Kubernetes SIG K8s Infra, 13 Sep 2023

Previous Meeting

Next Meeting

⏯

youtube image

►

From YouTube: [SIG K8s-Infra] Bi-Weekly meeting for 20230913

Description

[SIG K8s-Infra] Bi-Weekly meeting for 20230913

A

Hi, everyone welcome to kitchen for meeting. We have September 13 and this is the bi-weekly meeting as usual. Please be mindful of anyone in this call, so this car will be recorded and Publishing YouTube, so be careful about whatever you say and be kind to. Anyone in this call, so uh I don't see any new people on this call. So I'm just gonna go for the building with you.

A

uh I'll go on gcp.

A

I, don't see well, um I, think that's more like a communication for.

A

A

I would say it's more like a reverse of a cost. Consumption related to bigquery, so in order to I, would say duplicate the entry cloud provider in kka.

A

uh We revert the approach of the periodic project doing three days or collecting the data and pushing to bigquery that Supply I would say an upstick in the cost required I. Think that's fine, but just like we should kind of monitor the this.

A

A

It's more yeah James has full context of the situation, so I think we just monitor this. Just put it put it. There.

B

So when we did the change for the um entry cloud provider, switching the feature Flags uh to like force everybody to acknowledge that uh they they want to use the entry cloud provider, even though it is going to go away, um then we had to go, adjust all the ca jobs to adapt to it uh and like we had to monitor all the ca jobs to make sure that they are still functioning.

B

uh You know, so we turned on the bigquery um jobs um to run more frequently, uh I think I'm, almost done probably Monday or Friday I'm gonna go and switch it back to uh once a day. um You know right now. It's running every four hours, yeah I'm gonna switch it back to once a day.

A

Okay, I just want to put that basically I think that's like the only change. I would say: I noticed over the last few weeks on gcp, so I'm not really worried about that. I think we are in front of credit to cover that takes that spend so.

B

I think that's fine. uh Has the yeah no never mind the the packaging uh new URLs thing doesn't show up here, because that's uh Google infrastructure uh so never mind. Sorry, yeah.

A

But um basically James posted a graph about basically bandwidth consumption. So over the last month we I think we capture half of the traffic and it's kind of expressing like this amount we spent for the entire month of August. So we we spend 42k, uh which is when I checked on the platform. Fuzzy platform you can check it or I can give you a while to share.

A

We I think we bandwidth from fastly was 60 630, 35 terabytes, which is like not, which is a good thing, but I think we need to basically be patient and see how things happening. Once we change the GCS bucket. We might basically see a shift because I think I was discussing with sick release and yesterday about switching from kubernetes is better to commit your own one. So we can see change we stoppage there. We leave that thing.

A

We don't need, and we shift that so we can see bandwave changing but at least have full ownership on where we push the binaries.

A

That see, do we have no ticker by the way can we have no ticket? Please.

C

Oh and stuff, where I can.

A

Okay, thank you.

A

Is the if I check I will say.

A

Last month,.

A

Yeah I think we are around the same monthly cost, so I think I don't see any specific change.

A

Yeah, if I try to compare that to I would say.

D

A

Yeah I don't really see that with Delta of 1K, so I don't really see that as a huge change, so I think we find them on.

A

Oh where's, that.

A

Yeah I don't really see anything specifically regarding gcp.

A

Is there like anything regarding juicy.

A

Oh wow, let's see, let's get there.

A

Okay, AWS: uh let's see this.

E

A

So for I think we can go. We can try to compare.

A

A

Would you like to see things.

A

The registry Mark is mostly the same. Brow didn't really change over time.

A

uh We see changing yeah I think that the one changes like we now have like tests running in the scale account.

A

And by the way, James, we I think we'll blow up 20K this weekend.

B

In the weekend, yeah until we switch the janitor on yeah.

A

So from yeah it was like from I think Friday to Friday to until we switch we put the channel it was like. Then we spend 20K because there were like 3 000 instance running or.

B

A

B

A

B

Good good and bad.

A

It's a good thing: I mean like yeah. We now have like it's been for the scale. um I suspect we're gonna, see more of that, because I see we just I, see pull requests where we just increase the number of nodes and all of bought for testing. So yeah right.

