Kubernetes Community, 29 Jun 2017

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes Community Meeting 20170629

Description

We have PUBLIC and RECORDED weekly video meetings every Thursday at 10am US Pacific Time.

https://docs.google.com/document/d/1VQDIAB0OqiSjIHI8AWMvSdceWhnz56jNpZrLs6o7NJY

DEMO: k8guard
Releases covered: 1.7, 1.6
SIG Updates from: SIG Node, SIG Docs, and SIG Autoscaling

A

In the kubernetes community meeting today is Thursday June 29 and, as always, we record and publish this on YouTube, but for more excitement. Today, it's being live streamed, so if you are always hoping to be a YouTube star like Kris, is hoping to be a YouTube star she's, getting you two ice creams today. So we're gonna kick off our agenda and this is Chris Noba. Our moderator, hey,.

B

Everyone how's it going I am just getting them getting settled in to the Redmond office right now, so I'm opening up the docs I can.

A

Get us started with the demo, then, if that helps yes team's awesome, so then we have a demo this morning from many and start.

B

A

Are recording, thank you, we're going to start with Mehdi and he made yeah I'm sorry and he is going to be talking about cake guard and auditing system for kubernetes. So maybe yeah. Could you sure.

C

Good morning everybody, my name is media. You had me all.

A

Weekend, thank you.

A

C

A

Yes, please all.

C

Right so I want to tell you I work for targets, and I would like to tell you the reason that we came up with Keith guard and the reason that we open source said so kid dart is a ring system for kubernetes.

C

That's I and Ryan Belgrave have been designing and developing for the past few months and the way we came up with this when, after the reason, our team Nelson, he said we have hundreds of namespaces and hundreds of developers who are deploying to provide an ace, but we want security, efficiency, available, availability and stability, because not everybody was following the 12 factor recommendations or they were not doing the best practices and we didn't want to just manually, go and delete things without any trace in there.

C

We want an automated way of suppressing the violations in our company clusters and by the way, the name of the cage guard is not. Kate's. Cart is kate guard but kate, the guardian angel or kubernetes. So that could be relevant confusing sorry about that. So the kind of things that cage guard can identify and take action in them are things like this violations like image size.

C

If you have a super big image, if you're downloading from at unapproved image repo, let's say you want to restrict your images to be only downloaded from your own companies or your own repo, not from anywhere else or if you want to restrict containers with privilege, nor extract abilities for obvious reasons or if you want to enforce single replicas, not to be the case like if somebody's deploying that pods or containers with a single replica, you could also identify that or if you have an invalid ingress.

C

You have the few case that people were had a star in their ingress and that's with basically for all the traffic from your all the traffic from that cluster. To that specific part which you never want to be, or if somebody mounts a kubernetes clusters host volume to their container, that's also a big No. Another violation could be like we have a annotation for who is the owner of a namespace that annotation has an email or chat IDs. So if you don't have that annotation, we also can detect that, so you can also configure it.

C

So all of these violations that I mentioned are all configurable. You could say what is the image size approved on? What is the approved annotation? What is that for ingress? All of these things are, and all of these things are you can ignore them, or you say, I only want these type of violations to be activated. So when we first start developing this, we start the most simple basic thing. We just want something to discover all about things and then, by the same time, we do not.

C

We didn't want to take action on them in the same microservice, because we wanted to develop separately for it and also test separately. So we came up with two micro-services discover an action when the discovers bad things and when it does action in them, but layer. We also came up with a different micro service, called report that generates a human, readable reports out of all these violations and actions that happens in our clusters, so this is the overall design of it.

C

So we have three micro services, discover action and report, and the report is like reads off of a database Cassandra and will generate a UI for you. The action will take action on all the violations that could be scaling them down to zero. If a DRI deployment it could be suspending if there are a job or a con job in kubernetes, or it could be deleting a pearpod.

C

No and after enough warnings through chat or mail, they will do the hard action on them and you can configure that how many times they need to be warned before you do a hard action of an example of like I.

