| *** annabelleB has quit IRC | 00:39 | |
| *** zerocoolback has joined #kata-dev | 00:48 | |
| *** zerocoolback has quit IRC | 01:20 | |
| *** zerocoolback has joined #kata-dev | 01:50 | |
| *** annabelleB has joined #kata-dev | 02:10 | |
| kata-irc-bot | <eric.ernst> @bergwolf around? | 03:22 |
|---|---|---|
| kata-irc-bot | <eric.ernst> Received another query on a tired subject: ```Is there a document or an enumeration of the limitations of 9pfs ? Specifically these are the things we need in order to use our log and other file system apis: •fstat,stat •ftruncate •fallocate(FALLOC_FL_KEEP_SIZE) •ioctl(FS_IOC_FIEMAP) •fallocate(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE) •open(),close(),create(),fsync() •closedir(),readdir(),mkdir,opendir • | 03:23 |
| kata-irc-bot | syscall(SYS_renameat2) •preadv,pwritev ``` | 03:23 |
| kata-irc-bot | <bergwolf> @eric.ernst I did know one important limitation of 9pfs -- if you do open()->delete()->fstat(), the last one will fail because 9pfs does not support tracking deleted-open fd, while POSIX requires that and some user apps does rely on it. | 03:30 |
| kata-irc-bot | <eric.ernst> I wonder if we should have a 9p limitations document posted that we can point people at, and add to. | 03:30 |
| kata-irc-bot | <bergwolf> I didn't try IO to a deleted-open fd, but I think it will fail as well. -- If that is the case, many logger library will fail as well | 03:30 |
| kata-irc-bot | <bergwolf> yes I agree. | 03:31 |
| kata-irc-bot | <eric.ernst> That's what this failure is -- someone trying to do logger | 03:31 |
| kata-irc-bot | <eric.ernst> host fs | 03:31 |
| kata-irc-bot | <bergwolf> oh, you did encounter it already | 03:31 |
| kata-irc-bot | <bergwolf> we try to avoid 9pfs as much as possible in our production system | 03:31 |
| kata-irc-bot | <eric.ernst> this is someone else using kata/clear. | 03:31 |
| kata-irc-bot | <eric.ernst> Sounds similar though. So, in your experiencce, how do you handle this kind of design pattern? | 03:32 |
| kata-irc-bot | <bergwolf> we use distributed block storage for both rootfs and volumes | 03:33 |
| kata-irc-bot | <bergwolf> There are many ways to hotplug them to guest. and hypercontainer does have a storage description exposed to users | 03:34 |
| *** annabelleB has quit IRC | 03:34 | |
| kata-irc-bot | <bergwolf> kata still lacks that kind of user interface. The device hotplug API can support it now but we still need a user facing interface | 03:35 |
| kata-irc-bot | <bergwolf> Possibly we should try to modify the OCI spec for it | 03:40 |
| kata-irc-bot | <eric.ernst> in this case, its pretty custom, right, then? ie, the workload is modified considerably to work within the vm-container design? | 03:40 |
| kata-irc-bot | <bergwolf> nop, it's complete transparent for containers. graph drivers are internal of docker/containerd/crio. we need them to pass down the raw information instead of just mounted host path. | 03:41 |
| kata-irc-bot | <bergwolf> hypercontainer is able to use block storage directly because hyperd and frakti pass them to runv. | 03:42 |
| kata-irc-bot | <bergwolf> and flexvolumes allows k8s to pass storage info directly to frakti | 03:44 |
| kata-irc-bot | <eric.ernst> ah, i see, ok. | 03:44 |
| kata-irc-bot | <eric.ernst> hyperd/frakti is converting host dir to a block device then? | 03:44 |
| kata-irc-bot | <eric.ernst> that's the part I'm missing. | 03:44 |
| kata-irc-bot | <eric.ernst> the current user says "I want this directory from my host, thanks" | 03:45 |
| kata-irc-bot | <bergwolf> in hyperd's pod json -- which is a pod description for users, one can just say I want this block device for my volume | 03:46 |
| kata-irc-bot | <eric.ernst> sure, okay, that's fine then. | 03:46 |
| kata-irc-bot | <bergwolf> in kata, we need something similar I think | 03:47 |
