19:00:54 #startmeeting swift 19:00:55 Meeting started Wed Aug 27 19:00:54 2014 UTC and is due to finish in 60 minutes. The chair is notmyname. Information about MeetBot at http://wiki.debian.org/MeetBot. 19:00:56 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 19:00:58 The meeting name has been set to 'swift' 19:01:04 hello, world. who's here? 19:01:08 o/ 19:01:09 o/ 19:01:10 yo 19:01:15 hi 19:01:34 hi 19:01:35 o/ 19:01:40 ola 19:01:41 o/ 19:01:43 hello 19:01:58 #link https://wiki.openstack.org/wiki/Meetings/Swift 19:02:13 not a ton of specific things to cover, I think 19:02:20 #topic general 19:02:50 sign up for the hackathon if you're planning on attending. IIRC the hotel block (rates?) ends this week 19:03:01 #link https://swift-hackathon.eventbrite.com 19:03:19 #link https://www.eventbrite.com/e/openstack-swift-hackathon-tickets-12383404095 19:03:23 tdasilva_: heh. thanks 19:03:27 :-) 19:03:52 any questions about the hackathon? we've talked about it a bit the last 2 weeks 19:04:20 great 19:04:39 monday we tagged swift 2.1.0.rc1 19:04:53 please look in to that this week and submit any bugs you find asap 19:05:16 unless something comes up, 2.1.0 final will be next monday 19:05:46 I haven't seen anything come through that would require a backport, but I've also been traveling. 19:05:51 anyone else seen anything yet? 19:06:26 not yet 19:07:05 good :-) 19:07:19 #topic ops meetup summary 19:07:31 this week I was in san antonio at the openstack ops meetup 19:08:11 it was good. I learned a few things and got some good feedback 19:08:35 specifically relating to current things in swift, I asked about the gap analysis issues that we've been discussion 19:09:32 interestingly (to me), most of the people there are relying on distro packages and some are building their own packages. but nobody was pulling from an arbitrary commit or doing anything other than the integrated release. 19:09:46 eg they may patch some stuff in their own packages, but the base is the integrated release 19:10:33 it's important to note that the people in attendance were generally installing all of openstack and not jsut swift, so sticking with the integrated release makes more sense (a common complaint was doing upgrades) 19:11:06 any comments on 12.04/precise LTS being EOL and the push to 14.04 19:11:15 so, that relates to us in that it's good input on figuring out the impact of moving to a time based, milestone-focused release schedule 19:11:55 goodes: interestingly no. specific distro issues didn't come up. or, nobody knows or mentioned the disparity of what's in CI vs what's in prod 19:12:27 I also asked about oslo.config. 19:12:52 the consensus was a resounding "meh" with nobody saying that we should change anything 19:13:15 that is, 0 people though changing was inherently good, and everyone else was no or don't care 19:13:23 interesting! 19:13:24 so that was interesting feedback too 19:13:33 sounds like that's settled, then 19:13:47 cschwede: ya, the reasoning was (1) we have config management and (2) we use whatever you give us 19:14:21 so realizing that moving to oslo.config wouldn't remove what we currently have (ie we'd still have to support it for a long time), nobody really cares 19:14:31 at least the people in the room in san antonio 19:14:53 the last thing I asked about was logging, just to keep a future eye on oslo.logging 19:15:33 and there, the answer was that common log prefixes were nice (eg ...) 19:15:45 so that's something to consider for the future 19:16:03 it's not something new. portante brought it up a while back 19:16:32 and oslo.logging is still incubated (ie copy/paste code) so it's currently a non-starter 19:16:38 just something to keep in mind for the future 19:17:00 any other questions about the ops meetup? that's about all I was planning to say 19:17:13 (well I've got a blog post I'll publish tomorrow...) 19:17:32 notmyname: great, can’t wait to read it 19:17:59 any big concerns from the ops on operating swift? 