C

Can show you an example: even this is an example of chat. They have masks all the names here because it's public, so this is our chats application and we tag the people who did the violation and tell them this. Is your warning count number that this is the source of violation? This is the type of violation, don't do it, and if this is the last warning, is you read because this is your last warning, we're deleting your pants and also you send them an email?

C

We shame them in email anyway, so back to the design, the discover part also have an API that you can consume it and make integration at the top of this. The API I have have a mini cube version of it. This is an API response. If you click on, let's say, deploy to give you all the bad deploys. This API is being provided, so you can build at the top of Kade card, and this is the report in this report. This is human for human, readable and human searchable.

C

You can search for a specific violation and see what kind of violations were done by what namespace and you can search through them in a nice UI.

C

And also, we also collect metrics I'm running out of time, trying to squeeze as much information into this table. So we collect two kind of matters: violation, metrics and performance metrics. If you go to the API, the slash metrics gives you tens of more than 30 metrics that you can collect like, for example, one example would be yes, the percentage of bad deployments over time or the percentage of bad images over time.

C

Another metrics these metrics are provided in the slash metric endpoint that could be collected by something like Prometheus and you can generate graph, honor, dashboards or any kind of dashboards. You can do that in other metrics that we collect is performance metrics. So, while we make all this API calls to kubernetes api to see if something is good or bad or describe image or just describe apart. We also collect how long it took for that API call to come back so over time. You have a good idea. How slow is your kubernetes api over time?

C

We open sorts kid guard in Apache version two, and we have a website for it. That is I. Think is the one of the most developer friendly projects out there, because you can easily try it in mini cube in a few steps like if you follow the darts, it can build all the micro services and deploy everything, including Kafka, Cassandra memcache, everything in a mini cube as I. Have it actually, my computer I just build everything and deploy everything community in a few minutes less than a minute. I. Think so. I inquire encourage you.

C

If you want to give it a try to go to Kate guards that gets out that IO and click on try it, it will guide you through how to build the projects for either mini cube or dr. Campos, and also will give you all the documentation you need about what kind of violations kit guard is capable of doing I. Think I run out of my ten minutes.

B

Well, thank you and I think we agreed if you want to keep going. If you have more to say we have a few more minutes, we can spare in the call and if not, we can open up the floor to questions now. Oh sounds good. Yeah.

C

I can show you more of my presentations.

C

So this three microservices, if you, if you go to github, let's go just to get out now: github tough town, / kid card, you'll see all the repos, but don't let that scare you that there are so many repos. You don't know what to do. Just you start from kid card start from here, and we have a make file that will guide you through doing all the commands you need for doing it. We encourage contributing back to cake. Art is our first release.

C

And you see what's the best thing.

B

Does anybody have any questions.

A

Got some compliments about it, it looks neat and it's just like there would be people investigating it as well. Yeah.

C

Looking forward to checking it out, it looks really cool. Thank you and one thing that I would like to mention that. Currently we run this, the Keith care discover has two most an API mode and also a cron job mode. So the API mode, as you saw, is it returns an API for you in terms of a JSON response, but also you have a messaging mode which runs currently as a cron job list, every 30 minutes and goes through and discovers all about things and puts into a Kafka topic and the action consumes it.

C

So you might want to be interested to see how fast this can be. So if you want to consume the messages faster, you can scale, allows the action micro service to let's say 5, 10 or whatever you need, so they will try to do the actions faster and then what one thing that action does.

C

It stores the state of the violations of how many times they have been warned and all the actions that have been done on them in a database in Cassandra, so they will not step on each other's food by scaling them out, and one of the integrations that we are thinking to build at a top of a cait card is called we code name it kate radar, which is gonna, be a real-time detection, as opposed to a cron job that goes every 30 minute or every minute that you want.

C

It will be a real-time that as soon as it's a violation, it immediately sends it to the same kind of a topic that action is consuming, so I just wanted to make sure to demonstrate that this is not real-time currently, but we can easily add a real-time integration on top of it. Another thing that I would like to mention.

C

The word guard here is not about guarding in terms of a malicious Nick network activity. Don't detect any of those things. We focus on violations that they're actually talked about 12 factor or this list of violations that I've listed in the documentation.

