21:00:49 #startmeeting swift 21:00:50 Meeting started Wed Dec 16 21:00:49 2015 UTC and is due to finish in 60 minutes. The chair is notmyname. Information about MeetBot at http://wiki.debian.org/MeetBot. 21:00:51 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 21:00:54 The meeting name has been set to 'swift' 21:01:05 hello, everyone. who's here for the swift meeting? 21:01:09 o/ 21:01:12 o/ 21:01:14 o/ 21:01:16 o/ 21:01:17 o/ 21:01:18 o/ 21:01:20 \o 21:01:25 o/ 21:01:25 hello 21:01:32 hi 21:01:37 hi 21:01:47 sorry got lost looking at the new gerrit 21:02:05 yup. new gerrit was just deployed 21:02:11 have fun, and good luck :-) 21:02:24 mattoliverau: wow, new gerrit 21:02:38 agenda for this week 21:02:40 #link https://wiki.openstack.org/wiki/Meetings/Swift 21:02:53 I'm not sure how, but I think we got even longer than last week! 21:03:15 tdasilva: are you here? 21:03:49 hello 21:03:52 first topic is one that's a little different 21:03:58 #topic glance driver 21:04:11 #link http://lists.openstack.org/pipermail/openstack-dev/2015-December/081966.html 21:04:22 there was a mailing list message about this (linked above) 21:04:43 the glance team is looking for support in maintaining their drivers for various backend storage things 21:04:48 one of them is the swift driver 21:05:07 here's the current status 21:05:08 #link https://etherpad.openstack.org/p/glance-store-drivers-status 21:05:50 the basic idea is that each driver does the Right Thing for that particular storage system, and the glance team has neither the time nor the expertise to know them all 21:06:05 so I think there are a few options (listed in our meeting agenda) 21:06:27 1) we could move the swift driver from glance into some new repo that we maintain 21:06:48 2) we could have one person from our community as the point person for all questions and reviews 21:07:17 3) we could build a joint glance-swift-store review team made up of glance core and swift core, but the code live in glance 21:07:22 4) we could do nothing 21:07:51 how much code is it? 21:07:55 I don't like option 4, just because I don't think that's good or healthy for openstack in general to let this cross project stuff fall by the wayside 21:08:18 mattoliverau: hav eyou had a chance to look since we talked about it? 21:08:38 not yet 21:08:41 and +1 for do something :) 21:09:05 mattoliverau: who were we talking to from glance about ti? 21:09:14 I volenteered to be contact person so swift support wont be depreciated. But I like option 3 21:09:16 flaper87 ? 21:09:22 yep 21:09:25 yeah, that was him 21:09:47 there's really two options, I think 21:09:51 mattoliverau: having more than one person maintaining seems better than option 2 21:09:52 in option "2" is only "one person". this is not good. 3 is better 21:09:58 mostly around where the code lives 21:10:11 acoles: that's what I told him 21:10:22 does the code stay in glance and we poke at it (perhaps with commit rights) or do we move the code to be a project under the swift ubbrella? 21:10:31 https://github.com/openstack/glance_store/tree/master/glance_store/_drivers/swift <-- this is it? 21:10:40 onovy: yes 21:10:44 I think if we want people to use it, it should stay with the project that needs it (glance) 21:10:53 mattoliverau: +1 21:10:59 +- 1k lines of code 21:11:11 mattoliverau, +1 too 21:11:13 yeah, I agree 21:11:29 and I also think that it has to be a group (rather than a person) mainaining it 21:11:39 +1 for #3 21:11:42 option 3 makes sense 21:11:57 seeing a lot of support for that 21:12:17 so keep it inside glance, and create new group from glance+swift members 21:12:19 that was flaper87's idea. it's a good one 21:12:24 notmyname: my hpe glance core colleague, stuart, advised me against option 1, fwiw 21:12:35 acoles: ah, interesting. why? 21:13:08 I think it would actually be a core team for glance drivers, and swift-core would be a member with an agreement that we don't poke