Kubernetes SIG Node, 14 Mar 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Node 20230314

Description

SIG Node weekly meeting. Agenda and notes: https://docs.google.com/document/d/1Ne57gvidMEWXR70OxxnRkYquAoMpt56o75oZtg-OeBg/edit#heading=h.adoto8roitwq

GMT20230314-170537_Recording_1572x1120.mp4

A

Hello, hello, it's uh March, 14 2023! It's a signal. Weekly meeting welcome everybody I want to start this meeting with uh announcement that quad freeze is tomorrow um for 127. If you want your PR to be in 127, uh please work on it and make sure you have somebody who will review and approve it. It's.

B

A

5 PM Pacific for today today, thank you, Mark. You know better because you're, a dis manager but uh yeah. Thank you. uh Okay, it changes things for me a little bit but uh anyway, okay, um today code free, so if you haven't merged today and uh your back or fix is not critical enough, it's out so we have 200 PRS and we did amazing job emerging things. Thank you for all approvers who uh looked at numerous PRS.

A

Think of all reviewers who prepare this PRS, who are been approved, I wanted to uh just remind that there is still 41 PR, uh which, with lhtm label but without approved label.

A

um So if you want to so just a reminder for a programs that there's still something to approve, it doesn't mean that all of them are critical. I think some uh judgment needs to be applied, but uh to this secure queue of unapproved PRS that may need to be looked at. I also wanted to look at the issues. uh First start with critical origin issues for this, uh like typically priority critical origin. This was something that we have to fix in the cities.

A

So this uh we have three. One of them is separate. The topic for a little bit later in agenda I will leave it for for now. Then second one is. We have some failing tests.

A

It's very old issue, so I think it should write already One release in this priority critical origin. So we need to understand. I will see an urgent issue and last one is.

A

Yeah from Clayton uh static audience.

A

So it's also very old issue and Mark this critical, urgent.

C

Are you talking about the 116262.

A

No now I'm talking about this, uh a couple doesn't already started static Port when API server is down I have it on my screen.

C

Yeah yeah sorry I was looking at the wrong screen.

A

So this issue survived uh at least a few releases, and seal Market is critical. Origin I, don't know if Clayton is on the call right now.

C

This could be related to the issue I filed this morning, so I'll take a closer look. The short summary is that when you try and update the Manifest of a static pod in this example, you update the image it doesn't propagate.

C

I I know where the problem is I, don't want to make that fix so close to countries. It's dangerous. You've got to switch mirror opponent, part around I, wouldn't risk it.

A

Okay, so we need to clean up our critical agent um yeah, given like we have CI group meeting a couple of weeks. We've been cleaning up um important soon issues, so uh there are plenty important soon. We can take a look at this. We still have a lot. uh We just cleaned up many of them. uh We have 40 markets uh important soon.

A

um Some of them are for caps that were targeting this, at least so uh this is uh expected to have them, but some of them are long-standing, like stale issues that may need to be removed from this Milestone. So if you have time to review those, please do um we need to make sure that we only have something marked as important soon if it's really important soon. Otherwise, you just defeats the purpose.

A

Okay um and yeah, when I uh do you want to talk about block you post it in a chat or you want to go to the next item? uh An agenda.

C

um I'll, just briefly mention: what's going on at 116262, I think uh the we found another null pointer access and I fixed that, however I'm not able to verify it because I don't have the Repro environment that Clayton has uh or the words there in internal gke. With this couplet I tried different things. Hacking around can get to that. uh But what I uh there is a defensive fix, which probably is even the right fix at this point, given that we don't process updates for static pod manifests.

C

So what we do is if it's a static part. We skip calling uh the two functions: compute resize and do resize in the compute product rates, so um that fix the more I look at it. It seems right, given the bug I just posted, because uh if we're not handling updates for images, why should we you know not skip over static points? Why do you even want to call that function now? Kitten has some objections with I think he has some concerns, rather with whether it affects the existing static.