C

Another thing that we we have designed it like the violations are very extensible. So if you come up with your own violation, you can easily add it through. The code is currently it's not pluggable, hopefully, in the future new pluggable, we accept the Arts. By the way we they're very easy, accepting our peers and we'll be happy to review it and merge it back to the main repos.

C

We had a. We had a quick question: is this used in production anywhere? Yes, we are using it in Target Corporation. We are using it across all our environments in the data centers and clock public closet, er, yeah, they're, wonderful,.

B

This is really great any more questions from anyone else. We have some more compliments in chat, looks like Jorge's impressed. Thank.

C

You and the technology that we are, we are using they're using golang for the development, kafka, cassandra, memcache Prometheus and another thing that I would like to mention. You do not have to have all of this. If you want to deploy this like, you can basically go only discover you don't have to have reports or action, you can only have a Discoverer API. That shows you bad things. You don't need talked about Cassandra if you don't need action, but if you don't need reports, you need all of this.

C

Only if you meet all of the micro services. If you want only look, for example, no bad things, you can just deploy the discovery API and it doesn't need anything like Cassandra or Kafka. If you are scared of those things, but by the way, and all of these things, I have an example in the mini cube that if you want to just deploy the mini cube version, you can have Cassandra cup and everything included in it.

C

B

Was the inspiration for you're using Kafka as a broker? Is it the public publish/subscribe battle, or was it the actual Cuban side of things? Well,.

C

We want something that was available and the way that we wanted to scale out the action service. Well, Kafka was the best way, because you want that. We want the same message to be able to be consumed by different clients across like I say. If somebody wants to build an integration that also consumes Kafka at the same message that we discover they build their own integration, they can use that easy Kafka because they accept different client IDs as opposed to something like rabbit. Thank you yeah. That's really is like that's a great use of Kafka.

C

B

Really awesome. Thank you. Okay, let's go ahead and move on if nobody else has any more questions. Thank you again. That was a wonderful dinner demo and it looks like next in the agenda. Here is release, dates and I'm, not seeing anything for 1.8, so I probably think we should jump into 1.7 and it looks like we have dog Chen a release manager with some updates.

B

D

So this is boring this time. So basically we can't. We can't allow there be nice candidate this morning and hope this is the one. So we kind to be waiting for the final test results and last night we can't one has to be out cut one release candidate and the other has to be understood and then there's the one is missing, because P is missing exit and tonight, so we had another one so there. So we are also negative. We have the end, you are documentation needed here and, and maybe you can yeah.

E

So all the docs pr's are in and merged 41.7, so we're gonna go so when some releases of them will merge that into the case that IO site so.

D

We also have the other cycle group leaves and the contribute to the ladies nodes and the right. Now each team is kind of a cry at the final stage, punish the release, notes to make them more user-friendly language, so so we're pretty much it everything's rocket ready is waiting for the last amid our signal and then just say call chaos. It's awesome.

A

Yeah it freezes, so people in the room are asking me if this is working. One I actually share my screen. It freezes our visuals into this, so I'm gonna stop sharing our screen. Oh look, you're all moving again high.

A

But that was one point seven for you: okay,.

B

Moving on to one point, six point: seven updates Anthony yay; yes,.

F

So we're coming up on about two weeks since one six six, we do have a couple of issues that relate to fluency and juicy feed that they consider important enough that we within these two weeks so I'm aiming to do it next week, right after the holidays,.

A

Awesome, thank you. Do you need any input from anyone or you just you're all set.

F

I'm going to send out an announcement today just to say you know, this is the draft of what's going to be in the release. If you have anything that else that you want to have input in, let me know.

B

Ok, moving on to these sig updates I will hand it back over to you.

D

But what we really were in the queue, and so so in a secure signal and image YouTube, we we enhanced after our company webcam interface, and we introduced the continent matrix to our interface at the same time and the community make the autorun handler made more progress. So in the 1.0 we have the company integration.

D

In alpha- and we also have the prior at the company one time file in the 100's, and we also have the flag D, which it is RI the hypervisor base of the Red Hat and continuity, and it is also in the 100's and you're. Also in this release.

