| kata-irc-bot | <xu> should we update kernel for the SACK CVE-2019-11477? | 00:34 |
|---|---|---|
| kata-irc-bot | <eric.ernst> Yes, please @xu | 01:51 |
| *** tmhoang has quit IRC | 02:04 | |
| kata-irc-bot | <eric.ernst> If someone can send a PR we can get it in. Might be worth a quick release even | 02:05 |
| kata-irc-bot | <eric.ernst> https://github.com/kata-containers/runtime/pull/1817 :slightly_smiling_face: | 02:13 |
| kata-irc-bot | <xu> @eric.ernst you're super prompt, :+1: | 02:35 |
| kata-irc-bot | <xu> Maybe we should do a quick stable release for it @bergwolf :thinking_face: | 02:43 |
| kata-irc-bot | <bergwolf> Skipping the ongoing one and do a quick jump to next stable? | 02:47 |
| kata-irc-bot | <eric.ernst> I’m fine with it if you can cover it for 1.6/1.7 | 02:47 |
| kata-irc-bot | <eric.ernst> Arm patches need rebase tough. | 02:48 |
| kata-irc-bot | <eric.ernst> Though* | 02:48 |
| kata-irc-bot | <xu> Another thought is to pick the fix commits for CVE-2019-11477/11478/11479, but I think bumping kernel version is better though we need @penny.zheng or @hejianet to update the arm64 patch. | 02:50 |
| kata-irc-bot | <eric.ernst> Yes. I don’t want to cherry pick, as we are overdue for stable update anyway for kernel | 02:52 |
| kata-irc-bot | <penny.zheng> I will update asap, @xu follow you time_zone8) | 02:56 |
| kata-irc-bot | <xu> @penny.zheng I am in the same timezone as you, though my day time is a bit longer :sunny: | 02:57 |
| kata-irc-bot | <penny.zheng> are you sure is just a bit??? never mind, my english is poor. 哎嘿嘿 | 02:59 |
| kata-irc-bot | <xu> My latitude is about 10 degrees higher than you, that's all | 03:03 |
| kata-irc-bot | <eric.ernst> and my day is even longer | 03:23 |
| kata-irc-bot | <xu> yes, you live close to the north pole. | 03:25 |
| *** eernst has quit IRC | 04:22 | |
| *** sameo_ has joined #kata-dev | 04:34 | |
| *** tmhoang has joined #kata-dev | 05:56 | |
| *** tmhoang has quit IRC | 05:59 | |
| *** lpetrut has joined #kata-dev | 06:06 | |
| *** tmhoang has joined #kata-dev | 06:06 | |
| *** pcaruana has joined #kata-dev | 06:15 | |
| *** sameo_ has quit IRC | 06:45 | |
| *** jodh has joined #kata-dev | 07:35 | |
| *** sameo_ has joined #kata-dev | 07:43 | |
| *** dklyle has quit IRC | 07:52 | |
| *** gwhaley has joined #kata-dev | 08:03 | |
| *** davidgiluk has joined #kata-dev | 08:06 | |
| *** igordc has quit IRC | 08:20 | |
| kata-irc-bot | <graham.whaley> North pole, pah! I'm sat at a healthy lattitude of 53.8008° N :slightly_smiling_face: A friend of mine moved to a Swedish village that really is inside the arctic circle this year, up at 67.8558° N Heh, no nightime for them right now!! | 09:06 |
| kata-irc-bot | <xu> cool, so he may saw the aurora! | 09:15 |
| kata-irc-bot | <graham.whaley> yep, every now and then :slightly_smiling_face: I've seen it twice !! Once was from a plane when flying back from a holiday :slightly_smiling_face: | 09:19 |
| *** dinechin has joined #kata-dev | 10:34 | |
| *** gwhaley has quit IRC | 10:58 | |
| *** lpetrut has quit IRC | 11:35 | |
| *** devimc has joined #kata-dev | 11:56 | |
| *** gwhaley has joined #kata-dev | 12:31 | |
| *** fuentess has joined #kata-dev | 12:31 | |
| gwhaley | davidgiluk: I generated the report of 9p vs virtiofs from gmmaha data - attached the pdf to slack though, so let me fire you a copy in a pm.. | 12:56 |
| davidgiluk | Thanks! | 12:59 |
| *** pcaruana has quit IRC | 13:00 | |
| *** pcaruana has joined #kata-dev | 13:00 | |
| davidgiluk | gwhaley: OK, so the reads are great, writes need a bit of tweeking | 13:03 |
| kata-irc-bot | <graham.whaley> writes I suspect hit the caches in the guest a lot - I think iirc 9p might cache in the guest a lot on writes, so we don't see the real to-host measures maybe :( | 13:04 |
| kata-irc-bot | <graham.whaley> the tests are fairly 'tweakable' on the fio settings if we need... | 13:04 |
| kata-irc-bot | <graham.whaley> v.easy to run @davidgiluk, honest ;) (well, once you help me fix any bugs in the scripts...) | 13:04 |
| davidgiluk | yeh and we've got a bunch of cache settings as well | 13:05 |
| kata-irc-bot | <graham.whaley> the report generator will make reports comparing n>2 data sets as well, but, the graphs get pretty dense (especially the fio ones) at that point. So, fairly easy to go around the tweak/run/reportgen cycle | 13:05 |
| *** dinechin has quit IRC | 13:48 | |
| *** dinechin has joined #kata-dev | 14:24 | |
| *** dinechin is now known as c3d | 14:34 | |
| *** pcaruana has quit IRC | 15:14 | |
| *** dklyle has joined #kata-dev | 15:16 | |
| *** sameo_ has quit IRC | 15:53 | |
| *** jodh has quit IRC | 15:53 | |
| *** pcaruana has joined #kata-dev | 16:02 | |
| *** sameo has joined #kata-dev | 16:28 | |
| *** gwhaley has quit IRC | 16:52 | |
