Layer5 Meshery Development, 3 Aug 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Meshery Development Meeting (August 3rd, 2022)

Description

Meshery Development Meeting (August 3rd, 2022)

Agenda:
[Uzair] Fix Broker/MeshSync connectivity test #5866
Why we do assign an External IP, though IP is “pending” for meshery-broker svc?
[Uzair] Fix panic on reset of database #5893
[Abhi] Eslint Error Fixes: No-undef: the variable is not defined but called #5947
[Abhi] Eslint Github Workflow: #5954
[Abhi] Enable duplicate Key Check #5953
[Subedy] Updates on the component workflow.
[Subedy] Empty state on confirmation modal.
[Harkirat] Deployment Modal: Search fully functional. #5948 /merged
[Ruturaj] Snackbars on adding new k8s contexts #5884
[Ashish] Removal of “Meshery” version and standardizing version across components.
[Ashish] Removal of AMD hardcoding for ARM compatibility/ alpine image for Meshery
[Alpha] Fix relative file path location for token in config #5910
[Nithish] New endpoint(/api/meshmodel/validate) exposed in meshery server
Need to update docs
[Yash] Meshery-ui-restructuring branch updates PRs to merge as soon as possible Open PRs
Tracker
[Rohith] Improvement on menu #769.

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

Okay, let's get started, then uh hey guys, welcome to the machine development call. So uh it's it's a ritual around here to start our calls with getting introduced with our newcomers.

A

So are there any newcomers on the call who would introduce themselves or any old-timers who want to introduce themselves again.

B

Hi everybody I'm a fairly newcomer. uh My plan is to kind of lurk uh for a little bit, um but I've been following the project for some time uh from afar.

A

B

One of the co-chairs for tag observability and generally a nerd for service meshes and open source.

C

And it took me forever to find my my mute button. I he said nerd and yeah. I just couldn't help but comment. It's just like yeah, perfect, perfect uh one among the 19 other things that matt is doing. Matt is also reviewing the service mesh patterns book, which has a lot to do with a lot of the work that many of you have done or are doing uh the we might even talk about the catalog today, so uh so anyway. So that's great.

B

Oh fun, I I guess if I was to be a a little bit funny, um a long time listener. First time, caller.

B

If you ever lived in seattle, I don't know.

A

All right anyone else who would like to introduce themselves.

A

uh Cydene, I think this is your first call go ahead. Please.

B

A

My name is syrian gamuti, I'm from tunisia. I I joined the level five slack.

B

uh Two weeks ago and I'm.

A

Kind of discovering everything well, I already have an issue of gone but yeah still discovering things. That's it good to have you here uh all right, so that's all the new comments. I see all right. Let's get started with the agenda then uh you're up. First, would you react.

A

Open screen is visible, uh so fixed broker messaging connectivity test, so this is regarding uh we perform connectivity from server to to detect that whether our machinery, components like broker are missing, is uh functional correctly or not, and the tests currently are misleading because they don't represent the actual current state of the measuring components. So I was working towards making those tests reflect what the state what the correct status is. So this is the pr I have created, so this is uh here.

A

I've described what this does is, and while I was work working and testing this, uh this new test that I have created, I found a a weird behavior that uh when I was testing uh in multi-cluster environment I had mini cube and docker desktop up and running and and my and in mini cube, I, the the broker machine broker service was not exposed. That is uh external ip was not uh was not yet uh provided.

A

It was in pending state, but still I was able to uh connect to the measure broker like, like the the uh the external end point, the external address and port was populated for the cr. So when I, when I dig into that, what may be the problem, I found that we use this get endpoint for populating those fields or attributes in custom resource, and here we check that if it is a load balancer, we check whether the status is in pending state or what is the state and if it is still in pending state.

A

We then also populate this address like this node port is populated and addresses populated, because of which I think uh it is conflicting, the conflicting behavior. So if uh before proceeding right with removing this on making this function correct, I want to ask like this: behavior is relevant, or why is this such a behavior.

C

So I know that mr young is intending to to lurk a little more and he might not have heard just what you just described. But it strikes me that he might have an opinion here and so.

C

B

Tr truth be told, as you were, describing that I was like bringing up links for all of the things uh that that we're talking about. Could you do me the give me the embarrassing honor of like a tldr? uh I'm sorry guys so.

C

I think it would be good for everybody actually, including including me, which is, which is to say that so in order for mescheri to manage uh kubernetes clusters, it requires a deployment of the mesheri operator. The operator has two custom controllers: one is just a nat server to help with communication between the operator and measuring server.

C

The second one is mesh sync, which is uh which communicates with kubernetes or subscribes to kubernetes events using kubernetes and formers, and in order for um meschery server to be a nats client of that broker of that nats um broker.