| kata-irc-bot | <eric.ernst> K8S CSI should help here some? | 03:47 |
| kata-irc-bot | <eric.ernst> AFAIU they even describe the intended permissions (single writer, multiple reader, etc) | 03:47 |
| kata-irc-bot | <bergwolf> we used flexvolume drivers | 03:47 |
| kata-irc-bot | <eric.ernst> I'll be honest, I am not familiar with flexvolume :| | 03:48 |
| kata-irc-bot | <bergwolf> basically it allows you to do what ever you want when k8s wants to mount/unmount a block storage | 03:49 |
| kata-irc-bot | <bergwolf> so in frakti, we just do not mount at all | 03:49 |
| kata-irc-bot | <bergwolf> I think kata should try to do the same, -- because the current approach of mounting a block device on both host and guest, is broken by design. | 03:50 |
| kata-irc-bot | <eric.ernst> naive here - is CSI essentially replacing flexvolumes? | 03:52 |
| kata-irc-bot | <bergwolf> TBH, right now the two storage handling methods -- both 9pfs and the device mapper hack, are not really production ready. | 03:52 |
| kata-irc-bot | <bergwolf> Not a complete replacement, flexvolumes still allow you to do more flexible things | 03:53 |
| kata-irc-bot | <eric.ernst> ok. I wish we had someone with experience in this area on the team (oh, hi Tao). | 03:53 |
| kata-irc-bot | <bergwolf> ,:) | 03:54 |
| kata-irc-bot | <eric.ernst> Yeah, I took some basic notes on CSI and came away with this was a good method going forward, and that the storage spec was adding some good flags (rw, multiple rw, ro, etc) | 03:54 |
| kata-irc-bot | <eric.ernst> regardless we need to make sure kata works well for both flex and CSI imo | 03:55 |
| kata-irc-bot | <bergwolf> yes, I agree CSI is the direction to go | 03:55 |
| kata-irc-bot | <bergwolf> it's more standardized and flex is still for `hey, come and do whatever you want` | 03:56 |
| kata-irc-bot | Action: eric.ernst regrets we don't have PTG or another extended period with a quorum to go through this and come up with a solid plan. | 03:57 |
| kata-irc-bot | <bergwolf> If you want f2f discussion, you can try to catch @xu. He also knows all the details and you can find him in all these conferences in the season | 03:59 |
| kata-irc-bot | <eric.ernst> he just talks about running and not eating too many cupcakes :P | 04:00 |
| kata-irc-bot | <eric.ernst> (I hope he sees this) | 04:00 |
| kata-irc-bot | <eric.ernst> We have less experience in this area but are catching up with respect to CSI. I just want otm ake sure we can have a constructive conversation where the outcome is a plan on next steps, etc. | 04:01 |
| kata-irc-bot | <bergwolf> rob and eat all the cupcakes then he has nothing to worry about ;) | 04:01 |
| kata-irc-bot | <bergwolf> yup, it can really be part of kata's next step plan | 04:02 |
| kata-irc-bot | <bergwolf> walk through all the stack from k8s to CRI/CSI to runtime, add proper storage descriptions whenever necessary, to allow kata to use a block storage directly | 04:03 |
| kata-irc-bot | <eric.ernst> this is something to dwell on soon: ```kata's next step plan``` | 04:04 |
| kata-irc-bot | <bergwolf> Maybe we should have a list of items, to say that these are the missing features/requirements to make kata production ready -- which IMO is the most important thing for next step. | 04:05 |
| kata-irc-bot | <eric.ernst> Yes. | 04:07 |
| kata-irc-bot | <bergwolf> Some of them off my head: 1. stable branch maintenance (Thanks Eric!, it's almost done) 2. version compatibility (or rolling upgrades, live upgrades etc.) 3. block storage support w/o mounting on the host | 04:08 |
| kata-irc-bot | <eric.ernst> I have a hard time keeping track of our lists :slightly_smiling_face: | 04:08 |
| kata-irc-bot | <eric.ernst> I think #3 is the most interesting here. | 04:08 |
| kata-irc-bot | <bergwolf> put it on github ;) | 04:08 |
| kata-irc-bot | <eric.ernst> https://github.com/orgs/kata-containers/projects/12 | 04:08 |