| brtknr | davidgiluk: gwhaley: is there a way for me to see the pdf without signing up to slack? | 17:01 |
| davidgiluk | graham.whaley: You could post it to either the virtio-fs or kata lists | 17:04 |
| davidgiluk | brtknr: The summary is virtio-fs reads are great, the random write latency is a bit worse (probably due to the time to map the dax) | 17:11 |
| kata-irc-bot | <sebastien.boeuf> davidgiluk: around? | 17:54 |
| davidgiluk | Yep | 17:55 |
| davidgiluk | although I'll be grabbing dinner in a few mins | 17:55 |
| davidgiluk | seb: How can I help? | 17:56 |
| kata-irc-bot | <sebastien.boeuf> quick question, am I crazy or the virtiofsd daemon is not reporting the protocol feature MQ? | 17:56 |
| kata-irc-bot | <sebastien.boeuf> and I realized that get_queue_num is not supported | 17:56 |
| davidgiluk | seb: We did disable multi-queue at one point because we realised there's some races in the filesystem code | 17:57 |
| davidgiluk | seb: In the daemon that is | 17:57 |
| kata-irc-bot | <sebastien.boeuf> but the guest driver expects at least 2 queues, one hiprio and one or more requests, right? | 17:57 |
| davidgiluk | seb: yes I think that's right, I think we were counting the 'multi' in multi queues as all of the ones other than the hiprio - but stefanha is the one who knows the spec details more | 17:58 |
| davidgiluk | seb: Best check that one with Stefan | 17:59 |
| kata-irc-bot | <sebastien.boeuf> ok thanks Dave | 17:59 |
| kata-irc-bot | <sebastien.boeuf> davidgiluk: enjoy dinner :slightly_smiling_face: | 18:00 |
| kata-irc-bot | <sebastien.boeuf> stefanha: around? | 18:00 |
| stefanha | Hi, one sec. Call. | 18:02 |
| kata-irc-bot | <sebastien.boeuf> stefanha: of course, ping me when you have a sec | 18:04 |
| stefanha | sebastien.boeuf: Back now! This is an ambiguous area of the spec. I think it's only used for vhost-user-net devices (the spec refers to queue pairs, i.e rx/tx). The QEMU contrib/ vhost-user code doesn't use the feature bit at all. | 18:06 |
| stefanha | sebastien.boeuf: I'll investigate a little more and will look at DPDK. | 18:07 |
| stefanha | (Basically I think virtiofsd doesn't need this feature bit) | 18:07 |
| *** tmhoang_ has joined #kata-dev | 18:08 | |
| *** tmhoang has quit IRC | 18:09 | |
| stefanha | sebastien.boeuf: I've compared QEMU and DPDK vhost-user. This bit should always be set, otherwise QEMU won't fetch the number of queues from the vhost-user device backend. | 18:12 |
| stefanha | sebastien.boeuf: I'll post a vhost-user.txt spec patch to update the wording so it doesn't imply virtio-net. | 18:13 |
| stefanha | davidgiluk: I'll send a libvhost-user patch to always set this bit too. That will fix virtiofsd. | 18:15 |
| kata-irc-bot | <sebastien.boeuf> stefanha: yes exactly, the bit should be set (we're talking about MQ right?), but in case it's set, then the VMM should call into GET_QUEUE_NUM | 18:16 |
| stefanha | Ah, hang on, libvhost-user doesn't implement the GET_QUEUE_NUM message yet :( | 18:16 |
| kata-irc-bot | <sebastien.boeuf> that's my problem | 18:16 |
| kata-irc-bot | <sebastien.boeuf> I just realized that this morning | 18:17 |
| kata-irc-bot | <sebastien.boeuf> so virtiofsd is a bit off the spec right now | 18:17 |
| stefanha | sebastien.boeuf: libvhost-user needs the device (virtiofsd, vhost-user-scsi, etc) to tell it how many queues. | 18:18 |
| stefanha | sebastien.boeuf: I'll see how to add that to the code. | 18:18 |
| kata-irc-bot | <sebastien.boeuf> that sounds great | 18:18 |
| kata-irc-bot | <sebastien.boeuf> and yeah my understanding is that virtiofsd should provide to libvhost-user the num of queues and then qemu will know about the num of queues by calling into get_queue_num, but this has the dependency on having the bit MQ set | 18:19 |
| kata-irc-bot | <sebastien.boeuf> I think that's the whole dependency chain :slightly_smiling_face: | 18:19 |
| stefanha | sebastien.boeuf: libvhost-user can unconditionally set the MQ bit and default max_queues to 1. | 18:20 |
| stefanha | I think it won't be too bad. fingers crossed :-) | 18:21 |
| kata-irc-bot | <sebastien.boeuf> stefanha: but wait, if you report only 1 queue, that's wrong, right? because the VMM implementation will expect to have at least 2 queues supported | 18:24 |
| kata-irc-bot | <sebastien.boeuf> 1 hiprio + 1 request | 18:24 |
| *** igordc has joined #kata-dev | 18:27 | |
| *** tmhoang_ is now known as tmhoang | 18:28 | |
| *** igordc has quit IRC | 18:36 | |
| *** davidgiluk has quit IRC | 19:09 | |
| *** sameo has quit IRC | 20:42 | |
| *** pcaruana has quit IRC | 20:45 | |
| *** devimc has quit IRC | 21:02 | |
| *** david-lyle has joined #kata-dev | 22:01 | |
| *** dklyle has quit IRC | 22:01 | |
| *** david-lyle is now known as dklyle | 22:01 | |
| *** dklyle has quit IRC | 23:02 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!