C

We need to expose nats as a ser as a kubernetes service, so um particularly in in the case where meshware's server is deployed external to the clust. Excuse me to the cluster, so the question becomes: okay, like you know it's environment, specific. What type of um kubernetes networking and it's like. Well, we ship with a default of load balancer for the service type.

B

So there's a service type called external service, which is meant to expose a routable yet potentially external, although not necessarily external, to the kubernetes cluster um service, endpoints, it's an ip address, uh but to expose it uh in the context of cluster local dns. So you get like you know, cluster.local.comsvc.local.namespace.

B

You said so like in the ontology of services and service discovery, you can surface something outside the cluster.

C

B

A uh service type of external service or a location one. Second, I can get an example. We used to use this as part of like, um if you're familiar with, like a strangler fig pattern where you're decomposing a monolith into microservices. You know you might want to put the initial set of microservices. That's your end state and expose them as kubernetes services.

B

You know for a cloud native application so that, when you're running locally or in an arbitrary cluster, wherever you've got, you know a portable definition that is inclusive of configuration for services, um but they're, not there yet they're, still back in the monolith off on a rails instance, or something like that. That's where you would use one of these external service type services, I'll I'll I'll, grab a link and throw in the chat, but.

C

Okay, yeah: that's.

B

How I would model it and that way all the default kubernetes stuff just works, and it doesn't become like a bespoke thing that you have to maintain in an ancillary um system and or distributed store, which.

C

C

Which makes sense, and so as as um in the uh okay that begs, there's the question of the mesh sync custom resource, which has a few different attributes. One of those attributes is tracking of its cluster local ip address and its external ip address and for mesh reserver, the majority of the time well or in the case where mesh preserver is deployed external to the cluster.

C

It reads from that custom resource looking at the external ip address field and then uses that ip address or that host name to uh try to try to connect its nats client. Try to read, you know, talk to mastery broker and so user to the that's in part, that's more or less. What you you are articulating is that that's the behavior, but more than that.

C

Like so, we certainly run into the case where, depending it's like, it's very environment, specific how networking is configured, how dns can is configured and we're trying to reduce the time to value for people who for for users and we're trying to be a little bit intelligent about sort of the happy path like, particularly to the extent that someone is deployed on docker, desktop, for example, where it makes kubernetes networking on your local host very very easy.

C

If the external ip address here, if we ship by default with a service type of load balancer in the case of docker desktop, you would also then get an external ip address populated for docker desktop. It would be localhost if you did minicube it's more of a pain in the rump, because you might have to load up an external, an add-on like metal, lb or or some other networking solution to make sure that you can get an external ip address populated.

C

I guess so thus far loser we haven't helped at all one question before I potentially make some suggestions, so one question is: do you are you aware of the behavior in measuring today? If mesherie is deployed in cluster, so local to the broker?

C

Do you know if mescheri would will use the internal ip address, or will it continue to just always use the external ip address.

A

No, I'm not sure about that.

C

Something something to look into just to and then.

C

Yeah yeah there's this guy there's yeah. This is a.

C

Like it makes intuitive sense that if there's an external ip address populated that's what mesh reserver should be attempting to use or should be using to attempt to communicate with mystery broker and part of your question is there: it was the mechanics by which mastery examines the external ip address value and then records that into the custom resource one. I think that that's probably a one time thing that's happening. So, oh, no! No!

C

Actually, um I can't recall like if, if you go in and a user manually edits, the deployment type, the service type from load balancer to like cluster ip or node ip is the operator reconciling that back to load balancer? Is it undoing that configuration.

A

No, in that case, it is not done doing that.

C

C

A

A

We can discuss and the next second thing was uh when we were when we were the setting database and broker was not connected to machine server. The server was panicking, so I fixed that and another I'm working towards subscriptions uh for the context menu like uh now. If cluster is deleted, but then then also in the ui uh context is shown like it is not we have to refresh the page, so I'm working towards making a subscription for that. So these are my updates.

C

Nice, I suspect, on the first topic. We didn't quite help you we left you with two questions, not answers um of what of what matt has pointed out about external name. I think I have some reading to do because I yeah I.

B

Feel, like the point being like, instead of a selector, to identify a service in configuration. If you create an external name, you're, basically saying you can use this name in lieu of a selector and on the back side, you know instead of pointing to a load balancer or to a cluster ip.

B

It points to an arbitrary routable ip right, so it kind of demands that outside of the cluster, it's findable either because you know the the dns that the cluster is beholden to either a local dns resolver or you know we're seeing a rise in layer three. I think um you just talked about that at tlc meeting yesterday about network service match right, so there are other other ways to make network topologies happen other than reliance on on on dns hierarchies.

