21:00:16 #startmeeting swift 21:00:16 Meeting started Wed Jun 14 21:00:16 2017 UTC and is due to finish in 60 minutes. The chair is notmyname. Information about MeetBot at http://wiki.debian.org/MeetBot. 21:00:17 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 21:00:19 The meeting name has been set to 'swift' 21:00:23 who's here for the swift team meeting? 21:00:30 o/ 21:00:31 here 21:00:33 hi 21:00:34 o/ 21:00:37 hi o/ 21:00:54 hello 21:00:56 o/ 21:01:12 o/ 21:01:36 welcome 21:01:53 hello 21:03:26 timburke: clayg: torgomatic: courtesy ping 21:03:36 i'm here, i'm here 21:03:53 agenda is on the left hand side of... 21:03:59 #link https://wiki.openstack.org/wiki/Meetings/Swift 21:04:05 this is the 2100 meeting 21:04:26 acoles: thank you for leading the 0700 meeting earlier today 21:04:35 +1 21:04:36 log for that meeting can be seen at 21:04:38 #link http://eavesdrop.openstack.org/meetings/swift/2017/swift.2017-06-14-07.00.log.html 21:05:14 some duplicate housekeeping in this meeting, but hopefully not too much after that... 21:05:16 notmyname: NP 21:05:24 #topic housekeeping/announcements 21:05:45 the Denver PTG is open for registration, including a hotel block 21:05:53 details at 21:05:54 #link http://lists.openstack.org/pipermail/openstack-dev/2017-June/118002.html 21:06:33 swift will have a room for wednesday through friday (3 full days) in addition to the cross-project topics on monday and tuesday 21:06:59 I would like to especially encourage ops people to attend the PTG (and if you have to make a choice, choose the PTG) 21:07:12 rledisez: this includes you :-) 21:07:30 of course, i'll be here :) 21:07:42 and i'll try to bring alecuyer too (need to check with him he'll be availbale) 21:07:45 mahatic: is funny! 21:07:49 * clayg is reading logs 21:07:53 rledisez: awesome 21:07:55 y'all type slow anyway 21:08:11 based on the experiences at the first PTG and the summit/forum, the PTG is vastly more productive for getting swift community work done, and the ops feedback will be valuable at the PTG 21:08:27 I'm going to try to get one or two of the swiftstack support people to come to the PTG too 21:08:59 basically there's half a dozen people we need from the summit to go to the PTG and then we can stop doing the summit - (jrichli tell briancline) 21:09:21 has anyone seen onovy or pavel lately? we need them at the PTG too 21:09:25 hehe - i will mention the ptg to him. maybe they will support his going 21:09:33 Like I said in the 007 meeting, I don't think I'll be there. as I'm not going to self fund while unemployed, and might be too soon if a find a job soon to ask for travel :( 21:09:36 * onovy is here 21:09:50 onovy: come to Denver for the PTG! 21:09:56 mattoliverau: ahhh bummer! 21:09:59 what is Denver? 21:10:09 BUt yeah, PTG is totally worth it, better then a summit.. tho summit is in my neck of the woods so you should still come ;) 21:10:10 mattoliverau: :( 21:10:28 onovy: city in the mountains in the middle of the US. the location for the next project team gathering 21:10:31 onovy: a place with a lot of beer breweries 21:10:38 notmyname: just kiddin. Denver is too far from us to be there 21:10:46 heh 21:10:51 PTG fee is $100 still? 21:10:51 no money no funy 21:10:59 onovy: I understand, although it's unfortunate 21:11:18 zaitcev: I haven't actually clicked on the registration link yet. I assume so. can anyone confirm? 21:11:46 onovy: when budgeting swift community travel in the future, so far it seems that the PTG will be more valuable overall than the summit 21:11:48 it looks like "something" is changing in our company and i looks like we are going to deploy whole OpenStack, yea! so maybe we will be able to come to more (physical) meetings 21:11:55 cool 21:12:04 the other big openstack event is the summit in Sydney. details and CFP are at https://www.openstack.org/summit/sydney-2017/call-for-presentations/ 21:12:04 notmyname: zaitcev: yep, it says $100 21:12:09 rledisez: ack. thanks 21:12:11 so i will try to book PTG for next year 21:12:34 onovy: I think PTG next year might be in Europe, so that might help 21:12:41 tdasilva: cool! 21:12:43 tdasilva: \o/ 21:12:46 \o/ sydney, that should be one i can self fund :) 21:13:42 any questions on these events? anything I can track down for anyone? 