B

So just a update for the folks here uh we started with thousand nodes, we went to 2000 and then three thousand and then from 3000 to 4200.

B

uh along the way uh we had, we had to adjust uh quota limits and uh API throttle limits, and things like that, so we were working with AWS support on that um in like some internal AWS folks as well, so we are at 4200 um and we are maxing out uh the vcpus that uh we've been allocated. So we've requested one more increase from 8 000 vcpus to like 1600 bcpus, so uh that should cover us for running at least one CI job with 5000 nodes.

B

uh We are also trying to go from a large instance to a medium instance, so that we can run two CI jobs, one periodic and one you know for pre-summit testing. um So you know so we are doing all that uh I think we will start using more regions.

B

um You know running different jobs in different regions, so uh there is, uh the cost is going to increase um so once we have it stabilized at 5K. I think we're going to stop there for 129, uh because 5K is what what we usually test on the gcp side as well. uh But then, after you know, we can decide what to do at Chicago, um so um I I did you know.

B

I did a CF small, tiny cfp for the contributor Summit, so we can all sit down together and plan out like what do we want to do uh for 130 and Beyond? Okay,.

A

Okay sounds good.

C

Just crazy to think 130.

C

I still remember trying to deploy the pre-1.0 stuff back at uh umich.

A

So the playground account this is Microsoft Dimension yeah.

B

E

B

Answer that one, it is not a playground anymore, it is, it has a lot of jobs both on the epe side, as well as the node side. So I think we need to drop the word playground from the name, because it is, you know, being used for release informing job as well.

A

Can we can we move that to what are we running into a test of unit tests.

B

uh It's not on a unit test, it's running uh essentially Cube test2, ec2 and provider testing. For um you know from those repositories we are standing up different jobs uh for testing signode, as well as, uh like yesterday, I added Cloud product AWS as well, so that is running on that account as well.

A

Can we adjust your WhatsApp to use bus, basically, the account the bus goes out.

B

Yeah boss course is difficult at this point: I'm still investigating it. uh The problem is, there is a role uh that is needed and like I'm, trying to figure out if there is a way to not require a role at this point, so until then I think we're going to continue.

B

So the same thing is true for one more uh set of uh one more account that we, you know we were just talking about before the call. um There is an account that is being used, um which is which is which was set up by an Amazon employee a long time ago uh and he's no longer with the company, and it is uh the it is not very uh high volume or anything like that.

B

It is like a 1K per uh per year, but it is running like the AWS control manager, load balancer, the EVS CSI driver FSX, CSI driver, so he's running jobs from all those uh satellite um repositories that we have for AWS. So we need to switch them over to uh an official account uh because we don't know what will happen to that account uh in the longer term.

E

Yeah I want to add one more thing to that. um So yeah um poscus isn't actually a good approach because we've got a different set of tests right. So we run no DT test is one set of tests that we run and then there's the cloud provider AWS, and then these other extra pieces of projects that are test against AWS right. um They can all live in a single account where each work each group of projects occupies a single region with reasonable quotas. It's on there yeah.

B

I I'm not ruling Bosco's out, but you know the the way they are right now. I, don't want to add Bosco's and make our lives more difficult than it is right. Now.

E

Yeah I agree, and the other thing that Boston doesn't support is like I want to use one Google Cloud project or when I use account, but I want to shot it into regions, because you can do that. um Allows you to use one account or one good card project for a lot of things which you can't do today. Yeah.

B

Again, let's defer those things for, for you know later, uh I will open up an issue uh in one of the repositories to uh you know, look at Bosco's uh migration, uh you know of different uh jobs, but you know it's not a high priority at this point, given the sheer amount of other things that we need to do for the rest of the jobs.

A

Yeah I think I think the priorities right now is our full ownership of the things we want to use inside community and we can figure out the organization later yeah. As long, we are aware of basically where things are right.

B

E

Are the accounts we use? These are different regions. We test different sub-projects in, um and this is what the quotas are.

B

Mama, the one of the problem with not using boss course is that you know the the CI job gets really tied down to this specific account and we start doing like special things, and then we don't know how to migrate. The job to you know another account.

E

I, don't think that's true.

E

B

Now so, let's uh question here about, like the playground, account it's no longer playground right, Muhammad, so.