B

I've always thought pnrp would be cool but in any event that's kind of what I meant about the external ip, and I don't know if that's an external name. Rather, I don't know if that actually is a solution for you here, depending on your goals. But if it's to to have something portable, you know, then you could take the service type as configuration to an.

A

B

Action or the helm truck that sells the operator and then like the user, could you do something by default right, but then, like oh by the way, if you're going to run nats as a local docker container for iteration or in your own infrastructure, you have an existing. That's implementation, that's hardened at scale right, and I want to use that. Instead, I want to bring my own maps like.

B

That would be the scenario where you could optionally have a service type override to say: hey when you go make the endpoint that everyone standardizes on which looks and feels just like a kubernetes service, which I know how to talk to either through selectors and configuration for some things if you're doing tunneling or whatever by configuration. Or um you know uh you the the scenario I described.

B

um You can support that and just just make that an implementation detail, that's settled by configuration, but for all intents and purposes you don't know or care that it's not hosted in the cluster as a workload which you would refer to by a service right. So you can actually get some weird. You have to be careful too because, like I don't actually know what happens, if you have a service type external name, and then you, then you associate it with an ingress.

B

You know, like you know, you could bridge things inadvertently. I haven't explored those. We used it for the scenarios I described as well as a handy, quick way to glue in uh on-premise databases as part of a migration to something like.

C

Rds over time thing that if anybody else was disabled, almost able to keep up with matt um one thing that is one thing he was highlighting and is evident is that uh is that, well, um users need to have it. You know we need to make intelligent guesses that be somewhat opinionated about just working in.

C

Hopefully, the majority of cases will be any number of cases where people are are earnestly and seriously using measuring into their bespoke environment.

C

That, um given the high configurability of networking that uzer one thing to take away is um the need for to allow the user to reconfigure uh that might be, and that'll be helpful, just even in the various configurations that that you all deploy measury, that that someone could go in ideally to that the settings user interface and um dictate either a different, a different value for well what's inside of the cr and it to matt's point it's good to be aware of the this external name configuration which, if I, if to put it into different terms, the way that I digest the concept of external name is is in essence, like a point of redirection like a uh like a host name that you might specify.

C

uh That is, is just your default moniker for how you might how you might point to where mystery broker is. But if the underlying networking is different.

C

You could still potentially reference the same external name. Am I bastardizing that, if to think of it as like a dns uh or a host name in dns, that.

B

Yes, exactly like you could set it to a host. I might put something in chat, sometimes in the morning it's still the morning for me in the morning, I I I can write better than I can.

B

Oh, but but yeah I mean in my mind. You know whether it's the weird probably edge case scenario of someone who really really wants to run this seriously and has an existing nats thing. You know that's more of a database use case, but nats is fine too, uh like it.

B

It should be possible to to to use that like naps is just always looks like a kubernetes hosted service and if it's running in local docker, and then you know we're running this in kind, you know you can you can use one one way to configure it, but if you're running everything all in one so that it just works out of the out of the box, even though there's no boxes anymore, um you know just default to to a kubernetes service.

B

That's actually a hosted kubernetes service in cluster right, but you you're removing the concern of carrying where it is because it always looks local.

C

Yeah totally yeah that highly appropriate. uh um Thank you for that man. Thank you.

B

As an idea lacking context or depth of knowing anything, I'm gonna go back.

A

All right any other feedback on that.

A

All right, thank you, sir matt and b for sharing that uh moving on we have abhi abby. Would you like to speak about the slim terrors, yeah yeah, uh just sharing my screens.

A

Yeah, so uh we have been advancing in some uh yes in uh errors so uh past week or two uh we had uh you know experienced some uh uh idiotic, like a very uh childish behavior uh for our. You know, past mistakes that uh some uh ui were breaking because of uh some in-depth checks uh not getting caught on time.

A

So uh un-depth is actually very important check, so this needs to be caught, and this was in off state for yesterday for a very long time, so uh getting the help from yash and all we have been able to uh do some progress here so uh yeah so other than that uh there are two more important checks that needs to be uh enabled just like we have enabled on depth checks and therefore uh some code needs to be changed with respect to it so yeah we we have it.

A

So this is the uh issue depicting here. I I mean you can see it here and I already see uh interesting. So, thank you uh come on uh and yeah uh one more important thing uh is enable github workflows, so we have been running a whole lot of workflows on opening up, pull requests so uh adding one more uh to check the uh you know, ui links uh and you know when we are committing uh so the eslint checks runs as a plea commit rule, but those can be suppressed by the contributor.

A

But if it is inside the github workflows itself, uh it won't go missed, uh it won't get missed by any of the reviewers. So uh just a small check like I have classified those two issues as the uh first timers issue or one who is uh new to the project or other newcomers. uh I have labeled it as a newcomers because it is actually a newcomers issue. So those who are just wanting to start out with mystery earlier five and a eu stop so they can definitely show their interests.