the non-swift stuff 21:13:50 notmyname: because glance should own how glance talks to other core services, just like swift owns keystoneauth, nova has glance driver etc 21:14:20 acoles: ok 21:14:32 notmyname, maybe it's possible to setup permissions for this. just to make changes in some subdirs, etc. don't know 21:15:00 notmyname: the part of this that concerns me is the plan to change the driver API, at a time when they say they don't have sufficient resource to maintain 21:15:01 onovy: perhaps. that was actually a conversation in tokyo (wrt neutron, I think), and IIRC the conclusion was "trust people" 21:15:09 acoles: :-) 21:15:09 ok 21:15:26 notmyname: and in vancouver :) 21:15:44 lifeless: yeah, you were teh outspoken one there (with wisdom) ;-) 21:15:56 so I'll talk to flaper87 again and say that we want to have a joint core for glance drivers. unless anyone is opposed? 21:16:07 +1 21:16:35 +1 21:16:43 +1 21:16:46 +1 21:16:52 +1 21:16:56 +1 21:17:01 ack 21:17:32 #agreed swift-core team requests core rights for glance-drivers to help maintain the swift driver in the glance repo 21:17:37 great! 21:17:41 if making the joint team, do we need one +2 from glance core at least? 21:18:07 kota_: meh. we can work those details out. I'm not too worried about someone doing the wrong thing (intentionally) 21:18:08 i think swift and swiftclient core are good for this. 21:18:24 ok 21:18:24 FYI flaper87 ^ (for your scrollback later) 21:18:42 Well that's our recommendation, flaper87 will take it to the glance meeting tomorrow (if I remember correctly) 21:18:54 mattoliverau: ah, right. thanks 21:19:11 ok, hold on to your seats. time to move fast 21:19:25 #topic testing testr tests 21:19:35 the testr functests landed 21:19:45 thank you everyone who helped get it through! 21:20:02 acoles: I believe you have a final issue on it for follow-on patch? 21:20:34 oh, maybe I'm misrememerign? 21:20:35 some of os-testr options did not work, i found that switching to unittest2 fixed that 21:20:41 oh yeah 21:20:48 i just went to find the patch but...gerrit... 21:20:51 lol 21:20:57 acoles: patch 258578 21:20:57 gmmaha: https://review.openstack.org/#/c/258578/ - Fix func test --until-failure and --no-discover op... 21:20:59 gerrit is back 21:21:09 gmmaha: but the UI is ....new... 21:21:17 notmyname: aah yeah! :) 21:21:22 gmmaha: thanks 21:22:01 acoles: doesn't unittest2 also bring in that cool subtest thing? 21:22:22 notmyname: idk. tbh it was luck that led me to that fix. 21:22:37 notmyname: luck and the exception trace :) 21:22:50 gerrit sign in not working for me 21:22:54 yeah https://docs.python.org/3/library/unittest.html#distinguishing-test-iterations-using-subtests 21:23:04 so there might be other nice reasons to use unittest2 anyway 21:23:09 ok, so that's the status there 21:23:26 #topic patch 206105 21:23:27 notmyname: https://review.openstack.org/#/c/206105/ - Use entrypoints for storage policy implementation ... 21:23:33 this is the entrypoints patch 21:23:56 bringing it up again this week because it still needs reviews 21:24:12 I was suppose to take a better look at that, but have failed, but now that testr is in, i'll find some time 21:24:20 beyond the technical things for the patch itself, what does it imply we support in swift 21:24:28 the gerrit comments are good 21:24:41 so look at that 21:24:43 please 21:24:59 #topic other patches to look at 21:25:35 janonymous isn't online, but he asked for reviews on patch 227855 about keepalive settings 21:25:35 notmyname: https://review.openstack.org/#/c/227855/ - Eventlet green threads not released back to pool 21:26:02 ho: your keystone RBAC tests? did they all get messed up from the testr patch? 21:26:08 hello, sorry i'm late 21:26:41 notmyname: yeah, i will rebase them today. 21:26:58 ho: thanks. not that testr has landed, I'd like to spend some time on them 21:27:08 other patches that need review 21:27:13 notmyname: thanks! 