E

Yeah, let's rename it and make a new playground account yeah.

A

uh Renaming is not possible, you have to shut down the icon, that's a problem. With those accounts. You can rename accounts right accounts. You have to like, basically when you, when you try to do that, it's gonna try to destroy your account.

E

In the console and then purchase the state later and careful, no.

A

You in in the console you change the name yeah what's happening is like in the console. We change the name, but the ID is.

E

Not changing, we don't want to change the ID, though it's fine, okay,.

D

A

Okay, yeah I see why it's Young.

B

Yeah, so when somebody is looking at this cost, you know like what are these folks doing in a playground account right.

A

Okay, I will try to do that.

E

Later uh there was an open issue to actually putting terraform all the things we did by hand in that account. um There's quite a few things that we did um I think doing things like that will ease concerns about migrating things to different kind of feed through in the future.

A

Yeah, you have I think you both of you have have admin of the account so.

D

A

Have a S3 bracket when you put everything into your phone I, don't I, don't have a specific concern about how you want to handle that as long as like inside a remote, State accessible, that's fine, I, don't I! Think I! Think for the moment. I don't want to centralize remote state for terraform, because we don't have SSO.

A

We don't have SSO. So it's going to be painful to deal with all of this. So currently we it's fine. We have each account, has its own history, pocket for terraform, state posting and that's fine yeah.

E

The book The, the state bucket is an important important thing is the things we did are written in, terrible that we can clean that up. There.

A

So it's just it's just a matter of doing an important version.

E

Yeah, we'll we'll get it done as soon as possible.

A

A

University anything about bidding review.

B

um No, nothing nothing more, rather than that I.

A

Have one thing about the best account, but I wanted to talk to that it will open conversation, so we can move forward with that conversation right.

B

um So, are we still around 1 million for the year uh from the.

A

Last time, depending on how the scale tests are running, of.

B

Course uh it's September 15th, October, November December, so two and a half months: okay, yeah.

A

Yeah I think we're gonna be assuming, assuming we add, more I, think it's really up to how we want to handle the skill test like basically, if we say we're, gonna have 5K notes test by December other aspect. We wish 1 million because there's a few things we, because we go down on this yeah, so I don't really expect. We wish one video but I. I honestly, don't know that would depend on basically there's.

B

A lot of as long as we don't blow out, um you know like we did last weekend, yeah.

A

Okay, conversation.

E

About reserved instances and things like that.

A

Very early because we don't even know we, we are not done with the migration of the project right and.

E

It does like the first.

A

Thing so yeah, I'm I think it's not the one thing. We need to talk after all the project by migrate to gcp and AWS, so we have nothing less inside Google infrastructure, and now we can talk about instance, resolution and once we have like a I would say, a borrowing currency regarding scaling tests. When we say okay, we have 5K there's no, we have 3K, there's no one kitchen so running smoothly over time, and we can like say: okay for this. Second, we're doing is test reservation.

E

Right, okay, so.

B

Bob, uh you know I think the gcp per year has also come down right. It's hovering around 2 million this year.

A

uh Less than that, less.

B

A

C

We're in a good place: okay, yeah.

B

Okay, yeah: let's continue to make sure that we have the same level of investment, um so we don't have to jump around and you know do more heroic efforts to move things around right play.

C

Hide the test job.

A

That's why I do I think I think the goal and thank you so much Ricky for traveling. This is have all the proud job inside the community secretary. If you can have that by end of the year, that would be great because we can have a projection the next year and say: okay, this is the one thing we have. This is running or like an entire month.

A

We can do a projection thanks here and say: okay I will reserving fans in database in JCP is not possible because of how gcp builds stuff, but another device, I think is possible. I have to talk with about this, but I think our goals should be trying to have every project we want inside the community infrastructure. It's based on that we can project for 2024.

B

Sounds good so.

C

Actually sorry Ricky.

F

Oh um so I was just gonna say we're we're starting.

C

F

Get close on the prow uh job migration we're starting to get close to uh the end of the jobs that can like easily move at the jobs that uh you.

C

Know basically have little.

F

To no requirements to move so.

C

F

If anybody has any thoughts on how to start addressing we've got a little over, you know, chaos is, is a big piece of that. I know you're working that, but you know outside of chaos with God I, think six or seven hundred jobs that um I still don't right now have like a path to to migrate or are hung up on some. Some.