C

Pod image, update, I, think is referring to you, make an update to the Manifest file and then save it, and that should trigger a new part. I'll verify that it's triggering a pod delete ad in both the cases where this feature it is enabled and disabled.

C

That should I believe address it for this, but we can discuss that.

A

Yeah I, don't think it's a regression right. So it's it's not a new Behavior.

C

The null pointer is a regression, but once it's fixed, uh it is yeah. It should work the same way as it has been working for now. This particular bug which we are looking at has been there in 126 Master uh 126 release. So this is not a anything tired here is not regression.

A

Okay, yeah I, don't think uh something that is not regression should be fixed. Is that urgently yeah.

C

I I might have to go ask for an exception to give people more time to review this critical. uh Probably gonna raise a few days exception for this and Tim I. Think one thing that's not here is there is another PR, that's outstanding for Tim to look at. uh Let me just post it in the chat as well. Please add it to the doc there's so many things going on I'm just trying to get knock one things out. Look at this one.

C

That's whatever one week after this code, freeze author after the merge I'd be drinking beer. What had we missed all right, so this particular one is uh Tim asked for a follow-up changes to. um We want to rename the uh rename the variables for resize policy. This is not a risky change. I think it can go in. uh It's just I, don't think it's just the bandwidth. Tim has suggested some things. I made those changes, it's been sitting for a few days and um I wonder if we can get dimspan with on this one.

D

Do we want this one in 1.27.

C

I would prefer is.

A

It a name for API fuse yeah I would.

C

Prefer it I would strongly prefer it I, don't know how this will go if we were to do it after those photobuff stuff was all published, and you have to deal with that.

D

uh If he has this kind of the winning request, why are the first place approved the original implementation? This is this is this is faster next there's the discussing to rewards the orange ones and I, of course, I. Don't prefer to do so, but this is make the others either complexity to everyone.

A

Is no pointer right.

C

That is that one is that's the one which I think we want to merge for sure, because it does fix the issue without this.

E

Is that the state manager appear there, because I've merged I had proved that.

C

Yeah, thank you. Thank you for that I think I saw you approve that, and that looks like it's getting merged. I haven't checked, uh but I presume it will uh for.

E

The API one I know that we need uh I can't single-handedly approve that one uh I know it was Tim's desires to yeah.

C

E

C

E

Can chase him down that'd be great yeah.

C

I just bring him on Twitter, so hopefully he'll take a look at it and I picked him around the time when he walks his dog. So that's a good time to do. It it'll be in a good mood right.

C

A

Say it again, so we have this null pointer right, uh which, uh in Standalone mode.

C

This morning,.

E

A

C

It's a small, so that was uh just in case. You try to access the the state manager outside of the feature kit protection. Then you will hit null pointer to prevent to avoid that. That's more of a defensive! You know create a new Upstate manager to begin with and you'll, never even.

E

If there was another parent, that means.

A

Cool is that another note finder.

C

This this one, this one we're looking at yes, this is the one okay.

A

C

It panics further down the road uh when, for some reason, the the issue with this is that, uh even though the fix looks like it should fix it, I don't have a Repro and there is no I tried, different manifests and uh the.

C

um I think the root cause of this is really not wrong for whatever reason for static pod, uh the compute resize action decided that this pod needs to resize and uh we really need to get that particular many set of manifest that's going on in there and the fix that I made here is uh initially it looked like a defensive fix where I don't even invoke compute pod resize do pod resize for static Parts, but in light of this bug, I just created, it looks more like the right fix, because, if you're not gonna process updates like an image update or from the Manifest, why would you want to unnecessarily call compute pod resize to power resize at all.

E

When the feature gets off.

C

This is all happening in the future. This particular issue that null pointer is happening in the future grade is on for the static pod uh cubelet only mode. You.

E

C

It's on or off I'm sorry I couldn't hear it on on yeah, okay. So when it's on and you are in kublet only mode- and there is a static part- I- don't know which, what kind of manifest they used to, uh which triggered this and uh added the checks that I've added here. uh This nil checks should take care of it, because I saw one path, which was an existing code that I didn't really encounter ever before.

