Layer5 Meshery Development, 9 Mar 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Meshery Development Meeting (March 9th, 2022)

Description

Meshery Development Meeting - March 9th 2022

Agenda -
- Dry run before starting meshery. #5173
- e2e tests for lifecycle . #5171
- Two same commands under channel `set` and `switch`[PR]
- Meshery perf helm . #5166
- cobra genDocs functionality to auto generate pages #5108
- User Preferences
- Issue #2842

Join the community at https://layer5.io/community

Find Meshery on:
GitHub: https://github.com/meshery
Twitter: https://twitter.com/mesheryio
LinkedIn: https://www.linkedin.com/showcase/meshery/
Docker Hub: https://hub.docker.com/u/layer5/

A

So please please, if you can yeah here, we go.

B

B

Yeah, I do nice picture so yeah, I'm new here. Actually I was working with this project for quite some month, but due to some medical emergency, but now I'm just starting out. I just got stuck with that installation problem.

B

You know so I'm just still working on that problem.

A

Okay, good by the way so yeah any problem you got like we are here to help you.

A

Other than siddhant like, uh as he said like he is not a newborn by the way, so he has been endorsed in the past and do we have any other. uh I see like every like similar faces.

A

Yeah yeah, hello, everyone um should I should I just introduce myself yeah: okay, um I'm sorry gauche, um I'm a passionate golang developer and uh uh I'm currently working at uh vineyard. It's a small latex startup. uh I discovered mystery from cncf per set uh from lee's talk on a service mesh management plan and uh I'm new to sort of service mesh and uh so I've hopped on to the code base. You know to see how it is implemented, and I want to know more about the project.

A

Yeah. That's that's great. Sorry, again, yeah, starting with the least uh thought is like even better so yeah uh yeah and the goalie or you're calling on is like very helpful for uh both of us, like you will be also advancing and yeah. Our project will also so yeah. Let's see if we have someone else here.

A

No, I don't think right. So uh we can uh thanks for introducing everyone, like all the new gamers, very welcome to you. uh Let's go over our first topic, which is like uh dryden before starting measuring, so uh push.

C

Okay, I'll share.

A

C

Yeah definitely, okay, so let me give some context. I won't be showing what exactly is the problem? uh Let's just check? uh Okay, so if sometimes uh you run machete and some you will leave the system uh and then the state, and then it gets steal. Okay resources get steal and then you again run message system start and you like get these errors rendered. Many tests contains a resource that already exists. Okay, this is a very known problem.

C

A lot of people face it, especially when they are running misery via make run fast or run local, or maybe some other medium okay. So the issue is to remove these and handle the situation. Okay right now. What I have implemented here is like I try to dry, run the helm, installation, uh that's what we do with our mastery ctr. If that dry iron fails, I will clean up the cluster.

C

I will assume that there might be some scale resources uh and that's why it is failing so I'll start to clean the cluster and then make a fresh copy of all those uh deployments. Okay. So that's what I have implemented here, but this is not an ideal approach. This is a bad approach and it's it.

C

Someone might may like point it out that this is not good. Okay, I don't want my deployments to get deleted and then recreated uh can you just adopt those existing deployments and then deploy the rest of the rest of the adapters or measuring whatever?

C

So that is what uh the issue is that uh can we do that? The answer is kind of no okay. uh I will show you how so right now uh yeah, okay, the solution to do that like try to adopt the existing resources. Is that we release uh we make what should I say, we install these hem charts under the same release, name: okay, what help? uh Initially, uh the like server deploys machine operator under measuring operator release, name like the same name measuring operator and what message ctl does.

C

Is it deploys under measuring release them, so they conflict and says that it is under different name and you are trying to install on a different name. I can't adopt these existing resources.

C

Okay, so that's where it fails, but what if we try to hard code, it like. I am saying that release name is hard coded and it is measuring on mexico and also it is same in server. Okay, it also doesn't work I'll, try to show you, so let's try running return fast. Okay, not here, make one fast.

C

So what I'm doing right now is building this machine server locally. It will automatically detect that cluster is running and deploy uh operator with this release name, this hardcoded release name.

C

Let's see if it started.

C

I think I have to login here.

C

Okay, let me check if it started to deploy.