21:14:33 notmyname: i'm just interested in knowing who's planning to go from the team 21:15:00 i heard in boston there were some doubts about travel duration vs summit duration 21:15:26 rledisez: for sydney? 21:15:27 rledisez: which one? the summit or ptg? 21:15:33 tbh, i'm thinking of keeping a talk for vancouver instead of sidney 21:15:36 summit, sidney 21:15:42 yeah, I've got that concern too 21:15:56 it's a 3 day event that takes 3 days of travel for most of the world 21:15:58 yeah 3 days is way too short 21:16:05 for flying to the otherside of the world 21:16:19 I don't know many details for sydney yet. I'll pass them along as soon as I hear them 21:16:21 er - one quick question for ptg, are there some topics on cross/inter project (strongly) related to swift? 21:16:33 mattoliverau: you mean you like to spend *more* than 3 days travelling? ;) 21:16:36 if worse comes to worse, I can be the swift represntation.. though I'd love to see people :) 21:16:40 mattoliverau: you have to do that often, right? ;-) 21:16:46 kota_: are you wondering if you need to be there for monday and tuesday? 21:16:53 notmyname: yes 21:16:56 acoles: lol 21:17:25 what do others say, for kota_? (based on the last PTG) 21:17:28 kota_: it's a good time to network with -infra people? maybe pick peoples brains about keystone sessions or oslo something? 21:17:28 jrichli: yup, so welcome to my world if you do it :P 21:17:55 personally I liked it. 21:18:21 clayg: oic, perhaps, I could expect osprofiler session that we had discussed in the 007 meeting 21:18:52 well I was only there on the tuesday. hallway track was great, and I helped in the requirments stuff, so I could go participate. So depends on what you want to get involved in 21:18:53 it's not a James Bond meeting!!! (thanks mattoliverau /sarcasm) 21:18:56 kota_: yeah that's a pretty good example! 21:18:59 kota_: hehe, "007" 21:19:09 kota_: it can be at least a good jetlag day or 2 ;) 21:19:20 mattoliverau: exactly 21:19:24 thanks! 21:19:43 ok, last bit of housekeeping is on https://review.openstack.org/#/c/468105/ 21:19:44 patch 468105 - swift - Require that known-bad EC schemes be deprecated 21:19:45 kota_: good idea 21:20:01 notmyname: your welcome ;) 007 sounds much better tho :P 21:20:10 this is the patch that makes swift processes not start if there's a known-bad EC config 21:20:29 I sent an email to the mailing lists on monday 21:20:47 notmyname: I expect you saw from 007(0) meeting the suggestion that we add some docs somewhere (if we dont already hve them) to say 'do not use this scheme' 21:21:11 acoles: yes, and mahatic left that comment in gerrit. it's a good idea 21:21:27 oh good, thanks to mahatic 21:21:43 *non-deprecated* known-bad EC config. we've got at least a couple more steps before we refuse to start if there's any known-bad EC config at all 21:21:56 timburke: yes :-) 21:22:21 timburke: could you add a sentence or two in the docs about it? 21:22:38 (like adding some super-deprecated flag that refuses to create new objects, then requiring that any known-bad policies start using *that*) 21:22:40 I don't want to spend so much time on the details of this that we never land it 21:22:41 yeah, will do 21:23:02 but in this case I don't think we can spend too much time tellng people about it 21:23:04 timburke: maybe also in swift.conf-sample if it makes sense 21:23:09 timburke: thanks 21:23:32 and I think we're ready to land it, with that small docs addition 21:24:22 personally, I feel that as a group we've discussed and over discussed this one to death. we know it's happening. we've talked about it for weeks. we talked about it in person. we talked about it months ago when kota_ first found the original bug 21:24:45 given that, I don't think we need an extra step of getting everyone to add a +1 or +2 to this patch in gerrit 21:24:49 notmyname: yup, great work! over communicate like a boss. Let's break some clusters! 