| kata-irc-bot | <eric.ernst> if there's an issue, its there. | 04:08 |
| kata-irc-bot | <bergwolf> nice | 04:08 |
| kata-irc-bot | <eric.ernst> hmm, maybe a new github project... | 04:09 |
| kata-irc-bot | <bergwolf> I'm not familiar with github projects tho | 04:09 |
| kata-irc-bot | <bergwolf> learning/reading | 04:09 |
| kata-irc-bot | <eric.ernst> I would love to have another person to formulate/push a project with. | 04:10 |
| kata-irc-bot | <eric.ernst> I started this one because I kept getting asked about roadmap, and I thought listing out major features was a starting point for that. | 04:10 |
| kata-irc-bot | <eric.ernst> But.... yeah.. I could use someone to collaborate with here. | 04:10 |
| kata-irc-bot | <eric.ernst> the more established projects I watch usually use projects and/or a consistent and dilgent usage of labels | 04:13 |
| kata-irc-bot | <bergwolf> hmm, you mean kata -> production ready as a new project, or just #3? | 04:13 |
| kata-irc-bot | <eric.ernst> WDYT? I think it could be its own (it's a subset of this). | 04:14 |
| kata-irc-bot | <bergwolf> Does it support sub-projects? | 04:15 |
| kata-irc-bot | Action: bergwolf new to github projects | 04:16 |
| kata-irc-bot | <eric.ernst> ooh. no, just make a new project. | 04:17 |
| *** sjas_ has joined #kata-dev | 04:34 | |
| *** sjas has quit IRC | 04:37 | |
| kata-irc-bot | <caihaomin> hi @eric.adams @james.o.hunt my pullapprove has some bugs. I try to `enable repo` but got an `500 http error` any ideas？ | 06:32 |
| *** jodh has joined #kata-dev | 07:27 | |
| kata-irc-bot | <james.o.hunt> @caihaomin - sorry, what pullapprove? | 07:42 |
| kata-irc-bot | <zhangwei555> His LGTM doesnt count in CI @ts | 07:47 |
| *** gwhaley has joined #kata-dev | 07:50 | |
| kata-irc-bot | <james.o.hunt> Ah - thanks @zhangwei555! Which repository are you referring to @caihaomin? | 07:51 |
| *** marcov has quit IRC | 07:52 | |
| kata-irc-bot | <zhangwei555> Runtime Repo, take this as an example: https://github.com/kata-containers/runtime/pull/649 | 07:52 |
| *** davidgiluk has joined #kata-dev | 08:04 | |
| kata-irc-bot | <bergwolf> well, pullapprove is in sleep today | 08:09 |
| kata-irc-bot | <bergwolf> just ignore it | 08:09 |
| kata-irc-bot | <bergwolf> as I did ;) | 08:09 |
| kata-irc-bot | <james.o.hunt> That does look odd - @wei, @tallclair and @teawater (non maintainers) *can* approve, but @caihaomin and @caoruidong (also non maintainers) cannot :( | 08:10 |
| kata-irc-bot | <bergwolf> my approval doesn't count either today. | 08:10 |
| kata-irc-bot | <bergwolf> so I think it's pullapprove's fault | 08:10 |
| kata-irc-bot | <james.o.hunt> Yep - we have seen some issues with pullapprove. We can override it if we need to :slightly_smiling_face: But we should also review all the team memberships "Real Soon Now" (TM) :slightly_smiling_face: | 08:11 |
| kata-irc-bot | <james.o.hunt> @bergwolf - wtf? :( | 08:11 |
| kata-irc-bot | <james.o.hunt> yep | 08:11 |
| *** marcov has joined #kata-dev | 08:12 | |
| kata-irc-bot | <james.o.hunt> @archana.m.shinde - ptal at https://github.com/kata-containers/osbuilder/pull/144 - it would be awesome to have Ubuntu osbuilder support... :slightly_smiling_face: | 08:13 |
| *** marcov has quit IRC | 09:10 | |
| *** marcov has joined #kata-dev | 09:12 | |
| *** lsm5 has quit IRC | 09:41 | |
| *** lsm5 has joined #kata-dev | 09:43 | |
| *** lsm5 has left #kata-dev | 09:44 | |
| *** marcov has quit IRC | 09:45 | |
| *** gwhaley has quit IRC | 11:01 | |
| *** devimc has joined #kata-dev | 12:00 | |
| *** tonyb has quit IRC | 12:02 | |
| *** gwhaley has joined #kata-dev | 12:04 | |
| *** tonyb has joined #kata-dev | 12:19 | |
| *** devimc has quit IRC | 12:34 | |
| *** devimc has joined #kata-dev | 13:10 | |
| *** zerocoolback has quit IRC | 13:19 | |
| *** changcheng has quit IRC | 13:48 | |
