Layer5 Meshery Build & Release, 10 Feb 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Meshery Build & Release Meeting (Feb 10th, 2022)

Description

Meshery Build & Release Meeting (February 10th, 2022)

Agenda:
- Doubts on issue #33 (Scheduled benchmarks on different sample applications)
- GitHub Actions
- Status of extending SMP to CNCF community infrastructure labs
- Test coverage for v0.6.0 (Meshery Test Plan)
- Automated tests status (Automated Tests)
- mesheryctl
- Meshery Server
- Meshery UI
- Component tests
- Meshery Adapters and Service Mesh Lifecycle
- Functional
- Build and release pipeline for GetNighthawk
- Building and publishing Docker images
- Building and publishing binaries
- Cannot use two docker images in one. Look into how we have implemented load testing with Nighthawk

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

I'm not finding an appropriate like script or command to you know, uh run a particular application on a particular adapter say for say, for example, book info, it's configured uh to run an istio. But what are the commands that that will like help in running in linkedin? I'm not able to find that thing.

B

Yep, uh so I think this is where we can leverage the concept of service mesh patterns. So what we can do here is we can define a patent file which can basically uh which can basically onboard applications on different service measures, so that that would be the way to go and measuring also has a construct called applications.

B

So there is a measuring ctl command called measureyctl app deploy, so we can take in an application file which has which is basically a service mesh patterns file, so we can take. We can define our applications in in that file and use measuring ctl to deploy it, so that should be the way to go.

B

I think I can point you to point you to someone who can help you in writing those pattern files. So once we have the patterns files, it will be, it will be just uh invoking the measuring ctl command. With that particular pattern.

A

Okay, so we we can use measuring sheet, but right now the deploy is all bash commands. So.

B

A

Are we planning to change them or like should I change them in this pr.

B

Yep, I think there is a an open issue to use measuring feature instead of bash scripts. uh I think it has gone still.

B

uh Could you like, uh if you, if you are, if you are up for up to take that, then you can, uh you can comment on it and we can assign it to you, and lee mentions that has some comments here. So.

A

Let's see a pr open in this thing, but.

C

Hi, uh sorry, I just joined late, so I don't have uh the context.

B

So we were talking about the measuring smp action and using using mesh receiver instead of the bash scripts and the the other issue where, where we are trying to deploy the sample applications on service meshes. So lee mentioned that you might have some thoughts here.

C

Actually, right now, I am not dealing directly with this. I believe I am sort of trying- and you know, replicating this on the equinix uh bare metals uh that are provided separately by uh this by cncf. So my work is more or less focusing on that, so I'm just missing where you know I could sort of give you an update. So if you could like you're gonna prompt, uh if you could.

D

Yeah yeah, let me jump in here a little bit, I'm not sure if I'm connecting the right dots, but uh one of the things that one of the overarching themes to this particular meeting is well trying to get mysteries uh building with this process into a pristine place, where we've got lots of tests at different levels. Unit tests, integration tests, functional tests, your mama's tests- I don't know, there's different kinds and we use github's workflows to do that. We use a lot of github workflows um and we also want to not only drink.

D

Our own champagne like use mastery to test measuring, but we also want to empower others to use measuring within their workflows to generate load against uh and maybe do a traffic split when they're releasing a piece of software, so we've created github actions and anhirsh and aditya and navendu and others are familiar with those you guys created those um and- and so there's one angle to this, which is gosh. We just need more integration tests to cover other use cases to cover the various aspects of functionality that mastery does that way.

D

We can make a measuring release with high confidence related to this and really using a lot of the same tools like it's. A very similar effort is hirsch and xing, and and vendo and others have been working toward automating, a collection of service mesh performance tests.

D

um In this case- and this is, you know, like it's extremely related same same some of the same objectives, the overarching goal for what hersh is chasing down is under the auspices of one of our sibling projects, service mesh performance that he is.

D

We just recently gained access to the cncfs cluster, meaning we have access to bare metal servers, many of them and are have begun to work on automation, including the use of github actions.