C

Okay, so it's starting to deploy the operator it will deploy with the release name measuring, let's check, okay, good, so mystery operator has been deployed, but guess what it adds measuring in front of it, because the release name is different from the chart. Name. That's why I haven't add these this meshery release name in front of the deployment name?

C

Okay. So now this breaks a code in our backend. How here, okay I'll reload it.

C

Are they running okay? So, let's wait to like meshing to be running.

D

C

Okay, so operator is running, but it can't detect operator because of some code implementation, because we try to find the pod, which has only machinery operator, name, not measuring machinery operator and let's try now installing via mercedes okay. So what will happen it will adopt? It will try to adopt this message operator, pod or deployment. It will say. Yes, this is under the same messaging release name, but it will fail still. It will fail and let's see what.

C

Okay, so that's really good, it says that it has cids installed and it can't install now like it is failing again now. What I could do is skip the crds. This with mystery. Ctls, uh like here like I, can add, to skip crds start start.grow. I can like add a new variable like skip cid, but then again I will be assuming that it. It is a cid error that we are failing to install crds.

C

So what is the better way? The better way is to just clean up the cluster and try to make fresh copies of those instance. Instances.

C

I am already assuming that there are some steel resources I know like where, when we run this like run this dry run it it outputs, something like this like starting measuring, delete, still necessary resources, stopping naturally resources and deleting machine in space. This is bad. I can silently hide these logs and do it myself.

C

That's the one thing that I can improve here, and I think we should do this because, like uh trying to tweak the problem.

C

And if I try to like uh go with him and do things, it will just generate more and more errors. I like it now, it's like it did adopt the operator, but now it says that crds are already installed and I again will assume that let's skip crds now, so it will be kind of a loop with helm mirrors.

A

If just uh apply manifest files, instead of and so uh you like the same set of errors that would be getting on you know because I've seen like uh you already have a crd, you apply once again, then no changes actually so it gets light.

A

I I know the least about helm by.

C

It's just the problem with helm installation so like it will start to get like loops of errors like this error. This error this uh and we don't want to like make it hard for us to like get okay. Now. This is the existing resource. Let's skip it now, or this is the existing resource. Let's skip that now and also it makes changes like very weird changes. I showed you that measuring measuring operator it will break our code. Our code won't be managing okay, sometimes search for my stream machine operator. Sometimes virtual machine operator.

E

So please uh so I was uh yeah, so my question is: is the release name like bound to the helm installation in the when we do helm install? Is that where the release name goes.

C

Yeah we can change this to like a completely override in our hem charts. That's one thing that we can do okay, so then it will be only measure operator, but again the crd error that will stay.

E

Yeah, so I am, but is the crd error happening because uh crds are like name space and like it won't happen with other components which are namespace.

C

uh Like can you like be more brief, I I couldn't understand what exactly you are trying to ask.

E

Yeah, my question is uh in this: this crd error isn't happening because of the release name, conflict again right.

C

E

And like being.

F

C

F

Yeah, it doesn't actually talk about release, name right. So if you fix the release name issue, you use the same release, name which we should be whether helm was saying it or not. We should be using the same release. Name: okay,.

E

F

Here, under the crd error, it's not saying that there's a release, time issue it there could be, but it's not saying that it's saying um these things are different. Have you tried addressing that issue by.

C

Setting them yeah that's the next step would be like editing. The hem charts like first uh maintaining the full override of this uh deployment and then managing this like it's saying that it should be managed by help.

F

Yes, so so are we done? Are we not using the same charts to install? I don't know.

C

I think there is a like there is some some minor change in full override, like we are not fully riding in measuring operator hem chart. So I think that's that's where the problem is like. If I check here so full override is not written here. It's empty.

F

Okay, actually, let me actually a different question: are we the exact same code and the exact same charts should be used when, in all of the methods that measure operator is being deployed or measuring is being deployed?

F

So I wonder how we get the release name off in the first place like that's, because we're not using the same code, you need to go fix. We need to go change that fix that there needs to be one mechanism to install measuring and then the managed by the release name, all that stuff should be the same should be unless like post deployment- and this isn't your issue, push like you didn't this. Isn't this isn't um yeah? This isn't a moment of embarrassment for push. I just want to make that clear.