| *** changcheng has joined #kata-dev | 13:49 | |
| *** jugs has quit IRC | 14:37 | |
| *** jugs has joined #kata-dev | 14:38 | |
| *** sjas_ is now known as sjas | 14:39 | |
| kata-irc-bot | <james.o.hunt> Kalling all Kool Kats! http://lists.katacontainers.io/pipermail/kata-dev/2018-August/000357.html | 14:40 |
| *** devimc has quit IRC | 14:55 | |
| *** annabelleB has joined #kata-dev | 15:00 | |
| *** annabelleB_ has joined #kata-dev | 15:08 | |
| *** annabelleB_ has quit IRC | 15:08 | |
| kata-irc-bot | <graham.whaley> @salvador.fuentes (and all), just fyi, I've created a proxy test metrics CI job so I can test out the scripts at least, and at minimum get us up with memory footprint/density tests http://jenkins.katacontainers.io/job/kata-metrics-proxy-ubuntu-18-04-PR/ | 15:42 |
| kata-irc-bot | <graham.whaley> I predict a little bit of script massaging early next week, as they were written to have a per-bare-metal server 'what to expect' config file on each machine, which we won't get on cloud deploys - so, I need to figure a new way of doing that (previously the config file was defined by the uname - so now maybe I need some wildcarding...) | 15:43 |
| *** annabelleB_ has joined #kata-dev | 15:50 | |
| *** annabelleB has quit IRC | 15:57 | |
| *** annabelleB_ is now known as annabelleB | 15:57 | |
| *** kata-irc-bot has quit IRC | 15:58 | |
| *** kata-irc-bot has joined #kata-dev | 15:58 | |
| *** kata-irc-bot has quit IRC | 16:00 | |
| *** kata-irc-bot has joined #kata-dev | 16:00 | |
| *** devimc has joined #kata-dev | 16:04 | |
| kata-irc-bot | <salvador.fuentes> cool, thanks for the notice @graham.whaley | 16:10 |
| kata-irc-bot | <anne> One more little reminder that if you have burning question for the Architecture Committee Candidates, now is the time to submit it to the kata-dev mailing list! | 16:34 |
| *** gwhaley has quit IRC | 16:57 | |
| *** jodh has quit IRC | 17:01 | |
| *** kata-irc-bot has quit IRC | 17:15 | |
| *** annabelleB has quit IRC | 17:19 | |
| *** kata-irc-bot has joined #kata-dev | 17:20 | |
| *** kata-irc-bot has quit IRC | 17:21 | |
| *** kata-irc-bot has joined #kata-dev | 17:22 | |
| *** kata-irc-bot has quit IRC | 17:35 | |
| *** eernst has joined #kata-dev | 17:36 | |
| *** annabelleB has joined #kata-dev | 17:37 | |
| *** kata-irc-bot has joined #kata-dev | 17:38 | |
| *** kata-irc-bot has quit IRC | 17:40 | |
| *** kata-irc-bot has joined #kata-dev | 17:41 | |
| *** kata-irc-bot has quit IRC | 17:42 | |
| *** kata-irc-bot has joined #kata-dev | 17:43 | |
| *** kata-irc-bot has quit IRC | 17:43 | |
| *** kata-irc-bot has joined #kata-dev | 17:44 | |
| *** kata-irc-bot has quit IRC | 17:44 | |
| *** kata-irc-bot has joined #kata-dev | 17:44 | |
| *** annabelleB has quit IRC | 17:58 | |
| *** annabelleB has joined #kata-dev | 18:04 | |
| *** annabelleB has quit IRC | 18:30 | |
| *** annabelleB has joined #kata-dev | 18:46 | |
| *** eernst has quit IRC | 18:47 | |
| *** eernst has joined #kata-dev | 18:49 | |
| *** eernst has quit IRC | 18:50 | |
| *** eernst_ has joined #kata-dev | 18:50 | |
| *** eernst_ has quit IRC | 18:54 | |
| *** eernst has joined #kata-dev | 19:03 | |
| *** annabelleB has quit IRC | 19:32 | |
| *** davidgiluk has quit IRC | 19:33 | |
| *** eernst has quit IRC | 19:54 | |
| *** eernst has joined #kata-dev | 19:56 | |
| *** devimc has quit IRC | 20:36 | |
| *** annabelleB has joined #kata-dev | 20:55 | |
| *** eernst has quit IRC | 21:32 | |
| *** jugs has quit IRC | 21:33 | |
| *** jugs has joined #kata-dev | 21:34 | |
| *** annabelleB has quit IRC | 22:00 | |
| *** annabelleB has joined #kata-dev | 22:21 | |
| *** annabelleB has quit IRC | 22:51 | |
| *** annabelleB has joined #kata-dev | 22:58 | |
| *** annabelleB has quit IRC | 23:18 | |
| *** eernst has joined #kata-dev | 23:22 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!