21:27:18 copy middleware: patch 156923 21:27:19 notmyname: https://review.openstack.org/#/c/156923/ - Refactor server side copy as middleware 21:27:24 this one is blocking other work! 21:28:15 notmyname: i'm working on it. right now focusing on decoupling versioning from copy. Hoping to have something posted by friday 21:28:20 thanks 21:28:21 sorry i didnt get to that yet. did the copy_hook stuff get removed? 21:28:29 tdasilva: ok. thanks 21:28:35 acoles: not yet, but i'm working on it 21:28:42 concurrent reads patch 117710 21:28:42 notmyname: https://review.openstack.org/#/c/117710/ - Add concurrent reads option to proxy 21:29:06 mattoliverau: what's the status here? 21:29:12 I've marked it as WIP while I finish some left over work due toe clays last patch 21:29:28 I'll try and get that up by Friday (or at least pre-xmas) 21:29:40 ok, thanks 21:30:12 small segments in SLOs patch 252096 21:30:12 notmyname: https://review.openstack.org/#/c/252096/ - Allow smaller segments in static large objects 21:30:39 just needs one more +2 and it really helps out users, especially after the SLO-ranges feature 21:30:40 oops, here o/ 21:31:02 i started looking at that, didnt get too far yet though 21:31:12 cschwede: kota_: would either of you have a chance to look at that one? 21:31:19 * peluse says 'oh yeah, a meeting'... 21:31:20 sure 21:31:22 notmyname: yep, will do 21:31:25 thansk 21:31:56 oh, i already starred that. so it’s already on my list ;) 21:32:04 onovy: peterlisak: the two of you have a couple of patches mentioned 21:32:10 patch 238799 21:32:11 notmyname: https://review.openstack.org/#/c/238799/ - Change schedule priority of daemon/server in config 21:32:15 here i just want to know if: swift should support it natively or not. as i wrote in gerrit, it's usefull inside swift itself 21:32:51 that is a great question, especially for briancline, ahale, and other ops 21:33:09 i tried to discuss on irc few times but no answer ... so I put in agenda and hope to get more attention ... 21:33:30 it can't be done inside init script, because we have swift-init 21:33:44 only solution now is to use cron job which is really really weird 21:33:48 peterlisak: thanks 21:34:12 +1 on that. don’t need to modify any other scripts 21:34:25 hi here o/ yeah thats a good idea imo 21:34:45 onovy: I actually asked our ops folks about this recently. we already control these in production through other means, so we're fine with this going in as long as its optional, which it certainly seems to be 21:34:56 briancline, yep, it's 21:35:07 fyi ... most of lines are changes in doc 21:35:18 peterlisak: yeah 21:35:21 yes that's true. patch itself is simple 21:35:34 i can see situations later where it'll help us to have it in config 21:35:54 peterlisak: onovy: ok, so based on views from cschwede and ahale and briancline, definitely a good idea. thanks for working on it 21:36:29 perfect, so the second one 21:36:32 the other you had listed on the agenda is patch 229532 21:36:33 notmyname: https://review.openstack.org/#/c/229532/ - swift-init: SIGKILL daemon after kill_wait period. 21:36:34 and pls maybe test on other distros ... i just tested on debian 21:37:14 i need to do this hack in init script: http://anonscm.debian.org/cgit/openstack/swift.git/tree/debian/FOOTER.init.in in debian 21:37:20 which is really weird too :) 21:37:24 onovy: how does this compare to the swift-oldies script? 21:37:31 notmyname, problem is somewhere else 21:37:36 when i do swift-init .. stop 21:37:39 that daemons doesn't stop 21:37:45 so you can't start - port in use 21:38:14 we have this problem in production 21:38:49 so, now when i want to stop/restart daemon i signal process with sigterm/sighup (using swift-init), wait 60 seconds and then sigkill 21:39:01 if parent process didn't die 21:39:26 my patch moves this func from init script to swift-init itself, which is better place 21:39:31 onovy: i'll have a look through after meeting and likely +1 this, we've definitely seen this where the process will get stuck 21:39:49 briancline, perfect 21:39:52 as long as it doesn't affect graceful shutdowns, right? 