D

F

Of either infrastructure or something that yeah is disqualifying it from moving right now, so I just want to see advantage of thoughts.

B

Yeah I think we need to go look at them and what buckets they've fallen for sure. But I do have another related question Ricky, which is uh when you're talking about the number of jobs you're talking about like number of definitions of the job in our yaml files or the actual number of jobs that run.

F

The job jobs and our yaml files, okay,.

B

Like do you have be able to pull information on like how many jobs actually run um and like what is the frequency of each job uh for every two week period, or something like that.

F

I, don't have anything plumped up to to do that outside of The Proud website. Okay,.

F

A

I did I think there's like there's a mutual way to proceed about this. The most simple way should be to basically say we migrate everything to gcp and based on that, we can look on how we can do my my I would say my basic role would be to say: let's migrate, everything any it will test to JCP.

B

Ricky I think you're more thinking about from gcp to EK uh ec2 uh AWS rather, but then I think what Arno is talking about is from Google infrastructure to gcp right.

F

B

Means adding the cluster and adding the default cluster and making sure then making sure that the job doesn't blow.

F

Right yeah definitely my.

B

F

My focus has been moving to eks, but um right.

B

This might be easier to choose what we are saying. This might be something that we could script and try. Yeah.

F

What one concern that I I maybe have of.

F

There there are there, are there are projects and and jobs that kind of, like you said, are not run very frequently right, like the Haskell clients, that's that hasn't been updated in probably a year it was. Is there any concern from your guys's part on? You know, trying to make sure we trigger these jobs? um If I.

B

F

B

You know make the change make the change uh like if it breaks I, let people fix it. Okay, you know um I'm resurrecting the uh the effort where we look at. uh You know things that are broken for a long period of time, um so like we'll catch it using that uh mechanism right uh and like we'll tell people hey if you're not using it, and it's broken all the time, then you know we're just gonna shut it down. uh If you don't fix it right, yeah.

F

B

And, like the uh you know, if you don't make a noise in the woods, did you act when the tree falls right like so it's the same concept right if nobody's watching it and it it doesn't matter to anybody. Then you know: why should we run it right.

A

Yeah I think we in the past. What we did was try to do some adding games. They did that last year was live stream campaigning to the six, so they basically take care of the project and because we we I think there's a firework. I can look into that. Where basically collect the number of day where job is failing and we are we so jobs failing for like two years and no one care about this right, so I think.

A

Ultimately, the world is like trying to force push the differences to do the migration, and if it's failing it's up to them, if it's matter, they will fix it right, but our expectations will not to do basically to try to move things.

A

I mean trying to push from to migrated product jobs, and we can refer it if, like yeah, it's critical, I, don't I, don't think any I would say master blocking or mastering for me are inside Google infrastructure anymore. So the rush should be straight for a while. So right, people, the only situation where I see a river to the Google infrastructure, is where those projects need specific credential or external asset yeah right yeah. If if they don't need that, we should just migrate.

F

Yeah I do I, do have a PR in uh ktio or sorry, an issue in uh Kate's IO that I've been waiting for some feedback on.

B

Let's make sure that every job has a cluster and by you know, if it doesn't have a cluster, then it should. You should add the default um cluster yeah that is owned by you, know Kate's infra, and then you know we'll see where, where Things fall right like now, who is the right time to do it? Because you know later, when we start hitting deadlines, then people are going to yell more at us, um so the earlier we do it in the 129 cycle, the better it's going to be. Okay,.

F

Yeah I can I can absolutely get default. Added to everything. I've got okay in place to find all that, so it should be pretty quick. Thank.

B

C

This is jumping to my item in the agenda, but it kind of relates to this um I've talked to like Ben and a few others about this, um but I think it's coming time where we actually set a policy that uh we will not execute. We will not use. Ci will not use external resources unless they are owned by the community into a point where we can support them um things like any of the Microsoft tests.

C

um This is also to kind of like serve as a forcing function to like hey it's time you bring people to the table. To do this, we can't just um you know hey. This is broken. Let's poke some internal person at Microsoft to fix, yeah and, with you know, a lot of things getting out of the the you know we're at a good pace for getting out of all the Google own specific stuff.

