Kubernetes SIG Windows, 31 Mar 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes SIG Windows 20200331

Description

Kubernetes SIG Windows 20200331

A

Hello, everybody and welcome to another sip windows Meetup, it's the 31st of March. As always, this is a recorded meeting. So please adhere to the same CF code of conduct. No one and all the times are tough. Now you know everybody is working from home, pretty much all over the world. If any of you are either due to personal commitments or even other professional commitments. If you can't do something and you need some help, please ask what a small community- we usually don't know each other in person.

A

Maybe you've seen each other at some conferences here and there, but I know that everybody's willing to hop. So if one of us is getting in tough times and anything we can do to help, please ask the worst you can do. Is someone can help you? But if you ask the only way you can find out so for some of us working from home is the norm has always been like that for me, I've been working from home for 14 years, but for others this is new.

A

So if you're, looking for tips or ideas or anything to make you more productive, ask again as well so I know that a lot of folks are going stir-crazy working from home and not getting the social interaction in the office or not being able to go to lunch with people, or you know seeing your wife and kids 24/7.

A

Maybe so, there's different challenges that come with working from home and actually the one thing that I've always seen is that working from home- and you know, speaking from experience, your calendar seems to be more packed with meetings and when you're in the office, when you're going and seeing people and and talking to people. So when you work from home, you'll notice that you have a lot more meanings than otherwise and I. Think that's true. Universal feedback that I've seen from everybody that just started doing this.

A

You know you tend to find every stall in your calendar and you fill it up or in your office just thinking about all his lunch time- or this is you know, maybe if you have some time to go between buildings or or this, and that and when you're working from home that that those realities are not there. Hopefully, this virus situation doesn't last any longer. My my prayers and thoughts to all the star impacted, especially the ones our family in areas are more impacted than others like Italy and China and Spain.

A

But please ask there's anything: we can do not only a sick windows by this kubernetes as well. Please do ask cool any feedbacks or thoughts on that feel free to ask a question: have you one otherwise, who can get into our agenda all right, cool, quiet bunch today, all right since I saw deep there I did add one thing for him: Oh, someone is being a Nostradamus in saying: si si approxi release for next week.

A

B

I just want to announce that we did release the first alpha version of suicide proxy with a lot of help from KK and Jing.

B

So yeah, it's it's out there. We tagged the release, it's being tested, rip, I, heard disk and gcpd as well as a hair file. So pretty much. The file disk and volume subsystems are all working along with SMB.

A

Test with vSphere as Waller is a not part of the plan. We.

B

Need some? Ok, so that's a good follow up, so we probably need someone from the vSphere team to kind of take a look at the vSphere CSI driver and what it would take to integrate with CSI proxy at the same way can change it or at your desk, and you see DC PD a.

A

B

An action item there for.

A

C

I can find someone I.

A

Don't see something been shown today and I.

C

Did the work to integrate the CSI proxy with aks engine for our test? So you can ping me on slack. I can point you guys to that to do the equivalent in the fees, fair setup.

B

Super another missing item with a similar follow-up would be EBS, so we need to figure out who, from AWS can potentially own that piece of making sure the EBS driver can also support this mode. Abs USERRA.

A

Let me add that to UM remember there was someone that didn't for a second, a deep wasn't there, someone from from AWS that was pinging us to get back ported some of their functionality around disk. That's the person who find do you remember who that was yeah.

B

I'll try to dig up that person. Basically mainly I was working with Chang pan from AWS and EBS, but he left AWS recently. So.

D

B

Think it was Matthew Wong, maybe I.

A

Remember he had like the reason. I remember is because you wanted to cherry-pick features into old releases and then got something with it. We didn't want to do so. I will I'll do a little bit of digging as well and see which of us finds him faster, but I'll. We can follow up with him, so that's great, so yeah B's fee, so AWS and UCP, so juicy piece done roger is done and then we just need this fear and tcp and then it w sorry yep.

B

So I would just call it vSphere support, rather than vSphere tests.

A

Yeah yeah, that's basically what I can say cool. Thank you. This great progress is awesome to see so on the topic of that since you're on topic of CSI I'm. Assuming deep, are you gonna continue taking the lead for both CSI on CSI proxy moving forward.

B

I can help I think mark is also pretty involved. They're so happy to hand it off to mark I've.

C

Yeah I've done a lot of transition work from KK who switched off of our team to make sure that knowledge doesn't get lost. I, don't know if I'll have a lot of resources in the next month or two to.

B

C

A

What's our goal for CSI on CSI proxy for 119, for example, so one of the things with last week was start thinking about high-level priorities for 119 yep.

A

B

Of the big requests that came in from from like the NetApp folks in six storage is ice, cozy support, so basically building out API group for ice cozy and there potentially they would look at contributing. It was not clear whether he really wants to contribute, or just once the API implemented.