C

So when we fill out in the resource config for pods, then we fill out the structure uh depending on the kind of part, if it's best effort, if it's uh burstable burstable with CPU limits on CPU risks, requests only specified, it may give out some null pointers and we are accessing one of those.

C

The costac in the information, the cost Tech is not really helpful in telling which exact parameter was accessed. There uh really need to know I need a Repro for this to be sure, but this should fix it. The I think Jordan has a problem with uh the exclusion of static pods, and there are a couple of cases that I want to verify in there with the.

C

When you update the Manifest and then save it, we should be triggering a pod, remote and new bot ad for the Standalone couplet case and for uh the one with API server as well. I'll make sure that I go through those four cases without using debugger this time, so that it's traditional verification.

F

But I I have a kind of a naive question about in general, the static pods in general. Like I, is it has anybody ever talked about a more General way of managing static, pods or more like something like a I? Don't want to say it, but it Coop CTL edit of aesthetic of a of a you know, of the actual static pods on the kublet or.

C

We can't do that yeah what I used um I use my I use patch. Sorry.

D

Every time you see the you want to allow this vpa working on this static powder, thinking and comfortable I have to admit, because we, um the problem, is that your your source of the choose, the other statical part. It is that file, so we actually discourage people using that feature.

D

The reason it is really difficult to manage that the kind of the problem and as you cannot otherwise I, already approve your your appear, because I reviewed, but I couldn't figure out uh for this Standalone kubernetes how we are going to because we are missing in the standing Standalone kubernet cases kubernetes that particularly not registered to the API server. So in the app there's, no neural part that we created in the API server.

D

So the part of the administration is only at the node level defense, so so this is really difficult for for me to think about the connect, all the dots, how we are going to do. This only know the level of the defense to do the part admission. So that's why I connect the I'm still I, I'm I, read your fix and then go back to the original original implementation, which is the which is the hundreds files.

D

So I try to figure out how we are going to plan to support this static parts, and so this is why I haven't so so if we have the Mirror part, at least you could powder admission at the schedule hour. You basically say: oh kindness, is the part of the already occupied the resource right, so you, you start the rest of stuff that actually, but in here you basically could be over committed on the Node only node level, but that's also so so so.

D

This is why I'm not sure, because remote controller can do better job here and to adjust the static part, the resource usage and then make decision at the only Network know the level. So that's why connect I'm I'm still struggle otherwise I already? uh Yes, yes, I, agree with you director. This is why we talk about this many times.

D

We want to use the reserve static powder for the bootstrapping scenario, so this is exactly what we are doing here, but if we we allow this powerful feature, we see a feature to Dynamic update the static parts of the resource request, which is make me I have to think about more to make sure everything's coming, because we need to do out. So that's why I'm, okay, with your current lower Point fixed, but the the comment. What do you commonly see like this one will work with aesthetic part make me really nervous.

D

So that's why I have tried to go back to the.

C

Yeah I was thinking more from a user use case perspective. Let's say you configure a static board, some kind of a system critical part, but it's over provisioned by a lot- and you know you want to have vpa, manage that. uh Well, not I'm, not saying that we should do it for everything, but maybe for resources, and the right thing for now at this point to do is to block any updates to static paths, resources or anything else. The anything else part of it is taken care of by a bug in the kublet.

E

Epa should ignore, mirror pods, and we should definitely ignore them. On the qubit side, like the vpa is not the source of Truth yeah.

C

E

Pod definitions that the node um preparer is the source of Truth BPI, is only in the business of things that are managed by a control plane. Okay, what.

C

I'm wondering here is like.

E

Is there a test scenario, we're missing an ede that would have captured the flow that and Jordan identified, and then for these items we get the no pointer and maybe, as a part of a graduation of the future. We enrich these missing EDS in.

C

In the Ci or uh well in the SDI, there is I believe it's in the GK internal I. Don't have access to that. I can't even see the logs to make more sense of it. I have what Jordan gives me in the book. That's all I'm.