D

The get this the very same github actions to run a series of tests, and actually not just one time but like perpetually ongoing for many many months for the next year and the next couple of years to come, and so yeah there's just like there's a bunch of overlappings. That's really good because we can use. We can enhance and use the same action.

D

So hersh is able to has proven he's able to um take one of these github actions, the service mesh performance, one and you know and connect it to the cncf cluster and that's referred to as a self-hosted runner to use github's nomenclature.

D

So good hope that gives context to the discussion.

D

So, there's probably a couple of things to hit on here, so one is just in general hearing from hersh about the um kind of kind of the progress he's made sort of where he's getting where he's hitting his head on things um and how that the some of the very same tests these we are running. There um are also work very well as integration tests as functional tests as part of meshes build and release process.

D

So hirsch, do you want to maybe start there just give an update on your progress.

C

So, just uh talking about the progress uh actually, uh what uh jin- and I have mainly you know- the problems that we mainly have been facing are regarding uh permissions and uh on what all hard like you. You need pseudo permissions for some uh for running some things, but then sudo permissions. They say that this is unsafe to use for some uh like uh I'll I'll. Give you I'll give you an example. So, starting off, uh I made a debian server.

C

I realized that the mini the way we install mini cube right so that particular action, the manusa. I I'm not too sure that's the thing, but that github action that we use for installing kubernetes that works only on ubuntu, 18 and ubuntu 20.. So then I destroyed that recreated another one. Then uh I accidentally created 20.10.

C

I did not accidentally actually intended to create a 20.10, but uh then the then microsoft. You know some dotnet libraries do not support it, and now you have to switch again to 20.04, because that's the version that the microsoft some dot net libraries that are important for uh running github actions on that particular bare metal instance. So we need to do that. Then there are a couple of things that uh let's say: uh docker and all they say that okay, don't use uh sudo as the uh the person who is privileges to docker.

C

Instead, you create another user, so jin and I have mainly been firefighting- those things so uh both of us, like you know, we've come to uh this place, so the latest. You know that jin hang that jin. You know even sent me that if I had some uh idea on that uh was, regarding uh you know, uh measuring ctrl installed using pseudo permissions, so that was something that was uh that he particularly was facing, and then we've also updated these things in the top.

C

So I haven't been like this sort of a personal update, I've sort of busted my ankle, so I haven't been able to take this uh forward. I've sort of been nursing that uh so I'll just pick it up again. Once you know I this this gets. uh I get over with my ankle issues.

C

D

D

Did you say that you're typing, with your ankle, that you needed your ankle to type? Is that what you said.

D

Wow ouch did you yeah? uh Hopefully it's not bad. Are we talking about a sprain? Are we talking about it.

C

Like yeah, it's tissues and muscles.

D

Yeah things bending the way that they weren't meant to bend. Okay.

C

uh Yeah things bend that way once and no they're, not bending at all.

C

D

Okay, so uh aditya, that sounds like part of what hershey's saying kind of um aligns with part of your experiences. Those experiences you're referring to aditia. Are you referring to the catacota environment, specifically yeah.

A

Yeah yeah yeah. It was a category that docker thing I had. I encountered that.

D

Yeah got you: okay, not the persistent volume stuff, but a different.

A

Yes yeah, but you know, while experimenting with different ubuntu images and all that docker thing that came up irish, I gotta ask.

D

Do we do we need net like it comes into the picture because of the way in which the github actions themselves work? Yes, yes,.

C

Exactly exactly that, uh so it installs some binaries and like a dependency of those binaries, the dot, get dot net core library and that works uh like if you, when you see the chart for which versions it supports and which it doesn't. So that has an issue with 20.10 ubuntu, but that works on that works. Fine on 21.4.

C

uh So that's that.

D

It is okay, uh I'm gonna ask some questions that I hope are helpful. They might, they might be misguided, but to the extent that or yeah to the extent that we have a a bastion server. Like a you know, a man, a node dedicated to just being a management server um did do all nodes need to be like. Is this a management server uh concern, or is this a concern of every um kubernetes, every would-be kubernetes node.