F

That's not what's happening here! um This is actually piyush finishing off a long standing um initiative to that that husseina and others have worked on to well do what abi was saying is a solution and obvi's suggestion is correct, like one of the solutions here is, take it helm and kick it to the trash, because it's a piece of trash, and so you probably keep it with the rest of the trash in the trash.

F

It's it's really not a good um like his whole focus is about packaging and installing and it anyway, it's it's appallingly, not good uh anyway. Abby's um suggestion is straightforward, simple and will work absolutely hands down every single time, which is just used to manifest.

F

We moved away from the manifests so that we could, in concept, get economies of scope of using the same deployment utility uh pervasively, and so we figured we'd get really good at helm, given its popularity and universality in terms of um the fact that most of the things that we're that meshry is managing have helm, charts that we can um manage in that way,.

F

Yeah, I guess you're the only okay, um sorry, there was just feedback on the line, uh so I didn't mean to necessarily you know interrupt. This has been a really long conversation on helm as it turns out like we tried to save time. I don't mean today I mean like over the last few months and it's been and the pain continues, which is in part, why I despise helm.

F

So we're not reaping the benefit of having consolidated to a single tool now having said that, we're kind of not all the way there yet to using the tool in a universally consistent fashion. We're really close. I think this is the last area where.

F

There's two different styles of mesh, free server deployment or messaging operator deployment.

F

We just need to make sure between measuring ctl and mastery server. Those are the only two methods for deploying mastery operator today, no they're, not there's helm directly.

F

I think that's it. Those are the three I think so we removed the manifest, so those aren't gener generally available to people when someone uses the bash script to install mesherie. The bash script in turn uses metry ctl when mesri server detect connects to a kubernetes cluster and it doesn't have the operator, it uses a mechanism to install measuring operator that mechanism.

F

We want to make sure that that's the one and the same mechanism that meshtree ctl is using, which is a mesh kit based libraries around helm, but somewhere in there we're providing it different values and we're invoking it slightly differently, and so I wonder we might you know, I wonder if these all go away, if we make sure that it's the exact same implication, it's the exact same code.

F

It I have this sneaking suspicion that they may not go away because helm is a steaming pile of, and so even though we go fix this, we might still end with an issue, but we need to do the diligence and finish off all the work by consolidating to a single mechanism.

C

Okay, so one more comment from my side: uh I love this dry run idea and we should definitely implement across our uh like various installation code, because it's good like we know that it is failing. We should implement some counter attack to like remove that error, so uh we should keep it right.

F

um Yeah, maybe or not like it's it's this, it's no! No! It's this thing like it's a weird thing. So if everybody thinks for a moment like uh here's the thing so so push you're familiar with me, taking a quick analogy last time we spoke about kind of the.

F

What feels like really straightforward logic and then becomes really complex, is something like like what you're doing, which is. Let's say that the user wanted to ping a server and I'm not. I won't go through the whole thing, but but just the point is like yeah yeah: it makes it intuitively. It makes sense hey why don't you dry run, make sure everything's kosher, you know and gonna work and then let's do the command. It's like well, that's kind of the bane of management software in general, and that is okay.

F

Let's say that we're going to do a ping okay. So before we do the ping, um you know, because if you do a ping and if you're going to hit a host name well, there's a dns lookup that needs to happen first, otherwise you might be able to ping an ip address.

F

You know that same host if you had the ip, if you use so anyway, you there's an intermediary like there's prerequisites to being successfully pinging, something so hey before we ping it. Let's go ahead and do a dns lookup and make sure we can get the ip address. Okay, we got the ipad okay. Now, let's go ahead and ping and then in that time dns you know forgot itself and it changed, and um you can't ping that ip.

C

F

And so then you have to have logic.

C

It will be like making our lives harder.

F

Again, yeah, it's like like so the thing yeah and the thing about it is like just do the thing try to do the ping and if it fails, you're gonna have to deal with it anyway. So if you do a dry run, all you're doing is adding additional logic.

F

It still can be the case that, between the time that you did the dry run and when you actually invoke it that it still fails or the dry run, didn't cover all the use cases and it still fails and you're still going to have to handle the air. So it's like uh yeah sort of it's kind of nice to ha maybe you know anyway.