C

um I think we can. You know Define that going here and give a timeline where just like you need to be out of here or we're just going to stop testing.

B

uh Or publishing the testing.

C

B

And like making, uh actually people can do whatever they want in their opposed? We can't stop them from doing that, but you know they can't be release informing or release blocking from you know from a policy perspective right, like those other, you know you can't stop. uh So you know some specific group of people in a specific repository to like add CA jobs on whatever.

C

Yeah, if they use like a GitHub action or something like that and spun it up that way like we, we don't have significant control over that.

B

um So the knobs that we have is it can't be a release informing or a release blocking job right. That is what we can mandate and we can mandate that it's not going to show up and test grade. You can't I, you can't add it to test info. Basically right, um you know go nuts, but you know it just can't come to us so.

C

Ideally, honestly, I'd still like you know, sort of a global policy like it has to be some ugly I mean community-owned thing because, like all it takes, is an employee leaving and then suddenly things break absolutely.

B

Agree with you do you want to depend on drafting the policy? Would.

C

B

Do you want to take the pen on drafting a Google doc for the policy, so we can socialize it at uh kubecon and like make sure that the people uh hear from us directly uh rather than us, throwing a policy in their face as well right.

C

Yeah um this also, actually, we kind of want, like uh us, talking to Marco um I, think it was this week. It might have been last week about like setting a timeline to fully shut down apt.com.

C

um Potentially, to like announce that with the 129 yeah, the the next release um and and then give it like a year- and this was like okay- these are turned off on this date. You have yeah.

B

As you know, nobody listens to us until we shot picks off uh so anyway, we can.

C

Try uh you know I I can I can say you know internally that they're more than happy to shut it down whenever we want yeah.

A

Got it yeah, I, I, I, think I think, because we we don't want to basically spend the I would say the consensus is like we should draft a policy we need to embed in the chart of the seek, so we can clarify How, We Do ownership of the structure. Again, do you have that have a conversation about it in kubecon and try to put that in the chatter I think. That's the overall call in term modification.

A

I! Don't exactly know how we want to handle this because it's like sometimes it's sensitive, subject or a delicate subject, because you don't you don't necessarily want to shut down things, but you want to slowly make sure people shift away from the old infrastructure the we.

D

Can do brought.

A

Out we can ask you guys to do pronoun schedule, shutdown, but cut it out. Right away is not something it's on always the right thing to do so we can just slowly wait for people to.

B

C

Crow notes are okay, so yeah yeah, the the main thing like I've I've, talked to some of the various vendors and it's just been like. If there is something that they can a policy or something they can just like, hey look, this is gonna happen. We need to take action, yeah um that that is the that is the forcing function yeah.

B

Let's put that out, we can do the policy, we can do the comms, we are getting better at it. Yeah.

E

B

With regards to.

E

Vendors that don't give us anything I think Microsoft's the main one. um What other vendors IBM vendors to be used with.

A

E

A

Matter of communication, and if a vendor, we we give, we give a Global Communication about how we want to handle the different artificial asset we use and publish okay. Now it is up it's up to the vendor to adapt himself to what we Define.

A

It's not because, if render real product around kubernetes, they have reward and say they are basically have something to say about how we handle the different assets. We we basically have all policy is to ensure we maintain sustainability of the project long term, because we are very young project. We target 30-year life cycle like.

E

The difference, ambitious yeah.

A

I mean that's, I would say, that's I would say the the traditional for pencils a you can see. Yeah the products is successful when it's growing or there's like a commodity for other projects over a very long time. Right.

B

Okay, yeah successful right.

A

Now, but we don't exactly know, what's official gonna look like so we need to make sure we are successful for the next 20 years. Okay, as a vision, we shouldn't always think about. That's.

B

What I'm saying.

A

Whatever we're doing is like right now make sure we are sustainable as an open source project which is really really. uh If you don't treat. If you don't try draw some boundaries between you as an offensive project and different distribution.

B

Which basically means that we, the processes that we set in motion and the policies that right has to survive the people, the startups, the companies and so on. Right.

A

And so on, yeah.

D

I want to bring this to a level down for a moment. Speaking of what Bob said earlier about policy or community of a different structure, we actually tried to do something like that, then be initially started.