21:40:06 briancline, and that's a question 21:40:21 from my point of view we should kill parent process if it doesn't day in grace period too 21:40:25 die 21:40:31 in the situations where we've seen it get stuck i believe when we strace the process it's just sitting there trying to read() for data that never comes (we're talking hours at least) 21:40:48 didn't tried to strace it 21:41:17 how it should work: send sigterm/sighup to PARENT process. parent process die, childs keep running -> this is fine 21:41:34 but when parent process didn't die -> it's stuck, kill it all after grace period with sigkill 21:41:47 all=>process group 21:42:07 is there any condition to reproduce? 21:42:21 well, i would think on a graceful shutdown (swift-init shutdown / reload) it still should respect that a child proc could still be alive because its serving a request (keepalive sessions stick around like this).. the way eventlet doesn't expose enough information about the connections/etc seems like it's best to err on the side of caution on graceful shutdowns, but for hard stops (swift-init stop) it should do what your patch is proposing, i would think 21:42:23 ho, no :/ it hangs "sometimes" 21:42:48 briancline, so you are suggesting what peterlisak suggests too 21:42:56 when there are gracefull shutdown, kill only parent, not group 21:42:56 onovy: thanks. 21:43:15 same for reload (gracefull shutdown + start) 21:43:16 right? 21:43:42 well 21:44:00 what i'm thinking is don't try to force a kill on a graceful shutdown, only on the hard shutdown 21:44:14 how can you do gracefull reload then? 21:44:23 or, i guess we could set the grace period really high for graceful shutdowns, but that would mean hard shutdowns also use that same timeout too 21:44:48 I'm interested in the problem. i will investigate it. 21:45:06 same here, i think i need to revisit the patch and test with it a bit 21:45:29 onovy: thanks for bringing it up. ho and briancline, thanks for looking 21:45:42 ok, thanks you. me and peterlisak will respon to your questions 21:45:52 I'll see if anyone here (swiftstack) has any opinions on it too 21:46:03 notmyname, yeah, thanks 21:46:06 kota_: you had 2 patches listed on the agenda 21:46:18 yes 21:46:22 patch 198633 and patch 198634 21:46:22 notmyname: https://review.openstack.org/#/c/198633/ - Last-Modified header support on HEAD/GET container 21:46:23 notmyname: https://review.openstack.org/#/c/198634/ - Support last modified on listing containers 21:46:45 that allows to make container-listing show the last-modified timestamp for each container. 21:46:52 I talked in vancouver 21:47:15 i think they are too closer to land, I got +2 sometimes 21:47:29 but rebasing affects to loose it :/ 21:47:37 kota_: i need to cycle back round to those 21:47:48 so briging up to ask someone to re-review :) 21:48:08 ok, thanks 21:48:11 not so a problem and new changes 21:48:59 #topic general stuff 21:49:22 first up, are we going to meet during the next two weeks? there's often a lot of vacation (at least in the US) 21:49:49 so who would be here if we had a meeting the next two weeks? 21:50:08 notmyname: 23rd maybe, 30th no 21:50:26 maybe I'm 21:50:26 * onovy says: Merry Christmas and happy new year :) so no 21:50:32 next week is Xmas eve for me, I might be be available tho I do have to drive 8 hours that day, the week after I wont be around. 21:50:42 ok, then how about this 21:50:47 mattoliverau: you always have to be first :) 21:50:57 i'd be around the 23rd, the off for a while 21:51:06 let's not schedule any more meetings for this year, and stay active in IRC if there are questions 21:51:12 lol 21:51:30 +1 21:51:31 +1 21:51:33 notmyname: +1 21:51:35 ok 21:51:36 +1 21:51:37 lot's of time for reviews :-) 21:51:41 have an enjoyable holiday everyone 21:51:51 +1 21:51:55 +1 21:52:14 +2 21:52:19 any updates from anyone else on ongoing stuff? crypto? container sync? container sharding? ec-related patches? 