F

Okay, having said all that to because, because I'll lose my sanity, if we, if we keep, I want to go publicly flame, helm and the whole project on somewhere in in the next cncf toc meetings. What I'll do- um and that is the the solution you have here is a you know. um If, if we were at our wits end, then this this is feasible like this is um functional. What we, the best experience, might be um that you do measure ctl system start it resp. It says um deploying deployment oh detected, existing operator deployment.

F

Would you like to reinstall and then do what either like? If they say? No, then that's fine just execute the rest of the charts, the sub charts without the parent chart like like get more surgical or um or they say yes and you do and you do what you have right now.

F

F

And and that's not too bad, that's like like and in general, it is generally not a problem that we would reinstall mesh free server. Even if you just silently did it as you were proposing. That's that's um the best choice that you can make like that. That's a that's a good choice, because there's no stateful information persisted in those custom resources and there shouldn't be- and from I mean we don't, but maybe the user would and then yep. Thank you hobby, very good and then.

F

uh That you that yeah they asked for it to install and part of installing. If you take an application download an app from the web, you install it and then you download it again from the web and you try to install it. Do you think you're gonna reinstall uh yeah you're, probably going to try to overwrite your thing so like it's, not unexpected, behaviors, I'm done.

F

I stayed within time.

G

Plus five points.

A

Okay, uh yeah so push any closing comments.

C

Anything from you, so the idea is to like work on the handcuffs and maintain same analogy to install and charge across server and client.

A

Okay, so so, let's, uh let's move on to the next item, uh which is which I have to take it uh right now so uh end-to-end tests for uh life cycle so before that what is life cycle is that uh we are managing all the service mesh life cycle here in this uh uh in this uh ui, and this is uh using mesh ups p1. So we have our mesh of v2, also which, which is basically using the configuration section.

A

uh So so we want, uh since this ui is going to be uh there for some while so we want some tests uh to be done there so that you know uh we don't feel any drag uh in between.

A

So for that uh uh this is the pia that uh do very foundational basic tests, and uh uh we got some problems uh meantime, that is, you know, stopping us from uh doing the further testing of these uh life cycle management uh for service meshes, and these are the three issues that you have and uh yeah feel free to take any of it uh ping me on the slack or on the you know this uh or on any or on the public channel in the machine where you are anywhere.

A

uh So I will be very uh happy to you know: explain what are these and how to solve it. So uh once these three issues are solved, so we will be, we can move further further uh in yeah, so I will just take one more minute, so I have to like explain what are the issues uh in a very small time. So so you are saying like this is still so I'm supposed to go to istio page uh right now, because I have clicked on here, but I'm not going there or due to some g.

A

uh So this this error has been you know introduced uh recently. uh I don't know from which code part, but uh that's why we want ui test because someone doing some changes in the uh life cycle management, then it takes uh some other parts due to its side, effects and yeah. So so, basically you're saying like I can't change other than that since kumar adapter. I I'm only running r2 adapters right now, which is uh yeah it will. I think my.

D

A

Yeah, so yeah just believe me that I'm running two adapters right now. uh My server has been down right now, so um so a citrix service measure. This is showing this ui, which is not desired at all. So uh we we want that to be changed, and that is the issue number two. So if you see istio you go to like our query, parameter after localhost 1000, you go to citrix service mess which is not available.

A

You go to this, so there are two uh ui qualities written for the same components and that's what creating whole of the mess there. So we have to just uh get away from those. These are not very hard issues or they can be. You know taken as a beginner friendly issue, so uh whoever in the on the call like anyone who is interested in react so- and I want to try so yeah- please do so.

A

These are the issues, uh uh so the link is already there, so you can just go and see other than that uh yeah. Do we have any comments, sir?.

D

No okay, so where are the links? Sorry? Yes, I yeah also the links are they in the meeting yeah yeah? Here we are so so this is the link. So this is.

A

The issue link and uh here is the uh following:.

B

A

Okay, yeah thanks a lot by the way. uh Mario. I like uh those, are very great comments. I have one from get got from you, so I will you know uh we'll do the changes accordingly uh in some time other than that, let's move to views again for two same commands in the uh channel yeah.

A

C

uh I found something today but you're doing it like. I was reading some pr and I got to know that we are. We have two commands that do the same thing and we are managing the code that is so weird to me. Okay, so this is a set command in this channel. uh What it does is it changes the channel for you in mesh config. You can set it to stable, stable subversion, edge and edge latest. It should not be version here.