A

They better come and contribute. You can. Actually you can flat out tell them that hey, that's, not a priority for us, but you can make it if you're willing to come in and contribute coming, hobb us and we'll make it happen. Yep.

B

So yeah I'll talk to Ben who brought up that point about potentially contributing, and let's see other than that we like at a super high level, as we discussed during the CSI proxy meeting. It would be also good to understand where you know Microsoft's long term vision around privilege containers is, and if it's something coming in the future in the immediate future, then we have to balance that with how much investments we do it CSI proxy right output, that kind of be a 20-19 thing, mainly yeah.

C

I, don't have a lot of details but I. My kind of recommendation right now is assume it's we're not going to get privileged containers for 2019.

C

There's been a lot of conversations that are going back and forth finish, but yeah I would say it's pretty safe bet to say we're, not we're never going to get privileged containers for back ported to Windows Server, 2000 18.

E

A

And and that's very significant, all right, because most people are gonna stay on the LTS channel, specifically for production workloads very likely, even if they don't it takes a while for people to move to the new, a new channels to the to the sa c channel right. So it it may be important to keep these investments, especially they are not a huge investment from our end, because it's gonna enable backwards compatibility and maybe build for the next couple of years.

B

Yes, that's and one of the other main goals is just continue to improve the C ICD framework, potentially with you know, mark and others marching and others in in terms of making sure that we can potentially get at your desk or maybe DCPD CI framework just grabbing the latest code from the CSI proxy repo and building the whole thing and deploying so can I have like more end-to-end CI CD rather than just CSI proxy. It's not like unit tests, so.

A

So what does so our youth from a functionality standpoint beyond the things you mentioned, what would prevent us from going to beta? We see si proximal 19.

B

Yeah I think the main thing is just getting that CI CD framework really rock solid and overcoming some of the unknown questions around building containers and sidecars for Windows I. Think that's the main thing we have to figure out. As for a part of this work,.

A

Okay, so so, while we're looking for vSphere as well as AWS Storage support- and you guys are working on improving the CIC, the framework I think that we could comfortably say that it could take CSI to beta with 119.

C

A

We end up in this. Okay, sure sure- and you know, obviously you know things could happen and can change that. But you know one of the funny things I just realized. Microsoft are you when you guys release the semi-annual channel for 2020? This is gonna, be called 2003, so Windows 2003 I'm, assuming everybody, has Sawada from the marketing said on your end, that that's a really bad name, I. Think.

C

It's actually gonna be 2004 if it's equally as bad right, yeah. Yes, oh.

A

They didn't think about that one they started naming them like that. Did they no I, don't think they did so.

C

Technically the semiannual General leases are always Windows, Server version like 1909 and whatever and the windows, the LTS C channels are just Windows, Server 2000 nineteen, but I. Don't think anybody pays attention to that. Yeah.

A

It's almost people when you tell them hey, why don't you stole 2-0 2003 and then I say that what yeah.

C

A

Excellent, all right: let's, since we have D barn, let's continue one more item for deep there's, this PR ad 907 with the device capabilities deep that you reviewed and everything was put in your hand. I think, is just sitting on sick note and note has been impossible to work out in the future.

A

It was this PR was pushed out of 118 right.

C

A

There isn't they pushed out. It was because they said, oh, you don't have tests yet, but you already had mentioned in the thread that tests can come after no I. Think it's just this. Concerning that, you know how how slow is are things to basically get up proof from sick node? Does anybody have any ideas on how we can actually get this through? Does it mean that someone needs to just show up at signaled? Would that even work so.

B

Just to clarify my point was: is the test can come in a different PR, but I can.

C

B

In the same release cycle yeah, okay, I'll.

C

Just add that I've been having a hard time getting approvals from sig note recently as well. I've got a couple of back port pull requests that have been open for probably two months now.

C

A

Put them on the list channel so come, pin them and then I'll put this on as well. There's a bigger discussion around signal than the fact they're slow to approve pretty much everything so I want to show them as evidence. So if you can put the the PRS, you have and I'll add these. So so deep can.

D

Really draw letters on.

A

The sick windows leads channel okay, okay, deep! Can you do one thing community right to this PR and tell the guy that he needs to work on the test and if he gets the test, infrared e for 119, we will make sure his PR goes in sounds good. So.

C

There was another PR too, that was brought up at the end of 118 I. Think Jing actually offered it from CSI. The stateful sets workload for Windows.

C

Do we want to take a quick look at that one as well? If.

A

You have the PR yeah.

C

C

Because we can move on off, oh yeah,.

E

C

E

C

C

Having a hard time sharing between, if you tell me, the number I'll write it in here, it is 8 7, 5, 8, 1,.