E

Just saying um as a part of graduation, it seems like we have a test Gap in what we all see in community ede, and maybe we just for these items that are coming up late mode. We'll get the the no pointer merge, but following work should probably be to enrich our ease where we're missing my.

A

Follow-Up is this I created this um PR to introduce tenderloin mode tests yeah in E3 node. Unfortunately, I didn't quote this I called another new pointer that we may also called, and it's already fixed, uh but I didn't catch this one and I wonder like I mean we can add uh this one and uh start improving coverage for Standalone mode and like documentation as well, because we like documentation, the question would be: uh is it okay level of technical debt that we like accepting feature in a milestone? So we we have this one.

A

We have a lot of follow-ups like I know. There are at least few GitHub issues to follow up with uh more fixes I'm trying to then. If this is uh one thing. I reviewed, though, is that everything is protected by feature gate, so it feels like um if we just disable feature gate. Everything works as before. So I think this is a reasonably safe assumption and I I looked at quote uh twice, so it should be fine.

A

But my question would be like this one: this Panic we don't, we don't know how to reproduce, but- and it may resurface again and we have other follow-ups. Is it? Are we comfortable with acceptances in a milestone or we need to discuss, reversing it back.

C

um Or this particular fix, or the entire vpa feature.

A

Feature I just want to make a decision and go forward.

C

I think it's safe, so I look at it from okay. This is a fairly big feature, and you know uh this particular the Standalone kublet case was uh kind of sitting in our blind spot.

C

I'm sorry I knocked it over, but if I know I would not have it's a internal GK I had no clue it doesn't show up in any of our CI as watching CI and I didn't find anything and I asked around and the dims and- and uh somebody else mentioned- oh sorry, we don't have any of these and um I don't feel that badly about it.

C

Frankly, uh besides this, uh these terminate, it would be nice if we had not had this at all, but um besides these two uh I I think it's uh hopeful and see I pretty much look unaffected, it's hard to say, because there are quite a few uh CI that were running red. So when they do that they mask potential issues, then this might bring in right. So.

D

So may I suggest this is this is just important story. You have another API change it proposed. Obviously we want their all go together right. So, in this release, I mean not to the.

B

D

All go together in Swedish I. Even don't think we have the panelists to reveal that one uh before today's code freeze. Can you file the extension exception.

C

D

Yeah uh yeah, just just one minute and also the the can you file that extension I think I will prove that extension and okay, so make sure give us a little bit the time to review your new API change. If we couldn't agree on the API change later, we have to reboot all those things for this feature. Yeah.

G

D

I, hopefully we could have all the fix for the Panic. We try to merge your panic, but that still Connect the Dots here, so you fix for standard mode and the other new appointments things uh and then try to allocate the time for your API change. Is that okay? So we basically still move forward. This is okay, please!

D

So then, we've moved still trying to move forward with this feature as the alpha feature in a 1.27, because it's been so long and also it's Alpha feature and the feature gate and everywhere. Maybe we need to double check that one but then, but if we couldn't get the API change in so obviously you also prefer.

D

So, let's make a feature: use is so hard.

G

Well, um don't.

E

Just I would really like to get this in this release. I'd like to move any mountain, we can to get extra time. We get to look at that API change that he personally requested. That's.

C

My understanding.

E

Like I have trouble holding everyone else um back on that one I guess uh with respect to like I I, would really really like a big plus one to uh if we can't get this get the exception process going. But the idea of reverting this and then putting it back actually is very probably personally painful for Renee and personally painful for me, who spent a lot of time trying to review it. um I'd rather be able to knock out where we have test gaps like some of the Cross component.

E

Integration flows that are being called out here, I find um not surprising, with an alpha level feature for any feature going forward. I'm actually surprised that this this particular flow, or this particular issue, was even discovered.

D

But the Children Therapy, what do you say? What do you propose that would have proposed is pretty much single, no.

E

I know I 100, don't agree on a girl. I just want to uh be real with everyone on like our emotional weight behind this, which is just like it would ping me on a personal level. Yes,.