C

I I don't think this should be a concern of every kubernetes node, it's just the it's just the machine, I think which is running the action itself. This is what I feel right now and I don't feel that they should. You know, cascade into multiple nodes, uh the other kubernetes nodes. If we end up making them so got it.

D

So we don't want to go so far as to like try to run windows on the management.

C

Yeah so I mean I went through it, I mean, because that was I searched. Basically, I just looked it up. I looked up an error of which came up in 20.10, which redirected me to everything that you know. uh 20.10 is not supported, and then you need 20., because these these libraries are not supported or not, uh you know, are not maintained for these versions, so I mean I.

C

I still think that we should be okay with just the uh let's say the bare metal or whatever is running that so uh regarding uh regarding sort of automation and uh to that end, I I actually had two ideas.

C

Maybe if we wanted to uh keep on running it on the bare metal itself, maybe something like a uh you know, something that pushes uh let's say an ansible could help or give or if you do not want to use that, or you know incorporate that, uh maybe we could just uh containerize that entire flow and go ahead and make a docker container, which runs everything we handle, try and handle the permissions in the first step itself, so that we don't get bothered by those later.

D

Make sense yeah, it sounds like there could be any number of um approaches and.

C

Well, yeah, I think jen and I will try and you know see what both of us are comfortable and know how we can solve this. So we can do that.

D

Is there any one? Okay good? This is probably a good opportunity for now that we've kind of described the this particular initiative and some of the tech involved.

D

um If there is a good time for hersh to call for volunteers or those that think they might be able to be impactful here as well, so think about it, um contact hersh, if you think you can.

B

Okay and here's like, if you can like you, can open up issues on the measuring smp action so that, like people, can look into what what they ask is and uh volunteer.

B

All right so uh mario just joined and uh our next set of topics is revolve, revolves around testing so pusha. Let's start with uh the test coverage for v0.6. Could you give an overview of what it is now.

A

Okay, the current code coverage for mostly ctl is right now 33.40, which is as like almost same as before, because it is kind of like very less people are working on it right now. I think, as we move forward for 1.7, like after dot 6 release, we should like call out more people to work on it, because code is just increasing and increasing and tests are not being returned okay, so there is a whole list of things to do like unit tests are being written.

A

We also need to write many integration tests because um functions like integration functions are also increasing in measuring ctrl nowadays, other than that, we were planning to write some end to end e2e test for measuring ctf and I'm not sure how we will write them like running machine measuring behind the scene and actually running mesa residual commands, so that kind of testing we have to think about how we will execute it like how we are going to make sure that whatever is being printed on console is the one like is the thing that should be printed.

A

That's the one thing I need to figure out. We need to figure out and yeah rest of the thing like testing stuff will be sure, like mario will talk about that. He was working on setting up workflows.

B

uh Mario, are you here, uh would you like to talk.

E

B

E

I'm here and yeah, okay, so the subject is: is it for measuring ctl end-to-end test or for.

A

Your silver and measuring ui, no.

E

Okay, good sorry, I I thought I heard I should still thanks. uh Well yeah. um I was having some issues so uh still need to uh what I'm going to do. Is I'm going to grab the latest uh of the master branch and uh see if they're fixed, uh because I did see the thread we were talking about, but um a ashish tiwari mentioned that there shouldn't be any issues.

E

uh The approach suggested was to grab like add a weight to the to the github action, um but I don't know I was seeing some issues locally, so I was a bit reluctant to um to follow that path because of that um actually um yeah and about the coverage. uh Just uh is anyone able to look at the current coverage like um the one in the master branch, or actually this one like? Actually I'm not that familiar so so?

E

How would we see the coverage with this specific, uh a aha code golf, but isn't that in the branch itself I mean the pull request? I I just I kind of like have seen that there's also some kind of um coverage results being output problem. Is that like this vr is huge? You know like? Has it's been open for.

D

Yeah good time, not sure if this answers the question, but there is, there is a code cov.com or codecov.io or whatever it is um yeah.