A

Enables himself was test run for Windows okay. So what was a problem with this PR? Is it just sitting on someone's lap so.

F

Yeah, this is D, so this PR, like we I, think infrastructure team. Some people have concerns about doing like sophistic, like in store storage class for only GC gke right now, I think the workflow is. We need to restore storage class as and when I added I only like at support, easy gke, and it does not work for other providers cloud providers and they don't want to. They don't want to do that like cloud service, fake way and I think.

F

The conclusion is: why is in storage, test Suites I think we can kind of avoid this situation, and there is also once the four tests in storage test, suite and I can work on to enable that test first and if necessary, I can also kind of copy or replicate some of has to here in the storage test, suite I.

A

Think this is a terrific idea and great a great PR gene. Thank you. So my suggestion is work with Adelina as well to make sure that we add the appropriate coverage like the additional things you mentioned, and we can hope to push it through. So who are the reviewers here that are pending okay, so these are all six storage folks. These are correct.

F

No I believe one of them is like our testing petruk sure people got.

A

F

I think AB and then is our you see them feel first reviewer is the yeah. Is the our people yeah so.

A

So can we do something like this and Adelina you correct me if you're wrong, if I'm wrong, one gene adds more tests to these to have additional coverage, let's put them into one of our. Let's put this PR from a custom branch to run those tests on on the sick windows, testing that we do if everything's green, we can provide the feedback and and help get this PR merged.

F

Yeah sure so you mentioned someone I can work with someone yeah.

A

G

Sorry yeah we can. We can do that, but we already have a testing job for for PRS and, if let me let me take a look, but if, if these tests are tab, sick windows, that will pick them up so basically just need to run to leave a comment on the PR with the with the test name for say Windows. Let me just do that when I open the.

A

Okay, so so, if we targeted sick windows, it will be tested and then, as you add more tests to it, then will expand the coverage.

B

A

Right, I'll, let you guys me, let me see see you on the PR, then Adelina.

A

Well, thank you appreciate all right. Next up, I don't have an update on Cluster, API and figuring out who can kind of lead that effort I'm working on it to see if I can get someone from VMware to kind of step up for that, and then we have one item here that was added around introducing the networking workaround Nick Nick yeah.

H

That's, that's me. That's my.

H

So there was a little bit talk about privileged containers, and this is a quick workaround I had played with and I was wondering what other people thought of. So the goal was to essentially, if you wanted a privileged container use, something like wins to launch it on the host, but then you still potentially have to pop it the networking control into the container name space. So what the what this project accomplishes?

H

Is it proxies through a domain, socket I, didn't see an easy project to already did this on Windows, so you can run a binary on the host that attaches to localhost and binds to the domain socket and then in the container you can bind to that same domain socket and pop the port on the container socket.

A

H

That's very similar to.

A

What they're doing for cube ADM, you just did it for the networking stack.

A

Are you familiar with the work we're doing on cube, ADM Aneke and we're really using wins as well, so.

H

But wins doesn't tunnel TCP connections? Does it I.

D

Don't know I thought it does, because I thought the way been explained to me was that they're doing it at the process context level, so everything will be run as on the host, so they're they're doing it at the job object level. So I, don't know why we need to do this. Just for networking I. Think then you should take a look. Yes.

A

I added there that you should definitely review this with Ben moss and get his feedback since he's the one that kind of looked into wins extensively as part of the cube ADM work, he said: yeah some slackers, okay,.

H

Since last time, I looked at wins, it had a pipe, but it didn't have tunneling of sockets through that name. Pipe so has the name pipe to communicate to a process on the host, but not tunnels, connections.

D

But if the networking is, if I understand correctly, he had a networking, but he never used it because he was going through that process to the whole post, so that kind of otherwise we have to create one for every networking storage and then this.

H

Is any process that you would like to run as privileged from the host, but what the networking to be exposed on the container side, so you can access it through, like Tanner services and stuff like that, is how to expose it in that way, because you need essentially something that tunnel. The connection to those this is not to set up kubernetes. Now, okay,.

A

H

A

To do a deep dive with Ben so.

H

A

You get a chance talk with him he's pretty busy these days, but you know if it takes him a few days to respond. Just just have some patience. Oh thanks, all right, cool I think that's the end. I think you knew by the way for bringing this up. This is good for us to to know these things. Excellent, I, don't have anything else on the agenda for today anything else anybody wants to talk, asked any questions or anything else.

A

All right: well, someone said that you should 30-minute meeting, should be 25 minutes and one-hour meeting should be 50 to keep people some time back. So we'll do that here. I already have a great day. Please stay safe and see you all. Next week, thanks bye, Thanks, Thanks.