21:00:13 #startmeeting swift 21:00:14 Meeting started Wed Jan 9 21:00:13 2019 UTC and is due to finish in 60 minutes. The chair is notmyname. Information about MeetBot at http://wiki.debian.org/MeetBot. 21:00:15 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 21:00:17 The meeting name has been set to 'swift' 21:00:19 who's here for the swift team meeting? 21:00:23 o/ 21:00:24 hi 21:00:52 hi o/ 21:01:24 welcome 21:01:31 rledisez couldn't decide on a greating so used both :P 21:01:32 o/ 21:01:36 clayg: timburke: tdasilva: ? 21:01:43 hi 21:02:01 good to see everyone here 21:02:06 mattoliverau: as I always say, why choosing? ;) 21:02:09 o/ 21:02:14 welcome back and happy new year 21:02:20 lol 21:02:29 notmyname: thanks, you too! 21:02:32 lol 21:02:42 2019!!! 🎉 21:03:02 it's here, whether you're ready or not :-) 21:03:24 all right. first meeting of the new year. let's recap to see where we are 21:03:26 * mattoliverau notes that he's been up all night with a sick baby, so might be a little sleep deprived and therefore extra sarcastic this morn ;P 21:03:47 swift 2.20.0 has been released. best release ever 21:03:54 poor baby 21:03:55 \o/ 21:03:58 yeah :( 21:04:05 yey 21:04:06 mattoliverau: sorry to hear that. good luck 21:04:11 2.20.0 was solid 👍 21:04:20 "swift 2.20 is here, whether you're ready or not :-)" \o/ 21:04:24 heh 21:04:34 2.21 will be *even better* ;-) 21:04:50 we're in the middle of the openstack stein cycle. the openstack schedule is ... 21:04:54 #link https://releases.openstack.org/stein/schedule.html 21:05:22 key points are the end-of-february client release (ie python-swiftclient) 21:05:49 and the final swift release needed the first week of april (at the latest) 21:06:15 note that there is *not* an openstack event before then. the ptg and summit have combined and will be in denver in may 21:06:45 also, looking far-future, towards the end of the year there will be a summit in china (no details published that I've seen) 21:06:50 they must've had a 3 for 1 deal for denver :P 21:07:11 though this next one will be at a different venue 21:07:42 more immediately, along with the 2.20.0 release, we landed a lot of backports. timburke tells me that he's got more; I'd like to have a tag on the stable branches by the end of next week 21:07:43 April 29 - May 1 21:07:43 oh china, interesting 21:08:25 good idea. If we have backports lets release em 21:08:42 China might be cool 21:08:46 I need to check that the priority review page is up to date 21:08:49 #link https://wiki.openstack.org/wiki/Swift/PriorityReviews 21:08:56 * mattoliverau might keep talking to stay awake :P 21:08:57 Deven... when is Denver? Is it a design summit? are we going? 21:09:15 clayg: yes. we should all go. there will be developer sessions/rooms 21:09:31 april 29 -> may 1 21:10:15 in the latter half of the week? 21:10:28 #link https://www.openstack.org/summit/denver-2019/ 21:10:33 all the details are there 21:10:55 thx 21:11:10 that's everythign I can think of about current schedules for stuff. 21:11:24 #link https://www.openstack.org/ptg 21:11:29 ok, ptg is May 2 - 4, 2019 21:11:36 bah, Saturday 21:11:40 ah, ok. I see 21:11:41 idk... my kid's first birthday is 4/30... 21:12:02 yay 1st birthday! 21:12:06 cake cake cake! 21:12:06 yeah, you don't want to miss that :-) 21:12:56 the Big Things we've been talking about are still around and still in progress 21:13:10 if the PTG part is 2 -4 that's all we really need to be at right? 21:13:35 mattoliverau: 👍 21:13:41 we've got LOSF and task queue and py3 port are the most definite things 21:14:14 notice CFG for summit closes in two weeks from today 21:14:15 mattoliverau: clayg: definitely those days. one of the summit days may be interesting. hard to say right now. probably is different for each person 21:14:25 oh yeah, i should go look over https://review.openstack.org/#/c/629610/ ... 21:14:27 patch 629610 - swift - WIP py3: object server - 1 patch set 21:14:28 s/CFG/CFP 21:14:39 sure, it'll come down to my employer anyway 21:14:55 cool cool! 21:15:19 rledisez: how's LOSF? what are your next plans for it? 21:15:23 Who will go to the PTG or summit? 21:15:45 m_kazuhiro: I suspect most or all of us will be at the ptg part. and some will be at the summit too 21:16:03 m_kazuhiro: also, welcome! :-) 21:16:07 I plan too.. but again employer.. and I have to remind them as I missed the last few :( 21:16:07 great! 21:16:15 notmyname: right now alecuyer is taking a step back, working on some other projects. it's good enough for our production. I think at that point we need a plan for LOSF in Swift. what will happen, how, when, … 21:16:33 of course, we still want it upstream 21:16:36 rledisez: ah, ok. that's interesting 21:16:44 I think the rest of us want it upstream too :-) 21:16:56 +1 21:17:16 the current code run on 2.18, we will work soon to adapt it to 2.20 (or 2.21, depends how fast it's released :)) 21:17:21 rledisez: so are you saying that there are not any more obvious design changes. you are fairly happy with how it's working in prod now? 21:18:08 notmyname: yes, as far as we know, we've found no more bugs, and it matches all our needs/expectation on performance 21:18:19 there could be some perf improvment on some part, but nothing major 21:18:30 ok 21:18:45 do you have any plans on how to propose it upstream? 