D

Okay, and so I don't know if this is answering your question.

A

Okay, so I'll add a point right now: code code is only calculating for these three folders. We.

E

A

To we have to set it up for ui folders ui.

A

Then it will calculate for javascript files.

B

uh Is there a reason like why we are not we haven't done it.

E

Requires some setup right.

A

It will require some setup.

D

Also, I don't yeah, I guess we are running a few tests today, fairly nominal.

E

Yeah there is uh this uh cypress plugin for the coverage, but uh I think it only runs on on the master branch and this uh whenever pr is merged and also besides, that not sure if that those results are collected. So I do think we would need to allocate some time to work on that.

B

Can you help track this.

A

Yeah I'll open an issue to like create change the way code coverage is being calculated and see. If how can I improve that.

B

Thank you and uh another thing that uh we can discuss right now is the proposed test plan so uh a while back. We decided that we should consolidate what tests we should write and uh how we should structure the test so that uh we kind of know what s will be run with every pr with every release or every manual run. So we had a proposal on the master test tragedy document.

B

I think the link is in the chat, so there is a proposed uh list of automated tests that we are planning to run, and I am and mario had some comments on on it here as well. So, uh mario, like, would you like to articulate it a bit more and share your thoughts.

E

uh Yeah, it's been some time, um so uh there's one some comments over there. Can you just uh scroll down a bit yeah.

E

Oh yeah about the linting, so it's uh I think that's taken care of like now. It's a set as a static check, so I think that's good uh making the change. Okay, yep! Thanks! uh Let's keep going measuring. I feel.

E

Okay. uh Let me recall why I wrote that okay, it's added. Okay, that's taken care of wow! So such a productive meeting right.

E

Let me see our new proposals.

E

um Well, everything looks good, then I was just thinking that.

E

E

Are we planning on working on a adding, so I know there's a restructuring, ui branch, a that would add certain component tests to the to measure a ui right.

B

Yeah, I think it should.

E

I'm not sure does anyone know about that? Okay, we're just wondering that, of course we want to do end-to-end tests and we're working on them. There's some setbacks, but I'm hoping to get them solved this month. uh But what about uh a component test?

E

You know I just feel like there are several things that we could be catching way earlier if, if our components have uh tests that can run against a component and really do like those simple checks, so so we could get that uh way faster than an end-to-end test and it it's kind of there's a challenge because usually uh not everyone has that much context about mocking react.

E

But it's uh it's really simple using a just mocking features, uh so I think it makes sense just to get that um that's solved like enable more contributors to to add those kind of tests. You know.

B

So yep, so that would be in the feature branch right that the ui restructuring brand.

E

B

Okay, yeah yep. That makes sense like I think we can uh talk to people working on that initiative and uh maybe we.

E

Could create a blog? We could even create a blog to go over like um some basic tests added to a component and just a have that as a reference for contributors.

B

Yep yep, that sounds good. That sounds.

E

Good mm-hmm, I think I I mean I that's all I have you know I want to get those endurance done to to. uh You know work on some other things you know have have that ready and focus on what's next, I think I have performance a performance management thing in line so.

B

And this is the pr you have right now right.

E

Yeah, if you go to that uh failing one yeah, that's the pure. Can you go there's a failing check. Sorry in the end.

E

So that's the current failure and it's.

E

I don't know you: do you see the errors? I don't think that's due to some to some, um like native components, not being ready in the kubernetes cluster, because if that be the case, wouldn't the the filter there's three tests. If you hover up the filter, one is passing and it's adding a filter successfully and or maybe I'm wrong and there's a difference between a you know what what pots need to be ready and maybe the filter filtering deployments, don't need them.

B

Yeah, it is likely that that error and uh yeah- maybe that is the case- but here it is showing that the the on y filter that doesn't match, but the filter is successfully deployed. So.

E

The filter, no, no, the filter is working, but so.

B

There's three tests so the failures.

E

It for the pattern and the apps right.