D

We have it has, after our security, suite CPU CPU support, I'd known and in the image Copernicus and the better individuals of the 90s, and we also fix a single problem into the our enhance our of of the user space of the equation management for the node, which it is make. The weaker node is more robust when there's the resource establishing issues happen, and we also. So we also have the give me one.

D

Second, we also have a add more about connect, the no dominating of the security enhancement and denied, for example, an imitator of the Cuban egg access to the secret and about an object based on base notes. So those kind of the second husband Allen on the side. So we also have a more writer support and the image calendar and it takes some a lot of issues in our system and just one sec. Sorry.

D

So we also have the so we also support off the the syncopate, the namespace, that is another actual development plaintiff for the protocol, and so so we also in another comedy factor and we add a lot of the support. We add out the kernel kernel not support into the English release, so we also have to make the novelty factor more extensible and Platteville. So in this release we actually I have the Sun or night.

D

However, just be fabulous of the doctor without the issues we have those things and as the and writer side- and we are in a husband after supporting stalkers power over the to storage driver, and we also coming up some EVP metrics, so all those kind of things in the queue and the industry we are going, we are going to so once this I forgot in future, also not him how about pre visual in the resource management, and we cannot define of our autonomic for the release management.

D

You know in next year, so we talked about how we are going to use the point of the hydrophone is open all and how are going to continue to enhance our 32 you or we talked about how we are going to you in has finished for the notice back on the resource management side, so those idiots will found in YouTube and Industry can add those resource management. What is going to be executed?

D

So we have the work plan and so, for example, so we're going to add the call, a location and the node to to boost the some of the users, outputs of the high performance or closer. We also in the qq3 we are going to either more of those know. The common detector and also by the individual, is the remedy system, all those councils, and also we kick off the trip time to discuss and design.

B

A

Move on to sig Docs with Devon Donnelly, please, it won't be Devon. It will in fact the end or Chen, who is sitting here next to ya.

G

Know actually it'll be me I.

A

Or the combo we can talk to you, yeah.

E

Well, tech image: deafen you go first, okay,.

G

So I'd like to talk a little bit about cig, docks, past and future strategy, which is a place we've we've been discussing in the last couple of sig docks meeting, so sig docks in the past year has primarily concerned with getting the docks house in order.

G

One thing to be clear about the primary source of docks generation in kubernetes is the actual engineering SIG's themselves, so networking storage, node, etc. Most of these SIG's actually write their own docks, and these docks are generated at release. Time.

G

What sig docks has been doing for the past year is giving them a attempting to give the other SIG's a place to put their documentation until we got some greater involvement in sigdoc strategy.

G

Documentation was basically a big flat list under user guides, and the actual topics themselves were more or less freeform without much in the way of structure or guidance or standardization. So the thing that sig docks has been primarily concerned about for the past year is introducing that standardization and creating a taxonomy of the different types of docks that we want for kubernetes that are best for the project and then migrating the existing flat list of unorganized topics into that structure.

G

We're at a pretty good place with that now, where sig docks needs to step up is to.

G

Increase our involvement in the other SIG's in some way or another, because right now, the interface between the other SIG's and sig docks is when some member of say, sig, networking or six storage sends a pull request against the docks, repo and and all the negotiation and guidance about whether or not this docks fit. The dock fits and the overarching strategy happens at the pull request, level I would prefer sig docks to be involved earlier or vice-versa.

G

There are a couple ways we can go about this either.

G

One of them one way is for sig docks to start assigning its members as liaisons to the other SIG's and working closely with the other six to determine their documentation needs before it's release. Time and it's five days and still release and say. Networking, for example, is sending us PRS.

G

This would require a large investment of resources into sig docks that we're currently not staffed for the other way. It would be to do it in Reverse, which would be to designate members of other SIG's to liaised with sig docks, but we, however, we decide to do this if, indeed we decide that this is the best way to serve the projects. Documentation needs I, feel like we need to open this channel of communication.

G

That's the most important thing. Whichever way it goes, either sig docks reaches out or other other SIG's reach out to sig docks. Somebody needs to reach out in one direction or the other.

G

Since sig docks, sig Docs is undergoing a bit of a shift now that, now that the documentation itself is more stable and I think we're in a place where we can start doing this. As a sig like sig docks is.