C

uh Let's make sure it's not version now, okay, so it sets the release channel and version. That's it and then here is a command that is switch. It does the same thing, but just one extra little thing that it starts. It restarts the the.

C

uh What should I say measuring it also doesn't consider measuring might not be running, so I you don't have to actually start machining and only restart when something is running.

F

It doesn't need to consider whether or not like, don't we don't want to make it that smart, it doesn't need to consider whether or not mastery is running just um try to stop it, and if it's already stopped great, just fail forward. Yeah.

C

That's good, okay, that's good! Then we could have these.

F

Merged no well see yeah, it's deceiving all right. I mean I beg to differ, or let me let me allow me to submit to this to you for consideration, and that is uh it is you're totally right. It's such a small nuance like there's just a slight tweak in behavior and I'll, submit to you that it's material, the the slight tweak is so the difference be, and it's not necessarily obvious from the verbs set and switch kind of. If you take a moment to think about it, it becomes intuitive but probably up front.

F

It's like not necessarily intuitive you'd have to so so so here's the scenario you are you're running mesherie and you're running mess with a particular on a particular release, channel, let's say you're running unstable, let's say you're here, contributing and you're like oh yeah.

F

We just um that pr just merged I'd like to go ahead and try that out in a deployment I want to switch to the edge channel, the edge release channel, and so um there are two commands for doing so, and you kind of have your choice of behavior, and this is the so one is like this particularly becomes.

F

This potentially becomes really useful when you have multiple measuring deployments right now, you have to be a pretty advanced user to really care about the difference between these two things in the future. If you have multiple measuring deployments, this might be more helpful, and that is that is, if you say, mestre ctl,.

F

System is it release set or system channel? If you say set, then you're, just updating your static configuration your mesh config you're, rewriting that file to say that hey. I want the current context, whatever context you're in to be to use this release channel, which might be your current context or it might be a different context like you might want to say, hey the other context that I'm not in this one, because it's a flag um set that one to that channel and this one to that channel this one, and so and that's it.

F

It just writes to disk it updates your config. The switch to your point is like a one line change or what have you the switch command? Does that plus it like takes action right then it's like! Oh, I thought I said to switch to I, oh okay. I said to set the channel to that, but not necessarily actively roll. My like because when you say switch it'll, not it'll, then it'll actively um it'll actually switch the context as well like it will switch deployments on you, it'll it'll.

F

You know tear down your deployment and move it to the new release channel or the new stable version or whatever, so that that's the difference is like one yeah.

F

If you think of better verbs or if you think of um you, know something more obvious or maybe you're like well, why don't we just have that as a flag, not a separate sub command. It's like that's yeah. That's certainly debatable like it just it's sort of like yeah. Whichever way we go, let's just be consistent like we consistently have a bunch of verbs as sub commands, or we consistently have a bunch of flags that uh modify the behavior of and so.

F

Okay, I mean: does that mean? Is that help I mean like, like I'm really glad you brought this up. This gives us an opportunity to reflect on whether or not we've made the right decisions in the past whether or not we're being consistent with you know the design of that the ux of nashville cli as we go forth.

C

I don't have any arguments right now so yeah have we like it's not a big deal to merge it right now. Maybe we can if we feel, or we get some better work to merge them, because it's just like making uh maintaining more code and like marking errors and oh.

F

I see, I think I see maybe more of your point. Maybe it's so long. Okay, like I, would be in agreement with you if your proposal is to leave the two verbs leave the two sub commands set and switch because they're different behavior but eliminate the vast majority of the duplicative code other than like the three line. Difference like oh, okay, yeah yeah, that makes tons of sense.

A

Okay, yeah so yeah great discussion. So let's move on to uh another item here uh so since search is not here, so uh let's move on to aditya yeah, so add coverage and docs functionality to our 200 pages. Yeah you're. There.

H

Yeah, let me shine my screen. First.

H

All right, let me hide the midi controls too. Okay. So uh for recap the cobra cli library has the ability to generate dogs from the measure from the command line, uh golang files itself. So I have been working out that for past week and for updates, um I got to know that the cobra cli library doesn't have the option, to like uh say change the template for the um generating the markdown pages, so I had to refer some documents and finally, I was able to generate the equivalent amount of the markdown page with this syntax.