21:52:23 On that note then, I raised a new patch late yes arvo (for me) patch 258280 which once I push a new patchset up later this morning, I think would b e good to review cause it fixes up a problem an Op was having yesterday. 21:52:23 mattoliverau: https://review.openstack.org/#/c/258280/ - Pass HTTP_REFERER down to subrequests 21:52:50 mattoliverau: right! that's a good one. fixes what looks to be an annoying (to users) bug 21:52:59 mattoliverau: good catch 21:53:09 it allows you to set the same referer on a *LO segments container rather then having to have it global readable 21:53:47 1 lline fix, multi line in functional tests :) 21:54:19 last bit of stuff from me is about the hackathon. acoles and I are still working on logistics. will update people as we have more info. it's in bristol, UK the week of feb 28 21:54:32 anything else from anyone? 21:54:35 Making progress with acoles on container sync (thanks acoles!) otherwise, we have found a juicy one today: all client calls catch ClientException where they should catch urllib.HTTPError exceptions 21:54:38 Container sharding, minwoob and bjkeller have been working on fixing unit tests that do not agree with the sharding changes! 21:54:45 i have something, but can talk in channel if needed 21:54:46 just fyi: swauth 1.1.0 is out 21:54:49 is jrichli here to talk about encrypting content-type? 21:54:54 acoles, yes 21:54:59 some of us working on crypto have made a decision/proposal that needs agreement from everybody 21:54:59 jrichli: oops cross-posted 21:55:09 jrichli: go 21:55:16 (fast ;-) 21:55:17 we are suggesting that we no longer encrypt the content-type for our first version of encryption 21:55:19 blmartin et el: thanks guys 21:55:28 encrypting the content-type has caused many issues - and much time. and we still don't have solutions for all the problems that it causes 21:55:39 content-type in the container DB or in the object metadata too? 21:55:40 blmartin: now that testr's in I'll loop back around on sharding trello :) 21:55:53 mattoliverau: nice! 21:56:07 notmyname: it would not be encrypted at rest anywhere. (right, both) 21:56:12 ok 21:56:20 notmyname: object metadata creates the harder problems 21:56:46 not what I would have expected 21:56:48 we have considered whether it will be ok to implement this encryption later, and to the best of our knowledge, the answer to that is yes 21:56:49 notmyname: with multipart responses, we don't know how long they are a priori when decrypting 21:57:54 if someone can get me opinion for patch 251151 ; just say localtime or gmt, that's all :) 21:57:55 onovy: https://review.openstack.org/#/c/251151/ - Show local time in swift-recon in replication part 21:57:55 notmyname: because we don't know how may encrypted content-type headers will come in the multiple parts 21:58:17 s/may/many 21:58:28 acoles: ok. I'll have to look more to understand :-) 21:58:31 any concerns? thoughts? I think that we will want to confirm with clayg and torgomatic when they are around 21:58:44 but in general I trust that acoles and jrichli are making the right call 21:59:14 jrichli: yeah, there's the general concern of "not everything is encrypted", but that's now new 21:59:15 i see this as tactical scope reduction in order to make progress to some working artifact 21:59:35 acoles: which is exactly how jrichli described it to me. and that's reasonable 22:00:00 I think reducing scope is ok, things can be added later 22:00:09 we're at time 22:00:15 thanks :-). I guess we will go with that for now, but again, i will feel better when there is a thumb's up from clayg and torgomatic 22:00:24 thank you for coming today and thanks for working on swift 22:00:33 jrichli: yes. ping them in -swift or tomorrow 22:00:42 #endmeeting