G

Less concerned like now that, now that the documentation is has been reorganized into things like tasks, tutorial tasks and tutorials and concepts instead of user guides, we're definitely at a place where we can do this and we're at a place where we can start accepting a dish help from contributors.

G

The thing is, though, that most contributors to stick Docs can usually only give us a couple of hours a week and for somebody from sig Docs to be a liaison to a to another. Sig is going to be a much larger job than that. So if we can get contributors to sig Docs that are willing to give us some more time, that is a way we can use them. Does that make sense? I.

B

Think it makes sense.

B

How would we, how would we go about going through the process of answering this question of what we want to do.

G

So what we need to do is I think just this is just more or less unsolicited opinion here.

G

G

Come up with a better, more aggressive timeline for when documentation related information flows from engineering SIG's to sig Docs, because, right now it flows at exactly one point as a large pull request to the docs repo. We need to set some standards about if you're developing a feature for a release.

G

Here is when you need to have the user impact, slash user guide, information ready and here and like- and here is when sig engineering sig needs to start communicating with sig Doc's about their documentation needs, and those dates need to be slightly more aggressive or significantly more aggressive than what we currently allow for documentation PRS prior to a release.

G

What we allow like we, we allow documentation PRS to even be submitted up to as little as a few days before a release, and if a PR comes in and it doesn't fit in the docs, we kind of just have to eat it and fix it later, because we want to have the feature documented at all.

G

Ideally, I, like in a completely ideal world I, would see sigdoc staff with a whole bunch of technical writers from across the community. They would actually be generating the documentation, but sig Docs has four full-time technical writers at best, which is definitely not enough to cover a product with the massive surface area that kubernetes has so what few resources and full-time technical writers we do have are primarily concerned with organization and curation, and the actual content generation comes from the cig engineers themselves for the best for professional quality documentation.

G

I personally believe this isn't the best solution, but it's what we got I.

H

Proposed as an alternative that this discussion that followed extensively in the retrospective and potentially driven through cig release as a bit of additional release, criteria means you help us fight against the perception that code is immutable after the code freeze, but Doc's can keep on changing long after and perhaps if we put more stringent criteria earlier on in the release process for the existence or quality of documentation.

G

Mm-Hmm I I, don't even necessarily think we need to have draft documentation early, but what we have to have is some kind of almost like a user impact PRD for every feature right just write up a little piece of documentation. That says this is how this feature is going to impact the user. This is what the user is going.

G

This is how the users experience is going to change based on this feature, and then we can determine what kind of actual documentation topics need to be written to cover it and work with the members of the SIG's to make sure those get generated in one way or another.

A

You can share broadly with kubernetes dev. We can have spirited comment debate and we can resolve to some sort of plan forward from that I.

G

Would I would need to I think that what what we've shared back and forth is I'm not necessarily quite ready for the wider community I'm happy to write it up and send it out. It would just take me a few more hours of work on it. Yeah.

A

Awesome and you've gotten you've gotten some feedback that we in the chat that it might be really good to go with a similar model to what's 8 p.m. is asking for which is a representative from each of the six participating in their group or at least participating on a regular basis. Mm-Hmm.

G

I would I would I would think. That's that's one way to get there. Yeah.

A

G

From the other SIG's at sig, Docs yeah.

A

I think that I think what you are suggesting is patterning and having one one more standard way, unless we have a reason to do it. The other way.

A

So, as I said, discussion I just don't suspect we're going to come to any sort of answer right. In this conversation there.

I

Are some comments in the chat about requiring documentation or release before teachers get merged? That's really really hard right now, because often usually features are sprinkled across many PRS. If we don't really have a good way of blocking them or alerting them after whatever time we're deciding what unit nice or not we're going to be experimenting with feature branches as solutions, so we can actually make go/no-go decisions, not just based on documentation, but also weather features are ready, whether they have adequate testing and so on.

I

Up to 1.0, when the project was smaller, was easier to rip things out and then force people to recently or later, but that's not really.

G

There was one more thing: I wanted to address: that's somewhat less grand, but I think at least somewhat importance. If, if I have time.