H

So since we are using the pre and div tags, um I I was able to add it in deep in this program with the exact uh markdown as well as html syntax. So this is the output. I have got it, so it all depends on the uh whatever the data is present in the command right now. So some uh bit errors like the spaces and all might be there.

H

So we can just change that and for the yaml, so that that one gave me an headache right now so because, uh since there is no um preferred way to generate the uh yaml file, what we desire, so I had to create the way in which it which it gives only these four variables, like name, description, usage and example, but I still didn't figure out how to generate the uh yaml file in the way we desire, like I'll, show the example the present okay, so uh the ml file alone.

H

Couldn't we couldn't generate the wavy design. So what is uh what I'm thinking is like if we can generate the documentation from the data which is present in the any measuring ctl command itself? So I was uh thinking like whether we could ditch the yaml file and then depend upon the data present in this um commands alone, like this one.

H

So that's my idea. So any comments on this.

F

You you'd end up having to kind of pre-process the dot.

F

um No, it becomes um fragile because what happens is if someone renames the remove command to removes or something then there's fragility in the system where you'd have a mapping you'd have a static mapping of what commands are like you'd, be not only back to where you are right now, but also with some hidden bugs that would just pop up every once in a while, because there's a disconnect between the way you know between the docks in here you'd also have to pre-process to convert from that golang format to markdown and then um so.

F

What were you saying is the blocker for why um the auto-generated docs won't work.

H

The order generation doesn't work for the um yaml file alone, in which yaml file we make is a bit different than what we use here so for the intendation part and all we need to reconfigure it, and it may take a lot of time to figure out how to do that, since the cobra doesn't know how to format the animal file. Let me see here yeah. This is the function here actually.

F

So you're giving the base name of the dock to be generated cmds.yaml, so it the the docs that are being the yaml's that is being generated. Is it just a single file? Yes, is it it's a single file, okay, no, or rather, of the docs, that being being auto-generated through this new library or through cobra?

F

Is that a single file for the yaml file? Yes,.

H

For yeah they're, all single files and for the markdown files as well.

F

Okay, I'm not asking my question clearly, I think um cobra's ability to auto-generate documentation to auto-generate a yama yaml.

F

Does it have the ability to generate one yaml file per command, or is it restricted to generation of a single yano file.

H

Yeah generating yaml files for each and every command, yep it it's possible, since we are uh aiming to cut short the yaml file to a single one. I have been trying to do this, but to your question: yes, we can generate multiple yam files for each and every command, okay, and so.

F

Given that okay and today's architecture, today's um today's information architecture of the the way that we have the yaml files, the the yama laid out for the measuring commands, the mastery ctl commands those are on an individual command basis or no.

A

No they're in one file.

F

Okay, so they're in one file and I'm just a dtm's trending like they're in one file and so okay cobra, is also capable of consolidating all of the info that it generates to a single file as well, which I think you know you've. Just you just said all this, okay and but what you were saying is uh the so what's the issue like the formatting of that like.

H

Is challenging the problem is the formatting, because it depends upon the data what it is present in the commands like say.

H

Whatever the format we give here, it will just generate as it. So if we give like space or tap space, then it would generate this as well. That's the issue.

F

Okay, yeah hey um deal with it, I mean like yeah strip it out, let ben you know, put his hands on the wall, spread his legs like strips.

A

uh Yeah, actually uh can we use like uh replace all functions like uh since we are generating these examples, so we have been. We should have written this code somewhere, so uh just one replaceable function that will replace all slash and to an empty string and non slash, t2 empty screen two, so I hope it will or to its. uh You know, slash and slash 32, the two tab spaces or whatever it is so won't that.

H

Work yeah, I think that might work. I will try that.

A

Other than that uh you said like this is not in the correct format. So uh what do you want? Like uh you, want the three dashes uh online number five, then three dashes on line number ten, something like that.

H

Okay abc so currently, oh, I have to okay I'll close the file. Let's sing, seven reopen again.

H

Okay. This is the uh current yaml file which the measuring docs uses to refer the data so that it may generate the pages for the measuring ctl command reference, but what the cobra does is it generates. It generates data necessary data for the measuring cpl commands. That's not that's that not the issue, but the main problem is that the intendation right in the nation. So if I want the space, it doesn't generate the uh tap space.

H

It just gives like with name description, usage example and just with uh no space at all like this, that station okay, okay catch this month,.