A

Yeah, so actually that was something on yes.

A

If anyone's interested, you guys can either signal on slack or you can also write down your name in the minutes itself. Yeah volunteer right off yeah any feedback and comment on what shake share with us.

C

uh It's got to be done. um It's too big, it's too painful. When those checks aren't there, you know failed releases and that kind of a thing yeah.

A

Exactly uh so, these are kind of the high priority issues, so if no one is there to pick it up uh in a required time, so I will probably uh fix it ourselves. So, uh and these are, you know, really first timers issues, so if you are really wanted to get static source, that's a great first pick for you.

A

Thank you for sharing that with us moving on uh sivadhi. Would you go ahead with your updates.

A

I don't think somebody is on the call okay, we can circle back later on if he joins.

A

Would you like to share your topic here.

A

Sure yeah, you can open that here in in your desktop as well.

A

So earlier the search was like dysfunctional after this pr is, it has been merged now, so you can probably check the video like it's working. Fine like the search bar just like correct yep. There was a small update from my side. Thank.

B

A

Thank you, yeah.

C

Just yeah just to comment on that for a second, the the search is, um I guess great greedy is maybe the right way of putting it. Is that let's say that your connection was mini cube that if you started with, if you started searching- and you started your search with, I uh minicube would still be displayed because part of its letter matches and so.

A

Okay, I can like probably get a new pr for that.

C

A

C

A

Actually, I don't.

C

Know that I I was just kind of just a remark for everybody else, just to just to reinforce that one that I thought it was important enough to go to test and then to um sometimes that. Actually, that's really helpful.

C

Okay, because you know people might eventually, if someone loads up like 10 contexts, 50 contexts, maybe some of those refer to the same cluster. Maybe those are different clusters, but if they have a naming scheme, imagine as an example- maybe they have um you know dev east devwest, dev, south and uh instead of having to type out dev hyphen and then the search term, they can just jump directly to south and so so yeah. So you know good job.

C

I think it does well.

A

All right moving on, thank you. Moving on ruth raj, would you like to share your skin topic yep.

A

Yeah, so the pr uh this is about uh giving the user feedback when they upload a new kubernetes context. uh So let me just demo it so right. Now I've added a couple contacts and I have a config file which has one extra context. If I, if I import it like that, uh it will tell you that the some of the contexts already exist and that one new context was added in the uh snack bar so yeah. So there's changes on machinery, server and mastery cloud.

A

So yeah, please review it and there is some ui changes also. So this is the first time I have contributed to ui so make sure you make sure I didn't break anything and yeah.

C

There's a couple.

A

Couple problems with the uh ui some check, which I'll fix right now.

C

Yeah, that's that's fantastic! For my part, just I I, if I recall correctly, someone is working on a subscription for that page.

C

Nice yeah, which, given the inability to refresh the screen in the docker desktop extension or the docker extension that just puts pressure on subscriptions so that that's great but yeah, those um snack bars those toaster notifications.

C

That type of feedback to the user is really helpful.

A

Any other thoughts or feedback on that.

A

Okay, I guess not uh moving on ashish, you have two back-to-back items.

A

um Hello, inaudible, yep yeah, so actually I have a cold today, so my voice will be a little low. So if it's inaudible just tell me yep, I'm sharing myself now.

A

Yeah so the first, uh the first thing that I want to do. The first change was that uh the removal of meshi version, so in before this thing, what we had was that in version you could search a measuring version which didn't make sense actually and that mesh version were the core components and these components are the extra functionality that uh our adapters add on top of uh what the service meshes already have to offer, but those were categorized into their own different version, which wouldn't quite make sense.

A

So a pr was recently added in msc and then subsequent prs adapters, which have standardized the use of versioning. So all of the uh all of the components, whether they're our own custom components or there are the service mesh, specific components- all are now versions. So here you can see that authorization policy and mesh tls authentication- these are linkedid's own, you know, components and then the add-on and the linker d mesh. These are our components so, uh basically before this, if you were to drop it here.

A

Actually there is this small bug that I have pointed out of this. Only it happens only for linker dimension is theo mesh. This just one specific component, the rest are fine. Actually, so if I drag and drop lingerie add-on, it's fine. So basically, before this we had a version field around somewhere around here and I've removed that. So, if you draw, drag and drop a linked image from stable two point, one one point two and you deploy it. The linker dimension of that version would be deployed.

A

Same goes with the other add-ons as well, and same goes in uh htv as well so uh like. If I go onto any version here, you have an schematic component here. Also there is this uh issue, but I've again pointed it out. So this is with istio mesh lingard image, each adapter's mesh uh component that it has this uh certain styling issue. So please you can look into that other than that uh yeah. So that's that's pretty much it for this one update.