D

I, don't see the really convincing pipes, but the problem is the API change. Actually it is and not to address at this moment. So so, unless you folks all think about okay, we can pump that API change. The new API change to 1.28, because I, don't I, don't think about any current issue we find is the Blocker we need to go. This is what I earlier said right so, but API change. This is why I ask of it. Do you want these ones in the 1.27, but then API change? Actually it is unknown to me.

D

So that's why if we couldn't consensus other API I also don't want to rush. Just everybody say: okay, I, don't care, they said yeah and then just let's just move. This is kind of what I want to say here.

C

Yeah I think that's, okay, the I think so Sergey has some context on that. He already discussed with Tim on renaming that particular variable. Why we do why we chose that and uh really it's a pretty safe change. uh I don't know if if we don't merge that in 127 and go to 128 and try to make that change, we have to do a bunch of extra work for protobuf. uh That seems, like you know, an unnecessary headache to take on. Given the nature of the change is pretty innocuous.

C

It's just naming X to X bar, that's, that's all. It does and, of course, adding the defaulting which uh which Tim wanted and all this came up in the last uh like a month or so so this renaming thing. uh So it's pretty. We still have to update the cap to fix the documentation in this case, uh I I think I'll I'll take an exception for both these issues. uh To give us more time and I would like to see it but yeah, it's okay with me.

C

If it doesn't make it we'll figure it out, I think people have changed. We named stuff after something has been shipped. Yes, it's probably a little bit more headache with photograph naming. You may have to lose some number there and go for anywhere. That's fine, yeah.

A

And I think team approved the original PR because he's equally emotionally engaged involved in this PR. So um it feels like there are so many emotionally involved. People I just want to be level-headed and understand. Yeah.

B

A

Be put in uh amount of technical depth that we accumulate uh behind uh in just yeah.

C

You know I I, agree with that. My uh feelings aside, the way I look at it is uh is this hurting more than it's helping it's hurting a little bit on the on the um the blind spot scenario. The single mode Standalone go black, uh but overall I'm totally fine, it's very about it. That's that's all I'm trying to say I, know. I know you guys are not uh it's it's too much.

A

If this is, if.

C

It is hurting more, we should revert it. That's how my answer will be, regardless of how much effort went into this.

A

So, for this new pointer exception, do we um like uh any expectations that the test will be added or how do we want to reproduce it in-house with something like that, or we just saying that we will merge null pointer as we expect yeah.

C

I think uh I want to get some uh someone who has knowledge of the uh the CI that Jordan has been using for what internal. If there is someone who has knowledge about that, what kind of manifests were used to produce this uh take a look at this and see if it if we can reproduce that at least in-house internally in Google, if you can reproduce it, that would be great. I have tried a bunch of things. I just need the Manifest and I. Don't know what it looks like.

A

I spend a lot of time on that.

C

D

Even you cannot reproduce this problem right so, with the standard owner model.

C

With local cluster I'm not able to I, don't know how it is done internally in GK.

D

Because the circus.

A

Specific manifest from Jordan I I try to reproduce with different, uh like updates and static port and like creation, strategic for those kind of things. No.

C

Thank you for putting all that effort in it doubles it. Can it kind of confirms my own work.

D

Thanks thanks David also David's offered to follow up on this way, so I review I almost approved. It's just I try to connect all the dots just like what I said. Why we are thinking about the vpa can work with the static partner. So that's why I didn't prove because I don't want to so. This is also I. Try to figure out why everyone can offer those problems, because uh so this is so so.

C

Yeah, let's hold on let's hold on that I just want to get uh I want to convince Jordan. We don't want to kind of. You know push on that without us without a zip approval, um I I think last night what I did was I was using the debugger, so it's kind of a very it's not the way of verifying. But let me do a very formal uh way of uh like okay I'll.

C

Do these scenarios I'm gonna use uh this particular static, pod or two different static, pods and then try with feature, get enabled disable and post those update those and then see I think the say the exclusion of static Parts change is safe. It's a good change, I, just sort of convinced Jordan about it. I, don't know how he said. Maybe he's saying something that I don't see so I'll take a look. I'll take another look.