19:18:27 cschwede: nothing came up 19:18:42 the big concerns were around cinder migrations, keystone at scale, and glance (everything) 19:18:42 :) 19:19:02 keystone might get an issue for us too (regarding the token size) 19:19:08 us=swift 19:19:26 ya, turns out that nobody there uses pki tokens. everyone used uuid tokens (in part because of the size) 19:20:08 eg one of the problems with the keystone uuid tokens is something like to look it up in the keystone db it does a table scan matching on something like 5 columns. ie slow 19:20:36 good feedback all around, but not too much that swift was specifically called out on 19:20:38 wait, you mean a single 32KB header value is a bad idea? 19:20:39 who knew? 19:20:59 chocking, I know 19:21:09 whatever 19:21:15 notmyname, a token validate only scans the table based on 2 columns, id and expires (if it is expired) 19:21:34 notmyname, the times the token table are scaned for other things are on deletes/revokes 19:21:45 morganfainberg: ah. I'm not wanting to spread falshoods. soory. just relating what I heard in the room (from non-devs) 19:22:03 notmyname, no worries, just wanted to make sure people had the info :) 19:22:06 :-) 19:22:12 notmyname, a lot of operators don't understand that bit 19:22:28 revocations was a big concern with keystone, actually 19:22:36 but we're getting off topic from the meeting 19:22:37 notmyname, we *do* plan on spending a bunch of time @ the paris summit on working on the token issue. (pki is an issue) 19:22:52 notmyname, anyway i'll let you get back on topic :) 19:22:53 morganfainberg: I'd be happy to chat later about keystone feedback from the ops meetup. I think it's all on the etherpad 19:23:03 morganfainberg: great to hear 19:23:04 notmyname, ++ hit me up for sure (about to head to lunch) 19:23:19 ok, let's move on to ec status 19:23:23 #topic ec status 19:23:29 what's been going on this week? 19:23:29 sure.. based on con call last Fri there is a newly revised reconstructor chapter in the EC design spec for revew at https://review.openstack.org/#/c/116486/ 19:23:37 and the reconstructor frameork patch is up for review too (but I just saw a py26 failure so need to look at that) 19:23:44 tsg has a few in the hopper as well 19:23:50 and trello is up to date 19:23:53 :) 19:24:09 the new pictures in the docs are very helpful 19:24:29 yeah, those are from Ian. He did a good job visualizing the scenarios we need to map to 19:24:33 if you want to see how ec reconstruction works (ie the counterpart to replication), check it out 19:24:44 feedback welcome! 19:25:00 peluse: what's the most important thing outstanding right now? 19:25:04 we still have design work to do on the reconstructor side too... 19:25:12 I think the most important thing is movement on the existing patches 19:25:18 #link https://trello.com/b/LlvIFIQs/swift-erasure-codes 19:25:20 tsg and recon framework 19:25:22 peluse: reviewS? 19:25:25 yes 19:25:29 peluse: ok 19:25:36 I'm hoping that I can do that this week 19:25:41 cool 19:25:45 tsg, you there? 19:25:51 and maybe mattoliverau zaitcev and torgomatic can too ? 19:25:55 from my side - 1. EC policy patch merged 2. wsgi 100-continue headers patch merged to eventlet repo 3. EC quorum_size, generic payload footer support, PUT/GET in flight 19:25:59 Will do! 19:26:07 rock on! 19:26:16 tsg: nice 19:26:16 that's great news on the eventlet side 19:26:23 and pyeclib/liberasurecode nearing 1.0 (next week) 19:26:38 yeah, just worry about your item #3 right now :) 19:26:45 :) 19:26:55 tsg: torgomatic: is there anything to summarize on the footer/multipart changes? or should we just read gerrit? 19:26:55 JK, that's great news 19:27:15 just read gerrit, there's a small novel in there 19:27:23 torgomatic: heh, ok 19:27:29 most of it is on gerritt and torgomatic, has an update from a few minutes ago :) 19:27:36 s/,/ 19:27:57 torgomatic: sortof related to EC stuff (in that it's important for ec perf), what's the current status for the zerocopy patches? 