A

If someone has questions regarding this, I can answer otherwise I'll go on to the next item.

B

I had a quick one and just like, like guess, no, not don't hold. This. Are you integrating with the business component of linker.

A

uh Yeah, actually, these are these the this component that you're seeing these are the uh basically uh internally what we're doing with just grabbing the manifest that they give us. So each service mesh has some add-ons and they have these manifest for each version. So when we like, uh when we deploy a linked add-on or which add-on we don't, we just grab the manifest for that particular version of the add-on and just deploy that.

B

Awesome thanks. That's nice and future proof.

C

Yeah and so to want to like expound on matt's question, um one thing that isn't always obvious um and there's um there's some faith, maybe in here and some faith in the amount of like regression, testing that happens basically, every single time, there's a pr and every single single time there's a release of an adapter which is, um which is the notion that if uh ashish, if you were to click on the versions drop down, maybe just for link or d, we would see that you know there's a that.

C

There's a lot of versions in here and essentially that every instance of one of those components like the viz add-on that that component is version. You know every single one of those carries a version. It's a version specific to. In this case the service meshes version.

A

C

A

That happens whenever the these components are built. So when these components are built, this is a two-step process. The first step is to basically generate the service mesh components and just the second step is to grab the previous grab the core components from the previous latest version and just copy paste them onto the newest version that generated components for.

C

And then there's there's a few. You know. I don't know that this is necessarily written down anywhere, but it's certainly been accounted for that there's a few different.

C

You know um scenarios in which someone might have an existing linker d deployed at an older version. That's covered here. They might have a newer version of linkerd deployed like there's these there's various scenarios in which there's potential caveats um and we don't need to iterate through them all right now, but but it's just a no, maybe I'll. Just take a note that we should explicitly write down the behavior in each of those situations.

A

Yeah, so if you have insta, if you have already running linkedin control pane of a specific version, you can just drag drop, that version and uh click on undeploy and that will yeah.

B

I think what lee might be referring to iran, linker dean in production for three years, two and a half years uh fairly recently through a bunch of upgrades- and I believe it's largely resolved in in current uh linker d.

B

But if, if anyone was running not so old versions, like, I think, even less than a year or year and a half ago, um as many might be doing, uh the upgrade involves manually deleting like hard deleting explicitly the crds because they don't handle crt crt versioning didn't exist, contemporaneously with earlier versions of linker d, earlier versions of linker d2, I'm not referring to the older link or dv1 pre kubernetes variant.

C

And that particular use case actually that map points out again about upgrade as a scenario that people are facing, but that's the type of I mean to me sitting in the operator's seat. That's like hey, makes me quite uncomfortable, I scoot towards the edge of my seat, because I get nervous about an upgrade, particularly for core infrastructure like that, and so that would, I imagine, call me to use a tool like this, to the extent that it facilitates experimentation or verification or dry runs of uh of different versions.

C

You know running my services and helping facilitate kind of uh identify, upgrade catches, so that is exactly.

B

Specifically, how I discovered this project like a couple of years ago or layer five in general, because we had just gone from frank and istio just to low balance grpc, but not a full control plan um to link her d and then we were dealing with lingerie deployed by flux uh and and by the way, this scenario of crd versioning uh is is is is normative.

B

You know for for lots of other helm, charts and and products. So it's not just a link or d thing, but it is a wrinkle. That's now been.

A

B

B

um The other one that was a complicating thing before they added support for cert manager was, you know, most of the ways that most people follow the documentation to install link, redeem the right way, generated, self-signed search the time bombed after I forget how long a year right so like we actually had pieces of arm. Some of our clusters go down the first time that happened.

B

When you know someone forgot to go, update a cert, you know, and so we we worked with buoyant um uh directly and gave them a lot of feedback. uh We ever quote rather my team uh around this scenario in particular, uh and and gave them a bunch of feedback and they actually went, did cert manager. So.

C

With cert manager, now is it? Oh, I'm ignorant here, but are you uh are you did they make it uh easy to be notified about that type or they just? Are you just set up an auto rotation policy and.

B

So certificate manager kind of handles rotation of certificates based on the number of predicates, and it also exposes a number of time series metrics that are scrapable via a prometheus exporter. So you can.

C

B

Dashboard we had, we created dashboards and they might just be on grafana.com now, but we created dashboards and alerts that would they would start squawking a month out. You know when it was time for search to be rotated that had been created. That way, um I think again. This is more of an upgrade thing.

B

If you start fresh with cert manager integration, then you probably don't have this issue um outside of edge cases, which also we had happen to us um istio, for example, and we had the same problem with link4d if you, if you hammer on it hard enough uh the file change notifications for when the configuration changes in this case the certificate get missed because file, change, notifications and linux aren't guaranteed, they're usually delivered right, and so, if they're not, then suddenly you have certificate failures and you have to bounce pods, but like there's again like until certain manager came into linker d, and I suspect same is true for other meshes, there's all manner of hilarity in upgrade scenarios, but again that's that's edge cases and moving forward.