B

Yep so like like yeah, I think like from based on what this error we might expect the filter to fail as well, because the one-way filter that particular resource is missing.

E

Oh okay, unless it's depending on a different! uh uh What's that called, I always will forget the word uh on a different. uh Like version of that same.

E

um Kubernetes object.

D

E

Is that possible, or not? I don't know so so that you're saying that this is a fast, the filter deployment passing it's it's a false positive. It should fail right as well.

B

Yeah, it might feel like I I'm not sure what what filter we are applying and what patterns we are applying and what the environment looks like. So maybe I can try to replicate that particular test locally and like see if it is working.

E

Oh yeah that'd be great, yes, so that we can um like I'm using kubernetes. I'm sorry docker desktop kubernetes and I also see like some similar failures and so that'd be helpful. Just to uh have that run, I think the simplest way to replicate it would be just to uh pull that fork locally. I mean like through github, cli right and just uh just run that test. It already has the the files.

E

Because it's using book info and the other one for I can't recall it should be over here book info. It's failing.

E

Anyways yeah I'll just take that offline just to like not cut get all the meeting time.

B

Yep uh yeah yeah, like I, I'm not sure, like what the application is but like we can take it offline and maybe have ashley's look, look at it as well and.

D

Yeah, actually, now that I think, if it's starting to interrupt now that I think of it, if it's filter related, it's no doubt image hub but yeah to what mario had said is that yeah checking out that branch running those tests? Look whether you literally run the exact same test using the cyprus?

D

Excuse me, cyber cli, that's ideal, but also just running those tests. As a user in a mastery of deployment is yep.

E

Yeah, that's the important part.

D

And uh yeah: don't don't let ashish get away with hand waving? Oh, it works.

E

Maybe it's working on his side. You know there could be some differences that in his deployment of mercury and mine- and I think that the important thing is to you know find out what's happening and I even shared. If you go to the to the measuring ci uh thread, I did share that pretty much. The same issue happens uh like there's uh an istio issue, locked uh no matches for gateway and virtual service. So it's the same like uh v1, alpha 3 is still so. I'm not sure like.

E

Maybe there's some kind of setup that needs to be done when enabling the istio adapter, maybe not.

E

um Oh yeah, sorry it's in in the ci threat and uh in slack, so let me just share that sorry, just a sec.

E

Here, oh wrong button, just share it in the chat once this yep. Can you guys see it it's in the chat so.

E

Actually, let me just share the.

D

Yeah it's one of those things. It is um yeah.

E

There's things.

D

I think earlier to join this call, it'd be really helpful to have uh yeah or to have the to him. Show it in his environment running there's a couple of no doubt there's a couple issues, and I don't the difference might be that he has little magical leprechauns running around in his machine that are kind of fixing things on the fly.

D

E

Yeah, hopefully,.

D

E

Get his machine up, yeah, github runner,.

D

Yeah he can send one of those little leprechauns my way, but yeah.

F

D

Couple he he's, uh I will speak with him to make sure that he connects and goes through this thoroughly. We we um yeah, we wouldn't do need for that, pr to merge or a subset of it just to be able to you.

E

Know make some progress right.

D

E

All right I'll do that way.

D

The filters one to your to your credit or to not this isn't about you necessarily, but I mean to the credit of that particular piece of work. Filters are quite advanced and and fairly new, and so um to be expected. There might be a few hiccups around filters.

E

Yeah it's in alpha 3, so I mean it's something like very under continuous changes right. I would assume.

D

E

All right thanks.

B

All right, uh so, let's move on to the last topic for today, so I I just wanted to bring up the the work we are doing on the get nighthawk pipeline so get night talk.

B

E

To interrupt, I I have I've had this question for some time, so when a user deploys any kind of pattern or app uh and then he tries to do it one more time like is that item potent like I mean, would it cause issues for measuring to uh handle that, like, oh, like, oh, you can't do that that's already created it would update the existing uh deployment per se right.

B

Yeah, I I guess it should like uh it.

E

Should it should cause no issues like.

