yes, I will actually like I mean I'll ensure that like, you know, we put something together like and I'll share it with you guys and also I like I'll send you guys the links to the meshry repo So you can actually I mean you can point to it and and also specifically the ns I'm an app repo like, you know, where most of the action I just showed is is happening. So that'll be that'll be wonderful Thanks a lot guys. This has been a great That's actually that's a great suggestion. I guess I hadn't considered that nickel eyes a suggestion about the example apps was Was yeah was in part potentially to include all of those in meshry such that Such that the provisioning of each of those is Is facilitated actually now that I think of it nsm is Well, what we interface with More service measures than you see in UI Some of them most of them come with their own sample app Some of them don't But nsm takes the cake in terms of coming with As many as you guys do There's a lot of interesting and creative things to be done So So remind me what is what is your goal at the at the end of the day? Is this so for people to try try things out? Or is your plan to get these things so people can deploy them in production and have them Configured and managed by our system. Yeah The end of it is the ladder is production grade You know comfortable and confident Safe and there's a lot that's missing by the time that we would get there People would try initially we're trying to answer a couple of questions that people had asked us very frequently, which is You know in acknowledgement that there's a lot of service meshes and they say well, which one should I get started with? and the second thing they say is Hey, wait a second There's a lot of value coming out of the mess, but there's some overhead here as well now That my packets are being stopped and inspected more than they were before. So what's the overhead? We created mesh we'd answer those two questions most prominently so to provide to your point fedrick like a In a playground is the kind of the term that we'd used initially and over time We'll let that playground term fade and use more serious sounding words about Yeah, you know sort of day two operations and beyond that There's a bit of an aspiration around some things for developers some things that maybe people haven't seen out of a service mesh just yet Actually, I realized now when I ring this up In conversation with Ed. He I think he takes five steps further than than I have already So so yeah, the short answer is Yeah, we're hoping to acknowledge that there are many many people Better yet to come to service meshes that it's going to take a number of years And so trying to help educate them and walk them through and understand the differences between service meshes Allowing them to contrast them I'd be familiar but but in the end like the So there's a whole suite of people who feel much more comfortable when there's a user interface in front of them There's a whole suite of other folks who feel entirely uncomfortable if changes are being made through a user interface and not being synchronized with You know The code repository and so at some point we'll be facilitating for that as well cool, so my my recommendation on this as well then is going to be nsm like you yes, you're going to play nsm. You could display liquidy stio, etc um What what my recommendation would be is to think about the interface where Suppose that you wanted to have like a multi-compute cluster of istio and What would it look like for nsm to establish the low-level connectivity and then you install istio components related to networking on top of that so especially especially around envoy and so Definitely think about because right now it's a it's a flat hierarchy, which is the right thing to do at this at this point but You know definitely definitely keep that in mind because This could be something where people right now trying to do multi Trying to do like multi cluster federated cluster is really difficult. Like this is what keeps major Major companies up at night is how do we how do we even do this stuff? And so this this is a That the type of making that integration easy Would be I think would be a good Driver of people towards you because you could say well we have a point and click install that does that Just go install it and try it out and so so I think you can you can get the additional visibility Visibility with with that one use case on out Thanks for that. I agree We just had someone asking about that yesterday in the channel So so there's a there's a small window if you build it, they will come until somebody else builds it And then they'll go to them if they do it first so That totally um, yeah Another dreams Just to clarify real quick in terms of so that's actually one thing that measure is really yet to get to is to build in kind of the environment as a noun and hence kind of clusters and environments and Speaking of an nsm domain just to clarify for my own edification is You can have multiple domains within a kubernetes cluster. What defines an nsm domain is somewhat It is Pretty flexible as I understand it. Yeah, it's actually super super flexible I usually refer to them as network service registry domains because it makes it super clear what we're talking about which is It's a place where you have registered People who you know network services and the endpoints that provide them and those endpoints Anywhere frankly And clients can connect to them from Anywhere Which is the whole point behind inter domain network service mesh And we usually usually because it's just easier to get the demos up and going usually people will do network service registry domain that is isomorphic to your kubernetes cluster But the point where the whole thing gets super interesting is when You can get a network service Um independent of where you're running Yeah, get back or offer it What do you guys say to folks who who chalk up their understanding of nsm to nsm being a Well an orchestrator is wrong but a facilitator of uh additional container network interfaces Um, I think it's sort of missing the point Because the the interfaces are merely a mechanism Right they're they're just a tool any other tool would do it's sort of like saying that Envoy is really just uh a thing that that munges ip packets It misses the bulk of what it's doing Does that make sense? Yeah, you know It doesn't happen to be the case that all that stuff winds up in ip packets Sure But at the level that of most of the stuff that envoy is doing I don't think it actually cares Right Um, it's just that the existing users are already on ip networks. Nice. Uh, thanks. Thanks for this guys. This is actually um, so We'll have to see if uh Nikolai is is inclined to maybe add some of those sample apps here That would be fantastic to have Uh to have some people get familiar with the adapter harsheny has been fantastic from lumina Um, if if you guys have met her Yeah, I've prem told me about harsheny, but I've not met her yet. Um, if if you can Do me a favor and create a github issue on this, uh, that that would be fantastic and Uh, with that we're out of time and um, so Is there any last uh last minute request that people have or last minute comments? Okay, well with that I want to thank everyone for your time And we will see you at cube.com next week if you're there and we will see you here the week after Thank you. I'll enjoy the rest of your day. Awesome. Thanks a lot. Thanks. Bye. Bye guys. Bye. Good. Bye