B

It will be not a case, so I don't know what the support matrix is or how far back you need to support in this project. Yeah.

C

Yeah, the yeah, the upgrade scenarios get messy, you get oh yeah, I mean I mean messy for not messing for everyone all around, including this tool trying to keep up with those that's good. um I see she were going to show another item.

A

Yeah, the other other item was regarding the removal of amd, hard coding and um compatibility. So basically, we have a multi-platform workflow that generates uh docker images for both arm and amd platform. So if you go to docker dot, io, slash layer, 5, sorry.

A

Sorry I was putting the yeah, so if you uh basically see any image from layer, 5 or organization, most images have will have two: most images will have.

A

There would be two images, one would be arm compatible and one would be amd compatible and depending upon what machine you're running when you're the cubelet or your docker pulls the image, it will pull the appropriate image so either.

A

So, if I so basically, we had this thing earlier early had pushed this change of having both amd and arm support, but still that some things weren't working and the reason for that was. We had some hard coding going on in the docker file, where we explicitly had while creating the binary. We explicitly stated that we need to build it for x86, so that hard coding has been removed from here and from all the other adapters as well. So now, all the adapters are compatible with both arm and amd 64..

A

uh Basically, there is one little thing that is left and that thing is that mesh is still not compatible with arm because of one's caviar. If anyone has a suggestion here, I would really. I would welcome that physician, so the problem, I can state the problem real quick. So basically we use an alpine based uh image here in the uh in mexico.

A

We use an alpine as a base image, but we had some ellipse dependencies because we, when we built the google binary, we built it with the sigo, enabled because the sql package that we use requires some uh shared shared objects uh which are which require there to be lipsy and other such dependencies. So this was the image that we had found that had those dependencies, but the problem is this is only uh this is not for arm.

A

This is only for amd, which crashes the uh which, basically, if you try to build this image on arm platform, it will fail. So we are on a lookout for an uh alternative for this. I have searched, but I I couldn't found um an alternative, so the alternative would be an alpine based image that has all the uh glypc dependencies and which is both compatible with arm and md. I tried to manually, uh you know, install some of this stuff.

A

You know make some updates, but every time I used to get a new form of error. So after a point, I think I thought that maybe having a proper base image would work. So if anyone has any suggestion of have for an alpine based image, that has the lipstick dependencies and is both arm and amd compatible uh you're. Most welcome to create a pr on this docker file and update the base image. If you're able to successfully build it on an arm machine as well yeah yeah other than that, the adapters.

C

So she should we as a in either as a separate like a separate docker file or as a different, multi-stage or another stage in here. If you're, using a full-blown alpine base, image did did that not have the right headers.

A

Yeah yeah, you didn't.

C

Okay, so if we move to like a ubuntu or some something much yeah.

A

That will have, but that would load the image.

C

Yeah, and so that could be, that can be a work around in the meantime, where the image that uses the alpine or that uses ubuntu as the base image is the one that's tagged with like. There would just be a special tag, maybe just a point of configuration for folks like prima or whoever wants arm compatibility, that they would just you know when they did measure ctl system start when they were configuring, their mesh config that they would just modify the tag as a potential.

A

Okay, now that we can do that.

C

I don't know, I don't know if it's worth it, but I just I figured I'd point it out as you're thinking about the problem.

A

Yeah, if someone wants to run it um and they replace this with the ubuntu image that would basically make things work so yeah that can work in the mean. Meanwhile, I've created an issue and if I haven't, I will cross check and create one so that we know that this is an outstanding item yeah. So these were the two updates I had, and second one more thing would be that earlier we were using a cube. Basically, we were using our own uh binary, open api json schema binary.

A

It was a node.js based binary that we were using and I think we are okay, so you know that binary is still there in some of the. So I think there are pull requests being opened in all of the adapters. I would be merging those, so the small update is that we are no longer using this.

A

uh This whole step, because we don't no longer require this binary, because the schema validation that we do now we use qlang for that which is which is completely go native, use, go functions for that go packages, so this will be removed, and this also reduces the image size by 25. Megs is so yeah, that's that's all from myself.

A

Any thoughts or feedback on that.

C

And when you open the when you open that issue um who's there, if you ca or ashish, if you can like suggest that one of the the temporary workarounds is to permanently modify the workflow, that we have to build to have two different builds to produce two different images: two different tags: yeah, because that's a different solution and that's an enabling solution for prima, whom I know you're going to follow up with today about performance management. You could suggest that to her that's easy enough.