B

Yes quickly, please, okay,.

G

So it's been a long-standing thing for having support for previous versions of kubernetes or documentation versions for our documentation for previous versions of kubernetes. Apart from the latest version and in q3 we are migrating: the production server from github pages production, web server for kubernetes I/o from github pages to net LeFay, which will let us host branches of documentation. Apart from master, this means we can host the branches for previous releases of kubernetes, such as 1 6, 1, 5, etc.

G

I just want to be clear about the amount of versions the amount of support we were. We will provide for those version branches.

G

Basically, every time we launch will back merge the current version of master into the into say the previous, like 1/6 or 1/7 branch, so that all the bug fixes get in. But at that point it is a snapshot.

G

We do not maintain previous versions of documentation or actively start making pull requests and bug fixes against, say the 1 6 branch or the 1 7 branch. If 1/8 is the current version so once that release is finalized, that's what we got sig ducts probably doesn't have the resources to maintain 4 or 5 versions of documentation. At the same time, considering how much we're stretched with one version, but it's it's better than what we have now and it I think it will be pretty serviceable.

B

Ok, thank you very much.

E

Meaning for the sig Docs from one hour by weeklies to weekly one half hour meetings next week since it's a holiday, we're not meeting but then the following week will be one hour and then we'll start going to the half hour format and then yeah and then related to. There seems to be interest in contributors in the community to help write the docs, so I've been talking to some people at hefty Oh.

E

Heavy dough is generously like contributed sometime 7, so we're gonna anyone who's like interested and having sort of longer-term projects to like contribute to the talks. Please contact me or Steve Perry or Zack our new editor. That's it thank.

B

You, okay. Moving on to the next Sega update. We have sig auto-scaling with Sally Ross hi.

J

J

Anything I'm doing for the next release. So what went into this current release were a bunch of improvements to cluster autoscaler, probably the biggest of which was improved. Support for for heterogeneous deployments in which cluster autoscaler, instead of randomly choosing a node type to to scale up, will actually figure out what it needs based on the pods and then kind of use, an internal pricing model to figure out all right.

J

What's the what's the what's the minimum size that's needed to support what we we also landed, an input, an improvement for a horizontal pot, autoscaler called status conditions, here's the status conditions on things like pods and nodes and as well, that kind of indicates the current state of the horizontal pot autoscaler in terms of things that could be potentially causing it to not behave like you would expect so everything from oh hey. We can't scale because we're having a problem connecting to the metrics API, oh hey, we would.

J

We would be scaling you up by 30 replicas, but you've hit a cap of you put a cap of 10 in your horizontal pot autoscaler. So maybe check that out and increase your cap, so hopefully this will make it easier for people to see things that are blocking or currently affecting their horizontal, which is something that we struggled with in the past.

J

Those are those are the two major improvements looking towards the future we'd like to start you try to stabilize or move to beta rather horizontal pause, autoscaler v2, with its support for status conditions and custom. Metrics and also work is ongoing on the vertical protocol, quad auto-scaling initial concepts.

J

The proposal is still undergoing discussion and what we've started to get some initial through the concepts code merge or proposed in the kubernetes autoscaler depository. So hopefully we'll get some more of that in the upcoming days.

B

Awesome. Thank you very much for the update, I think moving right along. We have 18 and change minutes left so jumping into the announcements.

H

Hey real quickly share my screen if I can first off the first quick announcement is that state testing next week is on July 4th I hear that's a holiday where people blow things up, so we won't be having our meeting then too busy blowing up our tests right.

H

Okay, so about two weeks ago, I sent out an email asking for some clarification on triage guidelines for issues that have the needs: sig label, which corresponds to issues that have no say owning them, we're down to one from 1100 about two weeks ago. So I just wanted to call out huge congratulations to the community and to thank everybody who has put in the time to triage this out.

H

The other graph I wanted to show real quickly. If I can.

H

It's from Socko this is slightly out of date, I'm, not sure why. It shows, for example, that there are 39 open issues that don't have a snake, but this will give you an idea of roughly where the work is distributed. So, perhaps surprisingly, perhaps not signal followed by City API machinery, followed by six CLI, seems to have the lion's share of the issues. I'm sure this isn't actually representative of where these things belong, but this was the result of a lot of people taking their best guess.