19:28:13 torgomatic: for now, I will post the naive non-MIME version of the footer patch while we continue to discuss that topic 19:28:18 * peluse just remembered that was supposed to be on his review list for last week.... argh 19:28:28 notmyname: gathering dust, mostly 19:28:42 torgomatic: I'll look this week FWIF for sure 19:28:50 ok, need reviews there too 19:28:56 I still don't have benchmark results for the PUT one on real hardware though 19:29:14 torgomatic: plans to get some? 19:29:37 peluse: if possible; there's hardware contention at my office (as usual, there's not enough hardware to go around) 19:30:08 torgomatic: cool, I'll still look at the code but I think that's needed for approval of course 19:30:16 yup 19:31:47 ok, let's move on to patches 19:31:49 #topic patches 19:32:00 the services patch for swiftclient landed 19:32:05 thanks for working on that 19:32:16 note that there is a new dependency there 19:33:04 and I've got a todo item to send an email to the ML about global requirements. 19:33:22 #link https://wiki.openstack.org/wiki/Swift/PriorityReviews 19:33:37 I think we've already called out a few high priority things to review this week 19:33:48 the zero copy patches and some of the EC ones 19:34:20 I want to see the affinity+delete one and the global cluster replication one land by juno 19:34:42 that is, their respective functionality, if the patches as they currently are don't work 19:35:16 OK, I'll look at the zero copy and affinity delete this week for sure 19:35:20 and the keystone one too. basically, the stuff that is important to ensure success for people grabbing swift out of the integrated release 19:36:26 I haven't looked at the review queue (beyond the gerrit emails) since this weekend. is there anything else to call out? any other questions that can be addressed here? 19:36:28 was hoping portante would wrap up the keystone ACL one since its a followup on to the v2. I already +2'd it 19:36:37 and he also looked at the v2 one 19:37:59 seems as if many (all?) swift patches are blocked because of a pbr bug in the gate right now 19:38:12 https://review.openstack.org/#/c/116775/ <-- this should unblock it. -infra is discussing it now 19:38:27 peluse: iirc, portante won’t be able to do a lot of reviews atm (he said something in that direction in last weeks meeting). i will have a look at the keystone acl patch 19:38:37 cschwede: thanks 19:39:06 and right. portante probably won't be able to look at it. he's being asked to focus on some other things at red hat instead of swift 19:39:24 no prob, just mentioned it becuase he was already setup w/v2 19:39:45 any other patches? 19:39:47 peluse: yep, i had a look at the v2 too, so i think i can continue with acls 19:39:50 what else is there at red hat besides swift? 19:39:50 ...to discuss 19:40:00 peluse: that is a terrific question 19:40:31 #topic open discussion 19:40:39 anything else to bring up? 19:41:05 oh, 19:41:18 there's a ML thread about a different pattern for the design summit 19:41:50 there are more projects than ever and less space, so the current proposal is to not have full-day dedicated rooms to different projects 19:42:01 the mailing list topic is "Design Summit reloaded" 19:42:06 so check that out 19:42:06 oh yeah, schedule posted and keving and I will be representing the good work of everyone on the EC side with a talk on Mon I think 19:42:15 peluse: nice! 19:42:39 peluse: awesome! 19:42:57 hope to see a lot of you there, notmyname has like 100 sessions too I think :) 19:43:14 at lest 19:43:20 *least 19:43:37 that being said, I think the early registration price ends soon 19:43:50 but you all should have a free code since you contributed 19:44:04 but IIRC from the last summit, that may expire with the early registration 19:44:09 so if you're going, register today! 19:44:19 deadline is tomorrow for early registration! 19:44:23 ...and that's all I got. anyone else? 19:44:46 thanks for coming. have a good day 19:44:49 later 19:44:52 #endmeeting