B

Yeah, it shouldn't cause issues.

E

And what if the user wants to like delete it like the the measuring ui, is currently not exposing things to to to get that down right. Only through the ctl.

B

uh I think, like we have ex exposed uh delete, delete button in measuring patterns.

E

Oh, but it would delete just uh the file, not the deployment of that right.

D

I think so maybe.

E

That's not that.

F

E

Go ahead at least.

F

E

No yeah, no, I think you're right.

F

Actually, I think, we're I think, the delete button that we have refers to the pattern file.

D

Deployment yeah so we're we're.

F

D

No not to me anyway, I mean yeah with that. We need an undeploy or a d provision yeah and uh yeah or like.

E

D

Or do we probably probably not like because it.

D

Because there's some design that needs to go there, the commands are item, potent like it is, or measures designed such that um this is leveraging.

E

Existing commands right actually through the measuring ctl. I was able to delete it. It's just like a bit of a household like oh, what was that file and then I need to look for it and then just book the command. So it's nice to have. But it's not that critical anyways.

D

B

E

It sorry thanks.

B

Yep, uh so so what I was mentioning was the get nitro project and uh how we are trying to move uh nighthawk and other performance, benchmarking tools out of measuring to measure perf container, and we have a pull request open to to slowly build towards this change, and uh we also have a proposal. You.

E

Mean a container uh like a pot, or you mean, like a separate repository.

B

uh Both so the measuring puff container will run outside of measuring and it will be, it will be stored in a separate repo and measuring will measure will actually communicate with this container over grpc like it communicates with other adapters, so the adapters are adapters and machinery perf together will will now be called measuring components, as these are an extensible and extensible component to measure so yeah and and they communicate over grpc and the like similar to the adapters. The measuring perf container will register its capability to measure server and mesherie can work with it.

B

So a step towards that is to is to basically publish the get night talk server, binary separately in a docker image.

B

So I was working on adding that uh ci process to to to this get knight talk repo, and uh I I just wanted to put put this out there that maybe like we might need to create a new docker image in in docker hub that just stores this network service, or maybe we can have a separate tag on the existing docker image.

B

B

B

Oh, I don't know.

B

So I I I published the binaries uh yesterday, so this tag contains uh all all of nitrox binaries packaged into a file. So night talk has a nitoc server, binary night offline, binary, an output, transform binary and an adaptive load control binary, but uh measuring to use it within measuring buff. We only need the network server binary the network service binary. So what I'm proposing is.

B

Maybe we can add a new tag here that just that just has the network service binary and we can use it within measurable and get night talk can still remain, uh can still have the the original docker image where people can come in and use it so yeah so get night. Talk will remain the fastest way for users to get started with network, and this separate tag will uh will work for work for our internal users, and lee mentions in the comment that a new uh tag will work and yeah yep.

B

There is a measuring puff repo, so the so, while building the measuring puff container. I think we can pull in the put in the docker image with this particular tag and uh have the uh the network service binary within it.

B

uh uh Actually uh lee, like I think, from the discussions with uh sin sin and us we will have uh the get knight talk. Repo will remain remain as it is it will, it will have the grpc client steps and the binaries and the binary is published as assets as well, but the message of uh repo will be separate and it will be what we using.

D

Oh, oh yeah makes sense yeah. Let me jump in there. um The additional yeah.

E

D

Sleek the thing that you described about an additional tag that that's nice, um we do that in a couple of a couple of the docker hub, repos were publishing a container and, depending upon what tag you get you know, you'll get you know, potentially considering considerably different artifact you're, just you're identifying the notion of that. We could just have another personality of the nighthawk one.

D

That's got everything one that's slimmed down, and that makes perfect sense with the part that I got confused on was so there's also a docker hub repo for the the forthcoming new component for measuring perf. That might.

E

D

News to you, and so that might affect your thinking, yep, okay and you that still doesn't mean that you can't do what you're doing right now is just have a multi-personality.