21:25:10 but I could also be totally wrong. if the group thinks we need to do that, too, i'm totally ok with supporting that 21:25:13 what do you think? 21:25:25 :shipit: @acoles has swiftstack cover - good luck to the rest of ya'll suckers in the swift deployment and provisioning game! 21:25:31 lol 21:26:36 :D 21:26:38 I think its a great way to find out who else is using swift and EC out there (when they come and wonder whats wrong) :P 21:26:39 kota_: acoles: rledisez: tdasilva: jrichli: everyone ok with landing this patch as soon as timburke updates the docs? 21:26:40 clayg: he, swift works so fine, i just run a pip install -U on my cluster and go grab some cofee 21:26:44 but yes, ship it! 21:26:50 y 21:26:59 rledisez: :-) 21:27:08 sounds good 21:27:09 notmyname: ok for me 21:27:11 * timburke notices that notmyname is still looking for everyone's sign-off anyway... 21:27:19 lol 21:27:22 timburke: right ! :D 21:27:22 notmyname: yep 21:27:28 I'm in 21:27:29 overoveroveroverovershare 21:27:57 PTLin https://static.planetminecraft.com/files/resource_media/screenshot/1333/Like-A-Boss_6165401.jpg 21:28:01 great. doit 21:28:15 #topic priority reviews 21:28:17 #link https://wiki.openstack.org/wiki/Swift/PriorityReviews 21:28:36 I updated this (finally!) this morning based on what's been going on this past week 21:28:51 first of all, everyone's been doing a great job of getting some stuff landed. thanks! 21:29:06 kept checking links and "yep, it's already merged!" 21:29:35 I don't think there's a particular question on any of those patches or bugs that needs to be addressed in this meeting 21:29:47 just "keep it up! everyone's doing great!" 21:29:56 #topic open discussion 21:29:58 what else in on your mind? 21:30:09 tdasilva: mattoliverau: do you need help with your socket/centos thing? 21:30:35 well that popped up while playing with the tpool patch. 21:31:20 notmyname: yeah, if anybody has any ideas...not sure how much experience others have with server_per_port + centos 21:31:20 I'm glad tdasilva got the same thing. tho we got it using the same saio deployment (ansible-saio).. so if others have a different centos saio it would be good to confirm. 21:31:42 but we can take it to #openstack-swift channel 21:31:58 timburke: clayg: acoles: we like centos+servers_per_port. something we can help with? 21:32:03 I plan to have play with it.. but being a stay at home dad, so been a little hard to find long times infront of the laptop 21:32:32 just wanted to share I (finally!) found one real socket leak. In copy middleware. bugreport and patch are on the way 21:32:48 rledisez: \o/ 21:32:53 rledisez: yeah nice work! 21:32:54 rledisez: I don't want to say "good to hear", but thanks(?) 21:32:58 :-) 21:32:59 copy middleware, no way ;) 21:33:02 rledisez: awesome, nice one 21:33:12 also, some news on K/V memory usage, alecuyer just did the tests and gave me numbers few hours ago 21:33:24 that's great to hear 21:33:24 sweet 21:33:26 he'll send a mail to -dev tomorrow, but here a summary 21:33:29 rledisez: I'm pretty sure timburke has another socket leak bug with 411 responses... i was going to look at his WIP today but haven't yet 21:33:30 how'd they look? 21:33:39 the biggest the K/V, the more efficient it is 21:33:50 for a case of one device with 10 millions of objects 21:33:53 "biggest" by number of keys? or total bytes? 