H

This begs a couple questions on all shops: let's go up sharing my screen.

H

What do we do now?

H

I wish I had a better answer to this. Speaking as a representative of cig testing, my plan, for our sake, is to apply priority labels to the hundred or so issues that we now own and then decide whether or not we actually want to take some of these on as part of b18 I'm.

H

Also going to use these issues as an opportunity to figure out whether or not it makes sense for cig testing on these to help start drafting a charter for what it is, that I think save testing shouldn't and should not, as speaking as somebody who helped triage some of these issues, there were a couple of areas where I had questions or gaps, I, believe George and a couple of other of us.

H

During that you should either experience meeting said we were gonna, put out, bug triage document and I plan on adding to that some of the some of the learnings I encountered.

H

So one of the really obvious ones, I asked the mailing list and we think we'll we will have an answer for right is what are issues when we do with issues that are gke specific? Well, we think we ought to have a cig, GCP and eventually that will exist, but in the meantime we have a set of labels that belong to that. Some of the other hairier stuff was like.

H

There are a lot of things that live in the controller manager that don't seem like they apply to any one specific cig, like some of them were like workload, specific and I think workload management may be falls under cig apps, but there were a couple other things like the node controllers. Well, that's got the word node in it is that owned by state node, or is that about the lifecycle of the cluster or X etc?

H

I think these are questions that we're going to require SIG's to help answer as they go through and sort of look at the issues that have been assigned to them and things we could probably escalate to the steering committee. If and when that moves forward, yeah anything there so, like I, said I played on I can send out an email. We could start it throughout on kubernetes dev about this, or there was a.

H

It was a dock where we were gonna, try and codify like what guidelines did we follow when we were triage, including what things do we want to do and then what priorities and real, quick before you jump in another thing? That came up yesterday and I believe in email that that's not to Cooper. That he's talked about this. Is some users come to this community with a very component centric view, not a very sick centric view, so it's clear to them that they have a problem with coop proxy.

H

Now in my head, I know two proxies: will it's probably networking, so it's probably only by signet work, although maybe it's significance, it's a thing that runs on the node, but they're gonna come at it from a component century, so figuring out a way to align components with States and could allow SIG's that have ownership of a large number of components to further triage that out.

I

You talked about.

I

Historically, where things have landed, but you know I, think most of them are at least obvious to someone so having having the was there comments on that, but I envisioned doing there's a couple of things. One is: we've talked about putting take ownership into the motors files in the tree, so that would be one mechanism by which you know the queue proxy, particularly we can make it clear.

I

What's on that they're also, you know there was a question about you know: what do we do in the area labels I had envisioned having some file that maps aerial angles to say well,.

I

Automatically connect it, someone new to the area, they could specify the area and then the mapping would automatically apply to say I almost.

D

Don't have idea to me, this is a simple media. I think the art high screen, it's butylene, oculus, take a group person. Chica has the hyper eyes, so you are going to the type of great you can see here to stick know that what do you care about, and so we can stand Hamer so negative examples and Cuba proxy is the example people promised they should have at least a step after the test. Integration Hank the function Annika is that others can enhance and the some of the tasks in a scan, better signal.

D

It is you some others has. It is majority from home overnight of sick network, but the certain in ablution and Pat. It is can't write a signal because then we'd make it know that it's not too stable, so we care so so those kind of things it Ernst because monitor kind of book. If I don't sleep rules baby, please I'm a set of the tester. So when the particular class is file, we could trigger those kind of things and give to a private label and freedom will increase in has a focal proxy.

D

So if that feel, and then we can say, oh ultimately type the frogman x infrastructure, so that it is owned by the signal and a see a light worker, so both teams need the connector income exchange. Those idea was pretty problem. Where is it okay? So that's kind of a major like the release of the engineer for the first time and and so the uncle people can charge you based on that. So it's a little bit a negative signal. I'm called people thinking I said okay, so what's the new issue by default? Bezos cannot place.

D

The new has appear based on those consistent. No one can up the actual track, so those kind of things and find a little general. No thanks.