A

I, don't think it's about whether he see something that we don't see just uh once it will be graduated. It needs to work. And uh if you don't have enough testing in Alpha, we can start breaking people in beta and.

C

A

Yeah, we just cover this in sidecar as well like this change. It doesn't have any unit, even not a unit test and uh like any change without tests scares me, especially if it fixes some panic situation.

C

um Anyway, in this case, so we we have, uh we need a lot of I think we definitely need ETV test for this whole scenario and uh we need to Beef It Up, with your tests.

A

C

A

Okay, um I'll suggest we go to the next topic. um I want to bring on status of sidecar cap. I know there are a lot of interest and we have a working group uh working on sidecars. There was some delays on sidecars because we uh sometimes struggle like uh we needed to rework something after In-Place update and we also struggled with uh finding approvers for early pre-work PRS, but I think uh we just had a meeting tomorrow morning or today morning and we had a um that's.

A

What we're missing and I think uh everything we're missing. We can get done so we have we'll have PR ready and we have a very extensive end-to-end uh testing process. Pr. The only question is I thought it asked like freezes tomorrow. So I thought that maybe we have a chance if there is a lot enough interest for sidecar cap to be immersed in 127., uh so I wanted to bring it up like. Is there any uh like if there is like immediate blocker like it's?

A

Definitely not gonna be in 127 I want to know if people believe that uh there may be a chance, uh let's discuss and then let's see.

D

Okay Eric: how do you think.

E

Yeah I, don't necessarily know what the urgency of well this isn't the cap this. What is you have a link to a pool? That's not an enhancement.

A

E

Yeah I mean I, would I would struggle with this one versus? Let's just do it um not last minute, particularly I might have just the prior conversations we had where we thought we had a lot of stuff going great and.

A

E

Don't take it more.

A

Okay, yeah, uh okay, I think it's uh something that we can merge very early in 128, because we have all the ducks in at all I hope it wouldn't be the same early in the in the Milestone as an In-Place uh update.

C

Yeah this is this should be a good lesson. Okay, words such a big, clear, early, I.

A

Tried I tried: okay uh Parker. Do you want to talk about uh epr ome.

H

Oh yes, replied in the uh in the broadcast, and this is a special uh assist control that changed.

H

It was, firstly, not namespaced and after since Colonel 3.16, it changed to a namespace to one, and so after that this would be a safe assist, control, I'm, not sure if this would be accepted, as is it is a little special.

D

I agree and I totally. Okay with this I don't know sick. The only things I think uh someone proposed how to enforce that at the promotion days right so I feel like that. We haven't discussed that and besides that I'm totally okay from the signal, the person I'm totally okay, it's actually in-house. If I remember correctly, for the gke, we already made that namespaced this one to make a GK Fox a while there's actually so I didn't face the internal just based on the Kernel worship, yeah.

I

Do we know like how many distros are on an older kernel, or should we just make this change I'm wondering whether it makes sense to check the kernel here or because even the kernel version isn't guaranteed one right like Rel, typically backboards, a lot of features, yeah.

D

D

So there so I I think that I'm, okay with this one's the safe, because if I remember a while back, we did have the minimal uh Colonel worship is 3.18. This is a while back we already in tonic, but the only things that I just happened to find I didn't got time to find that uh uh change yet and also I'm, not I'm, not too sure we did enforce that, because the huge code base. Obviously we didn't impose that that is only in the system, some system, never tests invoked by the cube itemi.

D

So we did the enforce that part. So so, but on the other hand, uh do we need to check the kernel version here. I, don't see that we checking conversion. Do we want want. Also is the question right? So if what I say that we basically uh for the in the queue that has been put up, that you know that many people don't use in Cube admin and that's also another thing so actually we did say check of the kernel version on that perspective. So.

D

So I'm not sure, should we just make the document that more clear or we just need to add the check and another thing is for uh kubernetes worship. We kind of the suggest from the 1.27 and above and we support this right. So do we need to check that one there, because I do see the.