D

The jobs migration from default cluster to the eks cluster and the one let's say, conflict that we had is how these communities should be because there was uh different opinion so like that people who are the companies were going to the nature is essentially through cncf, rather the community disability, and this is something to pay attention uh when trying to draft a policy like how someone can help the kubernetes project and like what steps should be taken to um donate resource.

D

What we try to accomplish initially for that policy when creating it for your cash cluster, is that some minimal should be that you can the secrets infra can access the account research so that you can generate credential to see what's going on, but it doesn't really have to be donated in a way that you go through the CNC after donations process.

D

If you don't want uh I, have a quick link here for bottle growth back then, but like there were some disagreement, so that and like maybe, if you start from a Google Document, try to come up with something and to formalize this as part of Charter, as Anna said, I feel this might be helpful.

B

C

I'm going to drop that link in the docs notes.

A

Okay, uh I think we we have basically hey eyes for this, so I think we can move to or any comment because you have 15 minutes left. So you can move to next option.

A

Okay, I don't see anything so my next subject is about the cleanup of the organization, because uh I would like to start uh shut down the different account that was migrated from cncf to us, because some of the icons are either non-use or have resource doing nothing.

A

So I'm gonna open a Umbra issue and try to track down and try to shut down things like so from taking up a sub perspective. A shutdown of an account is 90 days operation. So at the moment you trigger that you have 90 days until the account is gone, so revert that action should be fine. If you, if somehow always notice, a critical Pro job is failing because the air conditioned down so I think.

A

Basically, this is more like open in different issue about that and try to investigate basically outline the account I want to shut down and shut out over time slowly. So we can buy end of the year. Have a cleanup of the organization because there's a lot of things we migrate, but no one ever used those things it just I think not Gateway is doing nothing and been there for like two years so I'm like interest to do that. Cleanup.

A

A

uh Any question.

A

Okay, next uh I think my question is also related to how we handle the oci artifact and Bob my good contributors add-on, because my question is related to Peter.

C

A

C

The like hosting.

A

Oci artifact inside GitHub content repo because I see so the idea is like uh for a long time. People were reaching out about this like where, if it's possible for the community to if it's possible for Community, to provide a platform where they can host oci artifacts, so basically not only control images but also M chart and so on. The oci specification evolved to the point like we can store non-commissioned.

F

A

So my question is how we want to handle this because I don't want uh I, don't want the suppression come to us and ask about this. So basically the privilege change through the registry, so I would like to be independent. So today you will get our service to all to basically all stay on yeah we packages.

C

Yeah we have not as a GitHub admins we've not talked about other artifacts. We have generally tried to push people to use our own stuff for images or to put publish the helm charts at um I'm blanking on the name of the thing, uh I think it's artifactor yeah.

E

Yeah yeah thanks.

C

um We haven't looked, we have we well, we haven't discussed anything other than that.

A

I did my my I wanted to get a conversation between kids and foreign, so I think the one thing is like because over time, obviously authorization on GitHub basically is like is evolving to the point we can deal it. We can I would say delegate, but there's also some kind of race become delegation to and Jerry Jones, basically I would say, not increase the surface attack because you can have an individual accidentally, releasing something with a high CPA and that's impact. That's impact.

A

The entire community, so I'm like I'm, like I, would like to have a conversation about this. Basically, how we become better at delegate, yeah.

C

Probably the best way is to to just create an issue in K org, and you know, tag us and uh keep on us uh regarding that discussion. We can also talk about it more at kubecon, too,.

A

Yeah, and also if it's I, don't want to say, talk to guitar and see what's doable and what's not doable, remember, I would say authorization perspective so can simplify, might be a future request to the entire GitHub platform or might be basically something we are not aware. So we can simplify delegation, but the idea is to stop basically, basically stop selling to sub-project to use the version 3 the project, because some of them are basically clad, Cloud, specific I'm, not sure we want to do that.

E

A quick question, though, I think there's one important question that you haven't answered. Yet: what's stopping us from using something like K promo, to publish oci artifacts, because you can store that stuff in artifact registry and.

A

Because we have a policy where we basically say the version is usual, should only host specific artifacts like the kubernetes components.

E

A

Or I think we have this the now we don't have a policy about how we handle suppression and I think this is where we need to.