I

For tests earlier, specifically, there is a biome out of maps all the tests.

D

I

D

This it is, we talk about in a one point for each, but that still is not let what I say that it is not specifically, it calls every woman's an example proxy. Actually it is Oh Matthew T, it is most of actually it is Mac multimedia. Try it first and invincible proxy good. It is integration, but not the problem. That's the glop here is Q proxies, and actually it is a people.

D

Think once know the problem, and so we we quickly the most in the past will be be the most of the tragic and thank you would find the problem then hand over blackboard, so the any of those integration problem. We need a clear and others we need also the remote has for those companies make most premium smokier. But what's that campaigns we need at the present time and it had the true group of people working together inside the negative blog. The progress.

H

One other great learning I had as a result of this is kind of metrics or measurements. If I hadn't sent that email out I said how many issues were open with Nate cyclical about two weeks ago, and there's no graph that I could show you that shows things over time. I really wish I could I'm sure it would look pretty awesome, because anecdotally I saw that number dropping by like a hundred a day and I. Don't actually know everybody who helped triage it down, because some people could use slash, dig things.

H

Other people could apply a label manually and thus far I haven't found a consistent inquiry that I can run against the get up archived data, so this is kind of a crime or if anybody knows how to do really good, interesting, analytic stuff like this, we were talking about the fact that there is exit coming up within Google and to trim back specifically. He was talking about what cool things did we do joining that contributes, and my opinion is one of the chopping wood and carrying water things. We can do during that time.

H

That can also be fun when he's sort of figure out. What are the goals that we want to accomplish, and how could we measure that definition of success, because I I feel like this assignment of things to SIG's, will help us for the better I solved that open issue counting bathroom about 4400 about 4200? In the time we've done that so I know like some of the really stale issues that have been open for a year have had some effect, but it would be great to really sort of define me.

H

What me think success looks like and measure that go. Do the policy change and then see if we had the effect we want, because it feels like a number of times in attempting to steer this project we're kind of flying blind. You think it's doing things for the better, but it's really tough to produce graphs. That say.

H

Yes, it went the direction that we thought it would go and so, like the only thing, I know how to do at the moment is: go play around with bigquery and the github archive data of events that are generated on a day-by-day basis, but maybe there's more useful stuff out there, and you would certainly love to hear about it. It's a contributor experience.

I

Yeah hoping that that will be useful for building.

H

Do you do you have any news to share about that? I mean at the moment I as a consumer of that can can sort of go login and click on various panels that that breakdown of issues by sig is really the only thing that I've I have personally found useful for like look getting a feedback, but the stillness of the data has made it difficult for me to rely upon that and I am as of yet unaware of how to explore or add my own charts or graphs. That sort of thing can.

I

You start a thread about that. One contributor experience.

K

There's a pretty good power, bi github integration. So if you've ever looked to it's sort of drag-and-drop, graphing and analytic stuff, that's another thing: we could look into. Turning on.

B

Okay, I think starting the thread and following up, there sounds like the best point of attack.

B

So we only have six men and change left, so I think sarah, you had a quick note. You wanted to call me just one note which just.

A

Says, thank you to george, who has been piloting an attempt to livestream this on youtube today and we had 15 viewers, three of whom were commenting while we were having this discussion, so we will publish this more widely and see what we can do about having this live stream, as well as the zoom connection, as well as having it available as a recording for all posterity and all your friends to watch and latha yeah. So thank you. Thank you, George for doing that and taking it on as a project.

B

Does anybody have any other last-minute announcements.

A

There's also a thank you to you, Chris for hosting oh you're,.

B

Welcome and then I have an announcement, and this is actually a call for forearms to I will be stepping down from the organizer position of sig AWS and that will be leaving us with two organizers. So if anybody knows anyone who is looking for a wonderful opportunity to step up and join the sig and get involved and possibly help out, that would be great. Please point them my way or to the rest of the organizers and we can get them on board it and ramped up.

B

Great, no other announcements from anyone.

B

Okay, I'm gonna call this meeting I'm gonna organizer, so I can do that happy Thursday. Thank you for joining.

I