D

So so I'm not sure this is this, this part I'm not sure but I'm totally. Okay with this is not safe. This cutter.

I

Yeah I agree, I think because on one hand we are adding features like username spaces that relies on latest kernel, features right and you don't have any kernel checks here and then this is looking for. A 360 is fairly old. Now yeah.

D

Yeah I remember the 1.18 that time I we did have this tall talk uh that's at least more than two years ago, then everyone agreed that the 3.18 back then or ID. So that's why I kind of feel this is super safe from our node perspective, Resource Management perspective.

A

Sorry indirect, do you want to say something.

J

Yeah sorry I just wanted to say, um like it'd, be nice if we I think this has come up a couple other times like if we document somewhere the minimum kernel version for various features, I, don't think we have it very clear anywhere so don. If you mentioned, like we said 318 somewhere before I, think if you can, we have a new pointer where that is, we can maybe make it more prominent so that we can use it.

D

You could just find it people just to help find that that's exactly what happens. Reference to yeah yeah! That's a sex, pickup I.

J

See perfect, okay,.

D

So it sounds like nobody really objected this one right so I'm going to approve. If will yeah, remove.

A

The version check from the spr.

D

Inkles at all only take off the uh our kubernetes release: worship right.

A

I, don't remember yeah.

G

D

A

It has this check for current version.

E

The checks are fine, I mean the danger on this. One is remove a check like I think this would be like someone runs on an old version of I, don't know, February, Ubuntu or something or rel7 I think might be hit by this check.

E

um I wouldn't make anything drastic right now, but I think like as an action unless we should come forward, come out with like a go forward policy but I tend to think 316 is old enough. That I'm not too concerned on this.

D

G

Yeah I agree Derek, it's it's not concerning we're. We have so many features that require four um and even five that it's not you.

E

Know this isn't a problem. Yeah I mean I'll, put my red hat hat on right now and be like if, if we had a men kernel version of 418, that would cover like our Rel latest minus one, which I think anybody who then would be following the rail family of derivative or red hat distros would be probably a decent policy. I, don't know for the Ubuntu family or the Google OS families, but.

G

Yeah, that sounds like a good. A good number I mean if you wanted to go root listening. You probably need five, but yeah 5 15.

D

K

Think but that's for yeah 418.

L

G

Fine, as a basement foreign, we should probably open an issue and do a survey on that.

A

Yeah, removing the Raspberry Pi builds right arm, uh so maybe a small devices will be affected, but we already removed for them. So I don't know.

A

A

Uh Kevin is Kevin on a call.

K

Hey Kevin is, but I can speak for this item basically is closely reviewing and monitoring. This PR and I also reviewed that and looks in good shape this approval for the future Gates. So everything else I think we can make it happen in meeting the the standards there and the in the timeline so raise attention to to the approvals for this.

K

No, it's yeah used to be, but it's not now reviewer, we you have it I, think it more correct state is approver because of.

A

You know the future.

A

K

Yeah I will in literally few minutes. Thank you for your help.

A

uh Thank you is there any questions about that I think we care for approved so should be straightforward. uh Ravine, hey.

L

Folks, um yeah so I'm just trying to figure out what the next steps for merging this node log query PR is. uh This was discussed uh end of Jan in this in this meeting uh at the moment, Jordan wants more reviews from Sig node to get this merged and I did get it reviewed by Renault and um and Ryan Phillips, and for the windows pieces, Mark Rossetti uh who's in the meeting also reviewed, uh but I'm. It's unclear to me at this point. What is Jordan's definition of I need more in-depth review, I.

I

Would I I think I I feel like this is safe enough to merge, given that it requires a feature flag and then additional configuration to enable it? Even after the featured flag, I mean- uh and this is Alpha so I I feel like we should be. Okay and I just wanted to wanted Arvin to raise it here, if folks think otherwise, or have thoughts.