E

Add a supporters could publish to registry.c.io and they do yeah.

C

There are definitely some that do.

D

Yes, there's a potential technical problem: there yeah.

A

A lot of sub-project published Contour images. Now do we want to be responsible for older oci artifacts inside the registry, I.

E

A

Yes, given that and you.

E

Can kick the GitHub conversation down to the future if our tooling supports it, which is very buggy I, think Marco can ask you a tooling.

D

uh Yeah, so I don't think that we really care about what's being published if it is also comfortable, it works for us. If you publish health chart, that's fine and the problematic part is that right now everything goes through the Google Cloud build. So you can't really push images on your own, like you can put it in the even with Pro. We usually run the Google Cloud build job, so that might be trickier to get working with him.

D

Whatever people want to post I can anticipate some problems on that side, because our pretty much doing what we have so far is made for images, but that would require some work. I think that might be one of the biggest blockers speaking.

A

Of that, but I would say, I would say it's uh it's a matter of technical implementation, because we don't need to use care of that. We can. At the moment we basically say we can have a different tooling that do the promotion, because all sorts of stroke will be the staging repo. We have a tooling exercise. Promotion to the private person doesn't need to be careful. It could be another tool, that's doable, but the problem is at the moment we say we use registry or any oci artifact.

A

We we increase the bandwidth for that, there's a cause problem yeah. No, it's not good, because there's a customer that we're gonna who's gonna pay for basically serving one of the most popular m sharp on the planet. Who's gonna pay for that.

A

E

It's a problem.

B

E

Currently, ship the nginx Ingress Hampshire, uh which I'm guessing is very popular.

A

The ancient links is basically uh through the race. They do promotion.

A

It's served through the bridge history, no.

E

I'm talking about the helm, chart itself.

A

It's a it's artifact if I, remember, correction, so they're like different, depending on basically what you, where you want to pick it. You can pick that from VMware. You can pick that for artifact, herb and or you can put pick that for basically even guitar. It's.

E

Really up, oh yeah I see what they're doing um the home chart the the girls from there.

A

Exactly directly I think GitHub can. Basically, if you apply there used to be a registry, a project that basically serve mshot directly for kids. So the the problem is not technical in the sense like we can, we basically cannot. We cannot cannot serve food. The problem is like the cost impact of it. If you become a central registry for the M shot of sub project, there's a cost increase the.

E

A

It's not something we just came out from it from a cost problem, so I don't want us to be in a situation where we basically say we become an alternative to artifact herb and we have to pay for that.

E

A

Sounds good right.

E

Thing um and then use that in your decision, Matrix for when you decide to allow people to publish things to get up, yeah.

D

Or not, there's one thing there that is.

F

D

Like we have to take a consideration that help charts are not disclosed in sizes, real container images right, so they are basically manifests. That's here, I'm going to apply so that is pretty much I, don't know how much exactly it can be, but I don't anticipate it to be or a few Mega ways. Maybe even that is a lot, uh and this is something where me we could potentially utilize CDN as well.

C

One other thing like uh thinking future proofing when it comes to shoving other things into oci.

C

um You know, Helm charts might be small, other things potentially not so much.

A

Exactly because it's not only on only about the M chart is basically any oci art if I would want to, because if we open the conversation about M sharp, that means we open the conversation about everything uh webass assembly extension does.

C

Artifact support sort of just whatever in terms of artifacts yeah,.

A

Yeah- and it could be basically just one- article I become 101 100, Mega megabytes. We have to serve and we have to pay for that. uh Yeah four.

B

Minutes left so uh can I ask you to open an issue with the discussion that that you have in you know you were talking about, uh uh because let me find you can make a decision right now. I I, don't even want to make a decision right now. There.

A

Was a there's, an open issue open a while ago by one of the sick, real estate? You can't have a conversation there.

B

Okay, uh let's cross link it to uh I.

C

Think that one was specifically about that one specific project, none overall right, yeah.

B

C

A

Can we can extend yeah I'll open an issue to extend that right right.

C

Yeah I'll just reference this one yeah.

B

As an example of one project, who wanted to do something.

A

Okay, so I will give you back here. Three minutes.

B

Thanks a lot, everyone bye.

E

All right, thank you.

C