A

I think this this one can be closed. The one the issues in the adapters can be closed, because the adapters now support the.

A

Thank you ashish for that. uh um Moving on alpha would you like to talk about the topic yep I'll share, my screen.

C

A

So uh there is an issue when the mastery ctel, which has been existing like two weeks ago, where the token location of the configuration fire. If the token has been given in the config file with this path, then it doesn't get expanded due to the fact that uh if you give like dot slash in the token location that gets expanded and it gets recognized. But um unlike npm, if, uh if you give the tilt, slash symbol, the user home uh path doesn't get added into the location, and does this concern cause an error.

A

So I haven't worked on an issue, and eventually I was able to fix this, but by use adding the um efforts block, which checks. If there is a tool slash in the location uh field, then just add the uh user home directory into the as a string then return the uh path along with it.

A

So also uh was it reviewed and it was uh good as well, but he suggested like if there is some way which can be included like if the path can be set as dollar token, like importing the file path from the environment variable- and I was just thinking if it would be better. If we ask feedback from all of the folks in the community, so any feedback from this and.

A

Just last night.

C

Yeah, so so to refresh my memory. So today there an environment variable, isn't supported. What is supported is either allowing people to override and specify the location of the token or by default, using the mesh config and the the token that's identified in their configuration. So this would be a third option like what would be the sequence of the order of precedence, environment, variable mesh config over parameterized, override.

C

Is there a use case for the environment variable over the other? Currently.

A

We yeah. Currently we don't have the support for the environment variable case that has to be added.

A

So this is the um sequence like if there is a absolute location given by the user, then return has to file if the user has given the relative part with uh till slash symbol, so add the user home directory along with the path given by the user and then return it or else. If that is uh no absolute part or given this path as well, then it is assumed that dot the mesh config is stored in the dot mastery directory and then return the config from that directory.

A

So that's how the sequence is present currently, so I think we can add the sequence for the environment variable as well.

C

But probably more important things for us to work on like I don't know that, there's a good like that's a good. We can, as a matter of fact, not probably, but I know for sure um who cena has had on her backlog, for I think about a year now like removing dependencies on uh other binaries, the aks eks docker compose, and I don't think that those have had much motion forward.

C

But it would be great to take on things like that, as opposed to as opposed to adding a third solution for people to add their token. Like I just uh uzer mentioned it was there, did you have a use case for it or, like was? Was it inconvenient to use the mesh configure.

A

I was thinking that if, if a user is uh using machinery to manage their application, so they would have in a directory or pathway, they are all storing their api keys or all the sorts of tokens. So.

C

C

Oh, that's nice yeah, that makes sense and which is configurable in mesh config today, because there's a file yeah a file path inside of there that you can specify.

B

A

Okay, so I will check on it and yeah. That's it for myself, for this.

A

Thank you alpha for saying that uh any thoughts or feedback anyone else.

A

Nope, okay, uh nithya shut up next yep. Let me share my screen.

A

A

Shared the link for the talk, so if you guys, are interested in like what uh okay, I'm sorry like my voice is low, because I have some protein fiction or something like that. So bear with me. uh Yeah. I've shared the link for the doc, explaining like what mesh model is and stuff like that in case you're interested but uh yeah at the top I'll. Just I'll. Try to keep it uh I'll. Try to not go too far into that.

A

So the update that I've made is uh regarding, like it's like a central place where you can validate stuff that, basically all the databases are internal to measuring, although this doesn't actually have much impact uh currently but then like uh in the future, like we'll, be having some uh basically like.

A

If we have mesh map- and we have other components inside measuring which you know want to validate stuff, and so earlier, we used to have different methods of validation, as asis has just mentioned, that we were using uh node-based binary for validating the schemas of crts. But we've now moved to qlang because yeah, because of various reasons that yeah right now we're actually using uh queue for generating the components and also for validating and queue actually offers a lot more than that. So we are actually trying to maximize like what queue has to offer.

A

uh You know so. Yeah validation with queue is uh very easy, and it's like you know. I mean it's like the natural thing to do for us, especially because we have been especially because, because we have actually implemented queue in various various parts of the code base, so yeah so like yeah, the update is actually regarding using. uh You know, exposing a central endpoint where other components of misery can actually, uh you know, send something and then like ask a measure server to validate it, and the validation is actually based on queue.

A

So there are uh two parts of uh updates. One is in messaging server, which basically, you know, exposes this endpoint and then like it's, it's being handled.

A

Yeah, this is the handler, which does that. Well, basically, the structure right now is that I mean it is subject to change. We might actually change this in the future, but yeah right now. It is expected to that. The request is expected to have something like schema, the schema type and the value and the value type. So currently we support- uh I mean uh yeah so essentially like this is the validation. You should give a schema and then like you should give the value and it will actually validate the stuff.