B

Yeah I commented an issue too and I feel like this architecture is what exactly what we discussed in the sake architecture meeting with with Jordan and Tim, and it seems like a lot of the concerns- are how we're putting together the queries but I feel like that. Shouldn't blacken Alpha implementation.

I

Yeah and the yeah, the only Windows comment is uh outstanding and that can be handled if we find it a better way, I think. On the Linux side, we introduced some code yesterday to make it more secure.

L

Yeah, in fact, that's general or not like even on the Windows side. Now um any services with certain like you will not be able to add services with new line characters or with dots. Okay and I've made it very conservative so that, actually, even on the Linux side, there are valid service names that is going to get flagged. But given how valid Jordan is about this I thought: okay, I'm just going to be conservative about this, so it's I, I, don't know what more I can do with that.

L

You know trying to make sure that a service doesn't have an escape uh probability. This is almost like saying: I want this feature to be I want this Alpha feature to be merged without any bugs which I'm not trying to introduce bugs, but I don't know at this point.

F

Yeah and I think it's similar to the logic we were talking about earlier. I mean these are alpha features so like if they work and they're, not obviously making things worse and they're doing what we agreed on it seems like seems like we should be able to short-circuit some class of conversations like this.

E

um So, just just to focus like it looks like we're not, you gave a pretty good review and Arvin. You made a bunch of updates like it's just a matter of knowledge. You tag it like I trust, your reviews and all.

I

And stuff yeah yeah, so yeah I'm, gonna attack it I just want to make sure you know we have this discussion here and we as a Sega like okay, with making calls like these, where you know, there's some potential uncertainty, but it's behind a flag in itself, so we can go it.

L

All right uh once you do that or not I'll reach out to Jordan and uh on maybe API reviews or someplace like that and ask him to approve. Thank you.

B

It doesn't merge today, so we've been waiting for this for a while.

L

D

D

B

I started uh I started talking to Xander the early sleep um on psych too, like that this may be coming so yeah.

C

Yeah Mark I'll bribe him with uh you know: a beer at plastic, Pub.

C

You will approve my exception right, Mark.

B

It goes to the early sleep, but I have something.

C

Oh okay, okay, it can be a supportive voice. Yeah.

A

Yeah, okay, so uh could you just cause this one and yeah? It's uh another became some monitoring agent rather than uh orchestrator, but uh I think this is where's. The root cause of Jordan compliances, we're still struggling with C advisor uh Port uh advisor metrics that we exposed through cubot and trying to replicate them anyway, but yeah. We discussed it and let's continue as we discussed uh next one uh mark.

B

Yeah, this is a very, very simple change. um Some folks and segmentos were doing a lot of performance testing and found that by setting the that perf counter-up day period for the windows stack collector from one second to ten seconds to kind of keep or prevent a lot of extra CPU usage.

B

um So we and we tested this pretty thoroughly and didn't see any kind of downsides to going to 10 seconds. So we're just wondering if signode had any concerns with this and.

K

Also, if we need approvals.

B

Because it's in the keyboard there's a lot more uh like issue or a lot more like perfect, uh like reports in the in the issue.

A

From my perspective, it should be sick, Windows uh responsibility to decide on that.

B

A

B

Discussed it in a couple of sick Windows meetings and um I think James and Jay the some of the tales.

D

B

But this is everything to do.

D

I will take a so that's not the interfere with Linux knows that's all just normally. We are doing right. So you folks are the exporter on the Windows. Please make sure enough to change Windows and also like the sign and the changing Linux notes. That's all and I can't approve this one. Yeah.

B

All right sounds good. Thank you.

A

And with that we get to the end of our agenda and we have five more minutes. Is there anything else? People want to talk about.

C

I just wanted to mention that I added a comment to the to the vpa Panic bug, uh based on our discussion, summarizing that we don't encourage uh manifest product updates, because the source of Truth is going to be the file, and in light of that, the change that I made looks not only defensive, more correct.

C

Hopefully that captures it. Please add to it. If I miss some point here,.

K

Okay, thank you.

A

Thank you, everybody uh is that um yeah. Let's conclusion bye.