21:19:01 or rather, plans on how to get it landed upstream? 21:19:47 I think we are waiting for a plan. alex cannot do that alone, because there will be of course reviews that will make changes to the design 21:19:57 ok 21:19:58 exciting 21:20:20 so we (all of us) need to work together on finding that plan 21:20:23 so, first would be reviews of the design, then working on code/helping us making the code matching quality expectation of reviewers, etc... 21:20:26 and tests... 21:20:32 notmyname: +1 21:21:08 what is your opinion of the code now? is it sufficiently tested and documented to land upstream? do you think the code quality matches what's in swift? 21:21:29 I'm not sure how to phrase that question. mostly wondering if you expect it to change much once proposed 21:21:39 ...or what work is needed before it's proposed 21:21:57 or differences between what needs to be proposed vs what's been proposed previously 21:22:01 notmyname: I don't think so. it's not much documented, and the tests are pretty limited (the golang RPC is fully tested I think, not the diskfile). but at least, func tests are passing :) 21:22:13 ok 21:22:56 there is not a wrong answer. I'm only trying to manage expectations for myself and everyone. so we understand if this is something nearly done or will need a few months of work from the whole team 21:23:33 rledisez: anything else to share about LOSF right now? 21:24:10 notmyname: no, as I said alex took a step back, so it didn't change much in the last few weeks 21:24:14 ok 21:24:29 we'll come back to LOSF, of course, but for this meeting I want to move on 21:24:41 next up is the same questions for the task queue work 21:24:51 m_kazuhiro: and kota_: how is this project going? 21:25:09 and what are your next plans for it? 21:25:25 i saw m_kazuhiro has updated the patches, but that's all I know so far. 21:26:08 First I want to merge Patch 601950. This is for general task queue patch. 21:26:09 https://review.openstack.org/#/c/601950/ - swift - Enable to configure object-expirer in object-serve... - 11 patch sets 21:27:00 I updated and fixed this patch at December. So I think the patch is ready for review now. 21:27:57 ok, that's good to hear 21:28:14 is there anythign else to add at this point, or should the conversation happen in gerrit about that patch? 21:29:01 I want to know who can review the patch. 21:30:21 I don't think I can answer that right now. I want to get a sense of all the stuff going on first, so that we all know 21:30:37 the final Big Thing that's going on is the py3 port 21:30:57 doesn't look like zaitcev made it today. but I know timburke has been helping with this some 21:31:02 (is there anything timburke doesn't do?) 21:31:35 and py3 is something that matters a lot to you mattoliverau, right? that is, it matters a lot to suse 21:31:43 like it does for redhat and canonical 21:32:08 it looks like there is a proposal today to port the object server to py3 21:32:10 yeah, it's one of the goals that SUSE is interested in OpenStack wide, for obvious reasons. 21:32:57 mattoliverau: is it something they're asking you to work on specifically? 21:33:21 So something I'll try and get back into, but probably not really until I'm back from Germany and Christ Church. But will try and review whats there in the mean time. 21:33:26 ok 21:34:05 once it's done, I think both canonical and suse need to send redhat a box of cupcakes to thank redhat for porting swift ;-) 21:34:18 mmmmm cupcakes 21:34:20 i want cupcakes! ;-) 21:34:30 lol 21:34:37 +1 :) 21:34:55 the py3 todo list is essentially tracked in the tox.ini file 21:34:57 #link https://github.com/openstack/swift/blob/master/tox.ini#L29-L34 21:35:07 that's ported stuff 21:35:36 that makes it seem like we havne't done much! really, it's https://github.com/openstack/swift/blob/master/tox.ini#L29-L87 21:36:03 noice 21:36:05 and two of those are directories 21:36:09 yeah 21:36:14 lol 21:36:24 wait, three! 21:36:29 in fact, we *want* that list to be very short because it means that we've eg ported everythign under common 21:37:14 so the list will keep getting longer until it shrinks by a lot (when a whole directory is ported). then it will grow again. then it will shrink to zero when everything is ported and we require py3 compat for 100% of the code 21:37:41 one of the real interesting things will be to compare coverage reports between py2 and py3 unit test runs 21:38:39 fwiw, i currently see 33143 total statements on py2, and 25384 on py3 21:38:57 (with 92% and 74% coverage, respectively) 21:39:13 the py3 work will be constant (with people getting more anxious about it as time goes by) until it's done 21:39:18 so we still have some 7k lines that never even get imported 21:39:38 anything else on the py3 work? 21:39:54 I think we've probably covered that topic enough 21:40:23 which leads us to... what? what else is there to bring up this week in the meeting? 