F

This one a dt like just format it like you, need it formatted I mean like. I don't know how smart it is. Oh we'll catch me in slack or something um because it's just string, manipulation.

G

H

A

Yeah thanks thanks victor, uh let's move on to now uh or sorry. Anyone here so uh some comments on user preferences, yeah.

G

A

uh My my peer is addressing this issue. It's a bug, while changing uh the user preference it panics due to nil pointer dereference, so.

A

So, um in order to uh look into this, uh I I started looking into the user handler and and the variables of the struct that is used to um unmarshal the request body.

A

uh It actually uh is a pointer to an object, so we need to take care of the nil case, and- uh and we all know uh that this occurs in the scenario where we have may have ignored a nil case, so uh so uh the before this, uh the previous code, the validation of the load test, preference, didn't thought about that it can be nil. So a better way of doing this would be to validate only if it uh actually present and not for the nil, so yeah um I've started the server and.

A

I'm sort of if I go to preference.

A

And uh say so see here, um as as the request body uh weak uh the front end does not send the load test uh preference at uh because there is no one so uh yeah and if we, if we change something, then it it gets applied and for the next time you can see it. It is applied for yeah. So currently, I'm I'm doing this for uh the loca, a local provider but provider as as the code base, the provider is a is in uh interface, uh it it implements the remote and the local.

A

So if I do for uh remote also it will work. So if I have to do it for remote and show it, I have to turn off my to authenticate myself.

G

A

So do we have any comments uh uh thanks, shot equator? This was this fix was needed for a long time, so yeah, okay.

A

So uh next item is uh by usana or hamza. uh So this is about like issue number 2842, which is migrate system, config, eks2, global sdk.

I

Yeah hi everyone, so this has been long bending issue actually, so the problem was that aws sdk they didn't have support to fetch the cube config file. So recently, hamza has come across issue in the aws sdk go repo saying that they are not going to support it as such, but their argument is that cube. Config file would be needed by uh either a cube, cutting or helm kind of cli tools.

I

So they are not going to support this, but instead they have uh redirected us to a stack overflow post where you can actually get all the information. Whichever is needed from the cube config file. So in a programmatic way, you can fetch all the tokens context everything so we took the snippet actually and we don't need all the things that they have proposed in the stack overflow post. Let me put it out there on the chat.

I

So using this uh code snippet we were able to get the uh once again yeah. So the answers section there is a code snippet right, so we were able to get the.

I

Tokens and everything, so we we didn't, really need a client set there as per our requirement. We we needed the config file only, otherwise there would be lot of changes required in the config code itself, so we reached a point where we get the structure which we have to actually convert and write it to a cube config file only eventually. Otherwise, uh maybe we have to rewrite the entire config code flow.

I

So we are thinking in the terms of whether we can convert this programmatic config into again a cube config by like appending it to the existing cube config file. That is one way. If that is not possible, we are going to change the existing flow so which with has a dependency on the cube config file right. So whenever we do the config, we eventually get a merged uh cube config file with the existing cluster information, as well as this eks cluster information that we are trying to connect to.

I

So we need to evaluate whether we need uh all this information in a cube, config file, or can we just uh have it in memory and use it whenever we are doing uh the next set of operations? So these.

F

I

Parts that we have next to tackle this problem, so that's where we are so any suggestions or comments on this.

F

This is exciting. This is great. It's potentially um you know tons of folks use eks, um and- and this will make it really, uh this will make it a lot more seamless, and the challenge that you just described makes a lot of sense that, like it's a debatable, it's kind of a debatable approach on one hand, you generally, we generally probably don't want to rewrite anyone's cube config, not that there's too much of a risk of us messing it up.

F

Just more like uh you know we that would probably be unexpected, but for them and we want to keep our hands off of it a bit um now, okay, but if we keep it that reconfigured cube config in memory only where are we going to run into issues it's like well actually, for the most part.

F

Yeah I mean for meshview ctl's part memory is.

F

Probably fine, considering that it's really the system start command, I think, is really the focus of where it needs this, and it's really about initialization and deployment of mesh reserver.

F

So if it performs performances procedure hasn't memory potentially it could pass that along to mescheri's server, which would be ideal because that would be ideal. It would probably be okay if it didn't pass it along and mesh reserver had to do a re-evaluation.