D

You know, multi-tagged um get nighthawk repo, that's still you know potentially helpful and the the bit that that probably didn't resonate was that it's the way that you described it made it sound like you were gonna. We were on a path to continue publishing two containers and images into one which- and that was just the way that you were saying it, but but yeah so long as we're um for publishing measuring perf as an individual component.

D

um It has its own github repo. It has its own docker hub. Repo measuring perf benefits from the output of the binary output of the get nighthawk builds, not necessarily it's not a consumer of the get nighthawk container images, though.

D

Because then, you'd have you'd be trying to combine two container images into one unless you're saying you're going to use okay, unless maybe I misunderstood, what you're saying is that you would the base image for the measuring perf container would be the nighthawk, the nighthawk flavor? That only has the associated binary.

D

Since we own both of those like it's probably more convoluted than is necessary, it's probably easier for measuring perf to be treated as any other component, any other measuring component. So it would have basically the exact same dock, very similar docker file to any mastery adapter the same base, image that way, security and economies of scope all apply to any component and that we're not saying well. We had to publish the get nighthawk image, as this c plus plus this thing had c plus kernel headers, so that it would work and then so that then.

D

Yeah anyways, since we own them both we don't have to go through a dependency. We don't have to inject a dependency, that's probably not necessary.

B

Oh yeah, I'm not quite sure like how we how we do that. Currently, like we use the do, we use the docker image the get nitrogen image within measuring right now like how do.

D

B

Test with okay, it's.

D

Just a binary on that in the measuring image. Oh.

B

B

Keep then, then what I said I might need to look into this again and.

E

Revisit yeah, but these two serve different purposes right.

D

Yeah, what one is um the get nighthawk image is. It is just loaded.

E

They're load generate.

D

Without any, um without a hook into a higher level intelligence like a management plane that can schedule those and run it and drive it and do distributed performance tests, that kind of thing.

B

Yep so yeah, I will look into look into it and get back and update my thinking process and maybe like make a new proposal on slack uh tomorrow.

E

All right so then the intent is that future versions of measuring mercury would not have uh the load generator or they would, but they would just reuse like that image or is, or is that totally not true like I'm? Just thinking like I have measuring bx and then I'm upgrading to b y like is there any kind of change that.

B

Like the the thought process was uh one of the reasons why we decided to split split night talk out was the the image size of measuring. So even if you are not using measuring to run performance test, these come packaged with the the machinery docker made. So user might not even use it, but it still contributes to the image size and another reason.

E

B

To run distributed performance test, so uh we are kind of moving moving towards uh measuring abstracting out the the load generator it uses underneath. So, instead of having the user to user configure which load regenerator he wants, we will basically.

E

B

To abstract it out and run tests with with night talk, and we can run these tests in a distributed fashion, having multiple instances of the measuring perf container running and sending load to your workloads. So it.

E

Means future versions of measuring sorry to interrupt. I would not have this enabled by default and it wouldn't even be part of the deployment right.

B

Yep so yeah, so it would be similar to how you deploy adapters so.

E

B

To work with istio, you should deploy an sdo adapter and I actually measuring ctl just.

E

Enable it in this yeah.

B

Okay, I think this support uh recently landed in measuring setting, and you can be- I mean like it- it is. It is set up to accommodate this change. I guess.

E

E

B

B

So before we uh before, we close this call today uh does anyone else have anything else to discuss or any comments on the things we have discussed.

B

And pushed this does make sense right like having it as a separate component and.

A

It makes sense to me like having it separate like a nighthawk, will serve different purpose, and we are focusing that people use nitro differently.

A

Like a public organization or private organization, um we want them to use nitro and I talk binary differently and we want to use our machine for ourselves differently.

D

B

All right, I think, that's all we have for today. uh Thank you, everyone for joining. I know it's a little late for uh at least most of us. So thanks again for joining, and I think this was a productive meeting and there there are a lot of action items to take away from this meeting, and if you are joining on the friday's community call, we will see you there, or else uh we will we'll meet back in two weeks with the mashery build and release meeting so see you folks, bye, bye,.

E

Thanks all bye.

C

E