21:40:44 have we missed any big existing ongoing work? anything else to bring up today? 21:41:07 https://review.openstack.org/#/c/629301/ would be good to have 21:41:08 patch 629301 - swift - Verify client input for v4 signatures - 1 patch set 21:41:30 wooo s3api 21:41:59 yes, that one would be good. it's been tracked as a private security issue until recently. we opened it to treat it as a security-hardening opportunity instead of a critical open vulnerability 21:42:08 woo s3api! 21:42:08 but it needs to be closed asap 21:42:14 see https://bugs.launchpad.net/swift3/+bug/1765834 21:42:15 Launchpad bug 1765834 in OpenStack Security Advisory "Need to verify content of v4-signed PUTs" [Undecided,Incomplete] 21:42:43 if there is one, single important patch right now, it's likely that one 21:42:58 I raised https://bugs.launchpad.net/swift/+bug/1811044 yesterday, cause I was pinged about it from 2 different people on 2 different openstack deployement tools internally yesterday. Nothing major, but something to think about. So will put some words on the bug report about it. 21:42:59 Launchpad bug 1811044 in OpenStack Object Storage (swift) "Allow_versions config option in Container-server is deprecated but removing can lead to some version containers failing to version objects" [Undecided,New] - Assigned to Matthew Oliver (matt-0) 21:43:00 ic 21:43:07 i really like the "In the future, Swift will hopefully grow a way to perform SHA256 verification on the object-server." idea 21:43:46 ie as I actually thnk about it. And not spitballing ideas 21:43:47 mattoliverau: ah, ok. is this something you're working on a patch for, or do you need somethign from the rest of us first? 21:44:01 nah, I can do it. Just letting y'all know. 21:45:05 mattoliverau: did they "Configure the versioned_writes middleware in the proxy-server instead" per the deprecation warning? or did they just forget that step? 21:45:05 mattoliverau: awesome, thanks for chasing that down! 21:45:58 no you can do it. but if you don't "reset" the versioning on a container that was set pre 2.4.0 the middleware wont check the old metadata anaymore 21:46:03 or am i misunderstanding the bug? i must admit, i only skimmed it... 21:46:15 oh... yeah, that's no good 21:46:22 I mean, they added the middleware properly. 21:46:59 oh yeah, speaking of audit watchers... https://review.openstack.org/#/c/212824 lives! 21:47:00 patch 212824 - swift - Let developers/operators add watchers to object audit - 12 patch sets 21:47:15 whoa 21:47:26 \o/ 21:47:38 I noticed, which is why I added at as an idea ;P 21:48:14 ok, anything else from anyone? otherwise, let's go ahead and end the meeting 21:48:19 to share what are my plans on Swift, I'll continue to work on SSYNC this year. my goal is to make it as fast as possible in 4 major steps: 21:48:19 1. concurrency per partitions (https://review.openstack.org/#/c/613987/ => need more tests) 21:48:19 2. delete reverted object ASAP (patch half written) 21:48:19 3. fsync() batch (need more thinking) 21:48:19 4. better coordination between servers (need a lot of thinking) 21:48:20 patch 613987 - swift - SSYNC: enable multiple SSYNC connections per job - 4 patch sets 21:48:55 btw, thx mattoliverau for the review on ^ 21:49:05 rledisez: nice! 21:49:29 nps, I'll try and loop back round on it 21:49:38 rledisez: clayg and i should post some of the ideas we've had around 4... 21:49:55 we drew out some diagrams a while back... i know i took a picture 21:50:26 timburke: that would be very interesting. we've got some, but it was just talking around a coffee, nothing much 21:50:26 +1 21:51:12 yeah please, share diagrams! 21:51:33 :-) 21:51:41 I'm really mad about SSYNC not really being un-interruptable - the whole protocol falls over if it hits a 4/5XX responses when it's streaming data :'( 21:51:56 everything will be better with gRPC - which we get with LOSF right!? 21:52:22 clayg: that point is addressed in 2 21:52:34 if i understand you well 21:52:55 probably not :P 21:53:17 haha, you're so mysterious :D 21:53:48 in point 2, because the receiver maintain a list of success/failure and send it back to the sender, we can ignore and continue, then the sender clean what was a success 21:53:57 well maybe... depending on how drastically you restructure the interface when you let ssync remove data files as it reverts (which agreed, would be great) 21:53:59 (that's the patch i'm working on) 21:54:23 probably a good idea to move that conversation into -swift 21:54:27 i'll end this meeting now 21:54:35 thank you everyone for coming 21:54:40 notmyname: thanks! 21:54:49 I'm excited about what the swift team will be able to do in 2019 :-) 21:54:55 +1 21:54:56 here here! 21:54:56 thanks for your work on the project 21:55:02 #endmeeting