21:34:13 with only one K/V, one object takes about 39 bytes 21:34:31 with one K/V per partition, if you have 100 partitions, one object consume 50 bytes 21:34:41 with 1000 partitions, one object consume 72 bytes 21:35:01 it was expected, it's because leveldb knows to optimize but sharing common prefix 21:35:08 sounds like a normal swift scaling story: "gets better as it gets bigger" 21:35:44 rledisez: I'm looking forward to reading more about it :-) 21:36:00 yeah, the mail will contains more details (and maybe shiny graph :)) 21:36:07 ooooohhh 21:36:19 * mattoliverau loves shiny graphs 21:36:52 any other topics to bring up in today's meeting? 21:36:57 i'd like some eyes on some swiftclient patches. it's been a while since we had a release (3.3.0 was in January) so it'd be good to work toward that 21:36:59 p 455470, p 455488, p 455489, p 449771 21:37:00 https://review.openstack.org/#/c/455470/ - python-swiftclient - Stop sending X-Static-Large-Object headers 21:37:01 https://review.openstack.org/#/c/455488/ - python-swiftclient - Tolerate RFC-compliant ETags 21:37:03 https://review.openstack.org/#/c/455489/ - python-swiftclient - Skip checksum validation on partial downloads 21:37:04 https://review.openstack.org/#/c/449771/ - python-swiftclient - Buffer reads from disk 21:37:13 clayg, timburke: in my case the leak is on 413 responses 21:37:27 doesn't timur have one he resurrected to stream stdin to the swift cli? 21:37:40 did I say 411? maybe I meant 413... 21:37:55 which one is "GTFO with all those big bytes!" 21:38:00 maybe the same then, i'll write bugreport tomorrow 21:38:01 timburke: sorry I've been rubbish at reviewing recently, I'll do my best to get some dome before the next meeting! 21:38:10 clayg: it's 413, it's the same bug 21:38:45 i don't think i found a bugreport for this one, i'll re-check before posting a new one 21:38:47 notmyname: oh yeah stdin - that's a good one! 21:38:51 seems worse when it's in copy, though -- who knows how long we'll keep that socket open? 21:39:19 timburke: That buffered reads thing calls to me. I cannot believe there's no better way to do it. 21:39:25 timburke: as long as the process live, i have some experience with that :D 21:39:33 timburke: I'll add swiftclient patches to the priority reviews page. thanks for bringing that up 21:39:47 you're right. we should do a release there soon (when these land) 21:40:21 speaking of releases, for swift itself, I'll start seriously looking at one when the EC policy blocker one lands 21:40:22 https://review.openstack.org/#/c/259410/ -- hasn't be updated yet 21:40:23 patch 259410 - python-swiftclient - Support uploading to an object in swift from stdin 21:40:30 ok 21:40:50 last call. anything else from anyone? 21:41:03 I've had some nice comments from cschwede on https://review.openstack.org/#/c/365371 21:41:04 patch 365371 - swift - Add Preamble and Postamble to SLO and SegmentedIte... 21:41:08 would be nice to get more 21:41:53 nice comments.. well that's me out :P 21:42:01 there's plenty of discussion in the comments about whether the API is what we want 21:42:20 lol @ mattoliverau 21:42:21 mattoliverau: hostile sarcastic comments are also an option ;) 21:42:26 joeljwright: what's your summary of the current state of that api discussion? 21:42:32 \o/ 21:42:32 lol @ joeljwright 21:42:53 the current API adds optional pre- or post-ambles to object segments 21:43:19 clayg has suggested that maybe we just allow data as segments 21:43:49 i.e. not tied to real objects in the store 21:44:40 would be nice to get opinions (especially if someone can see a use case where pre/post amble tied to a segment is too restrictive etc.) 21:44:56 joeljwright: ok, thanks for the update 21:45:51 ok, let's end 15 minutes early :-) 21:46:30 although I say it every time, I sincerely mean it: thanks for your work on swift. you've got a lot to be proud of 21:46:36 thank you for coming today 21:46:38 #endmeeting