A

So currently we support uh three different types: uh the the schema or the value can actually be in json, yaml and or json schema. This json schema is actually only specific to the schema, so uh this is this is basically you know directly uh direct from like what q internally supports. Yeah q supports, uh uh you know, conversion to and from json yaml and json schema. So that's, basically it and uh as long as like the all of these, things are actually valid.

A

That is, if it's a valid json, if it's a valid ammo, then we go ahead and validate it. So yeah we actually use uh the update in mesquite. Is that uh okay, just second, whether meanwhile anita had a question, and the question was that okay, you can uh so.

C

A

And go ahead, I was okay, so my question was: can you go back to mystery? So basically we have inside of the pattern engine. We have a stage of validation stage there as well, and there we use a different uh go package to do the validation so, like the the whole dryer and logic, we have actually make sure that our design actually goes to only till that stage until that validation stage, which is already present in the pattern engine and then comes back from there- and uh this is a separate endpoint.

A

So to me, why like? Why are we using? So if we are using qlang shouldn't the case? Be that we completely remove that that schema validation that we are using there if you go inside of the remember the exact name? It is like it's a third stage after filler it's, I guess it's validation or something like that. If you go inside of models, yeah actually yeah, I get it like it's actually. The the intended use case is, you know more generalized than patent engine that is like uh by by other components.

A

I mean that, like it can be meshed up. It can't be anything else, so it can either be you can in the future like we can even go to an extent of saying that, like we would still have, uh I mean well yeah. You could actually still replace the the implementation that we have inside the patent engine and use keywords uh in place, but then you could also use the same functionality that we have over here and instead of you know, sending a request to this particular endpoint.

A

You could just internally make use of the packages offered by uh meshkit, so.

C

B

A

Was to generate you, you, the registry, that you're using sorry for interrupting, but uh the registry that basically the thing that against which you're doing this validation, it's our registry right, the where the components are registered? No, no! No! This is no. This is completely the schema can be anything. Actually it's not uh related to our yeah. There is no registry, actually, okay, uh so, um okay, so when you receive, let's say that you receive a json, so the current the inside of pat just tell me: how is it different? Okay?

A

So the current thing that we use inside of the pattern engine stage? What it does is that the schema that is registered by the adapter or any other component. It validates the coming yaml against that schema. So here what the uh like by validation, when you take that json against what schema? Is that being validated yeah? The schema is also being provided as a payload. The schema and the value also actually is being provided as a value.

A

So the thing is that right now, as you can see like the endpoint was slash, ap, slash mesh model and then slash validate in the future. We would have something like slash api, slash, mesh model, slash design, slash validate, so in that case the the value that we would be validating it against would actually be the I mean exactly. It would still have the same thing, but then the schema would be hardcoded, which would be the schema of a design. So it's something like that.

A

So it's sort of like uh you get the point right or is it yeah yeah yeah? So the client supplies the schema in this case, but but in the in the existing stage that we have the back end has the schema? Yes, okay, so I'll have to think about it for and then I'll come back to you because you know yes,.

C

A

Have like schema at two different places and I'll discuss. Let's keep uh take this yeah.

B

We can have the.

A

Discussion later, yeah.

A

Okay, as I was saying, like uh yeah, the other part of the changes is I've seen biscuit, and I okay. These are basically the I mean it's. It's actually like we're just trying we're just directly exposing the capabilities that are offered by queue and then like, where it's sort of like uh we're providing an interface on top of it, so that like we, can actually make use of the things so yeah. That's it. If anyone has any questions, uh feel free to shoot them.

C

Nice yeah we're at we're about a time too. So what.

A

Yeah, just one thing is that, like we'll, we'll actually have to update the dogs uh in regards sorry sorry yeah in response to this change, but that can be different.

C

And there's a review of the pattern engine that needs to um is overdue anyway, oh good. What what are we missing at dt.

A

uh Nothing actually there's just this uh one last topic for uh yes to give any updates uh for the ui restructuring. Yes, are there any updates, or is it just uh to go through the uh to review the prs and watch them? There are small update from the on the performance profile page, but the main thing I wanted to talk is we had to merge these two uh prs.

A

I have a link here yeah, because one of the pr was merged the patterns page, but those uh are related to these two pr's, so it it is showing errors. It won't work until the merged.

A

So yeah I've shared the link to the pr's on chat, so please feel free to review them and, if they're good to go we'll merge them.

A

Alrighty, so we are a bit over time, uh but we have mostly covered all the topics rohit. We can talk about this topic on the website. It's called that's: okay, yeah! uh It's just like one line of code. I think they would have already.

A

All right, uh thank you all for joining, then uh I'll see you guys next week. At the same time, bye everyone.

C

Goodbye guys bye.

A