F

The reason that it would be if it has to do a re-evaluation if, if mesh reserver is deployed in cluster, it doesn't need to re-evaluate it. Just it can start going.

F

If mestre server is deployed in a local docker environment, um then it does need that updated, config passed along because otherwise it's going to try to it's going to try to look to the environment variables for the aws access id. You know access key and secret and those are not likely to be within that docker container it won't be and so yeah. Ideally, I guess I guess what I'm saying is like, ideally in memory. We just need to make sure that that gets passed along.

F

This is the other issue, that's gonna happen and josh, I'm just going to mute you for a second, as we tie this off. The other issue that could potentially occur is: let's say that we are successful in performing this procedure using mastery cto it has that updated, cube config in memory. It passes it along to mesh reservoir, which initializes successfully um using that config.

F

I think we'll. This will probably require a small logic change and probably for the better in meshre server, where a lot of times when mesh reserver when measure ui refreshes it might be, it might look to your local cube config. It needs to stop looking to your local cube. Config meschery server needs to be have cube, configs handed to it and then persist them, and it does with the start of support for multi-cluster. It starts to do that, and so there might be some residual issues with how it's doing that.

F

But what I'm saying is in general, measuring servers, structure and approach to how it manages individual clusters and consequently, individual cube configs, aligns it's going to persist that and and save those separately from looking them up. Every time on, the file system, which aligns very well with.

I

Okay, so a machine server purchases them in a actual uh goal and structure format, or would it uh have a copy of the cube config altogether? uh I believe it.

F

I

Structure right, config.

F

That's structure, good question. I would tend to agree with you. I believe that if you're using the local provider it will persist to sql light in. Like I guess that's that's not a you know. It'll persist a sequel light uh and I can go I'll. Go look at to see if it what structure it keeps.

F

um I would guess the same way, though, that you just said.

F

And so yeah, so I mean all signs point to storing. You know using storing this in memory, okay, at least for measuring ctl right, like as it passes it to mastery server and then mesh reservoir should generally like we need to. I don't know if the logic today is.

F

If it is continuing to look to the local cube config every time that measure ui is refreshed and if it's trying to load that like and and basically it it shouldn't, it should stop doing that. We just we need to figure out the logic where, basically, if it has one or more cube clusters connected, then like don't look locally.

F

If it has zero clusters connected, it should probably look locally, because it's super convenient for mesh reservoir to spin up grab that config and keep going.

F

um Yeah, which, which would you know if that logic is in place, that that lends itself well to wh as measuring server initializes it receiving a cube config handed off to it from ctrl. So.

I

Okay, so, anyway, like we are starting to look at uh having this golang structure config, we need to do some pre-processing of it, because it's in the aws sdk format, it's not really a kubernetes kind of config, either rest or the client cmd config format. It is a different format altogether. So from that, if we can, uh we need to figure out a way to if we can write it into a file format. First, we are also looking into that.

I

So I mean that would uh reduce the cold churn on config side as well, because that's what we expect as of now so, but I mean we would. We would know right if uh machine server needs it in a in memory structure format, it would be easy to do so. Anyway, we have to do the processing on this config structure anyways. So we will explore that for now.

F

Nice and the prerequisites for the end user, that's using mastery ctl, the prerequisite is. Is it just? I suppose it's just kind of one thing that they would have they would have configured. They would have a couple of environment variables present, which generally means they would have probably configured an aws cli.

F

I

uh Actually, right now we are taking these as parameters right, the region as well as the cluster name, so these uh are actually given as a prompt. As of now, we could change them to uh sub command arguments as well. So when we move to this programming model.

D

F

Yeah or no or yeah, no that's great, I mean maybe better to be explicit like like we are today, um yeah, yeah, sure yeah.

F

Anybody else have comments, questions we're at time.

F

Yeah this, this is really helpful. It's one of those things that will delight people. It's gonna, be good.

C

F

All right uh abby did we miss anything.

A

Yeah, uh no, no, no, I don't think so yeah. uh So there are. There are few uh future items uh uh which which we will discuss.

A

So I hope we can, you know, uh take it forward to the another week or this community meeting right.

A

Other than that uh yeah, that's it for the today and thanks everyone for joining and yeah good night, all good evening or good day, according to your region,.

B