14:01:07 #startmeeting glance 14:01:08 Meeting started Thu Jan 17 14:01:07 2019 UTC and is due to finish in 60 minutes. The chair is jokke_. Information about MeetBot at http://wiki.debian.org/MeetBot. 14:01:09 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 14:01:12 The meeting name has been set to 'glance' 14:01:15 #topic roll-call 14:01:16 o/ 14:01:17 o/ 14:01:19 o/ 14:01:23 o/ 14:01:26 o/ 14:01:32 o/ 14:01:42 #link https://etherpad.openstack.org/p/glance-team-meeting-agenda 14:02:18 #topic updates 14:02:29 really quick two 14:02:41 first of all I have requested a room for us in PTG 14:03:12 \o/ 14:03:18 So hoping we can get interested parties joining and we can get some work planning moving on there 14:03:26 yey 14:03:59 second thing I have in my mind. IIRC the submissions for talks for the summit is closing 14:04:18 so if you have something planned make sure you have submitted your proposals 14:04:46 ack 14:04:58 #topic release updates 14:05:04 abhishekk stage is yours 14:05:14 thanks 14:05:24 we are in R-12 weeks 14:05:38 S2 milestone has just passed 14:06:08 It's time for us to be prepared for non-client library, i.e. glance-store release 14:06:36 (We have one important patch up for glance-store, rosmaita added that to discussion) 14:06:57 Periodic jobs, all is well 14:07:15 everything is green at the moment 14:07:21 thanks 14:07:54 One more thing I almost forgot. You might have noticed we have RP column now in the gerrit 14:08:01 over to you again 14:08:17 saw that 14:08:34 So keep an eye on that and please review those pathes flagged with +votes as priority 14:08:59 ack 14:09:32 and also cores should have rights to vote on that column so do not expect me being only one flagging those 14:09:33 on that note, do either of you have a link to the default viz change patch? 14:09:46 I can dig it out for you 14:10:05 looks like that will be up to me and sean since you and abhishek are co-authors 14:10:20 https://review.openstack.org/#/c/628430/ 14:10:26 thanks 14:11:08 i have to build a new devstack this weekend, i will test the migration on my old one, and then check the fresh install on the new one 14:11:25 cool 14:11:29 great 14:11:38 ok lets move on 14:12:21 #topic unicode object TypeError from swift store hashing 14:12:28 rosmaita: you wanted to take this one 14:12:43 yeah, just wanted to discuss the patch real quick 14:13:03 i am wondering whether the fix should be in the readers instead of in the driver? 14:13:13 also, i am worried about not having a test 14:13:43 because i am confused about how this error happens in the first place 14:13:50 I'm even more worried about the fact that doing the encoding likely breaks anyone who upgrades to py3 14:13:53 i thought we were dealing with byte streams everywhere 14:14:31 So py27 does not use utf-8 encoding and using that with py3 _will_ give different hash 14:15:21 which means that the hash verifications will fail once you have cross version platform or the python version changes over upgrade 14:16:06 rosmaita: yeah, with the md5, the problem is that it's hashing string, not bytes 14:17:48 you'd think we'd see other problems, because a raw byte stream has to have all sorts of crap in it that would make it illegal utf-8 14:18:22 I don't think so 14:18:59 well utf-8 uses the whole range of 8bits so it's always possible ot encode to that 14:19:29 I just have no idea how the md5 actually works on py2.7 which has default encoding of 'ascii' that is only 7 bits 14:19:56 i thought it just worked on the raw bytes 14:19:57 and I think we need to verify that before jumping the gun and adding some encoding there that will produce different results 14:20:47 see, the problem you mentioned earlier is that the string -> byte translation is different for py2 and py3 14:20:53 but if we don't need to translate 14:21:01 then we have the exact same bytes 14:21:25 i am confused about what in our code thinks the image data is a unicode stream 14:21:44 as said, we need to figure this out before just going and starting to encode shit and breaking whole world with it 14:22:02 right, i think we need to have an online meeting so we can talk it out 14:22:04 rosmaita: ++ and wy this seems to break only with the swift driver 14:22:15 i admit, i am talking out of my butt here since i havne't had time to dig in 14:22:22 there must be something different there we're missing 14:22:23 but the whole situation is a bit weird 14:22:31 jokke_: ++ 14:23:21 Who ever is interested to dig into this with us, please let me know 14:23:32 yeah, i would feel much better if we knew why this is only the swift driver that's affected 14:23:33 I'll try to schedule bluejeans for us for tomorrow 14:23:49 i'm on pto tomorrow, will be travelling 14:23:54 oh ok 14:24:02 but you don't need me if you want to get started 14:24:12 for monday then unless we find the reason out before that 14:24:16 because encoding is used in swift driver only 14:24:37 abhishekk: so our swift driver actually encodes the data? 14:24:39 :o 14:24:42 i'm not clear why it is used there at all 14:24:56 jokke_, I guess so 14:25:28 this is a very weird situation 14:25:31 ok lets dig into this and get someone with swift/swiftclient knowledge involved 14:25:51 if there is no reason to encode it, we should get rid of that encoding in the driver 14:25:53 i think it must be on our end, because swift itself is all py2 14:26:01 and iirc, they have no plans to make it py3 14:27:09 swift is not so popular now, is replaced by ceph 14:27:10 anyway, schedule a monday meeting at a good time for abhishek, i can get up early to make up for not being available tomorrow 14:27:15 #action jokke will try to get hold of some swift folks to see if there is actual reason to do encoding on the data and we decide how we move on after 14:27:21 I am confused, its not on the master 14:28:01 ok, lets not burn our meeting time for the investigation and find way forward offline 14:28:08 https://github.com/openstack/glance_store/blob/master/glance_store/_drivers/swift/store.py#L1629 14:28:13 according to wxy's comment in the bug, it only shows up when we start chunking the object 14:28:55 #topic cache prefetcher redesing 14:29:00 abhishekk: this was your 14:29:54 jokke_, ? 14:30:12 cache-prefetcher in agenda 14:30:13 item #4 on agenda 14:30:22 ohh, sorry 14:30:30 Just to highlight 14:30:57 cache-prefetcher is using registry which is deprecated and is in line to remove this cycle 14:31:01 I moved it up from the Open Discussion 14:31:06 yes 14:31:25 so we need to change the entire design pf cache-prefetcher to use v2 api instead of registry 14:31:45 GR8 ... do we need actual api extensions for this? 14:32:12 Earlier I have submitted a spec-lite for this but I guess it will require full spec now 14:32:17 I guess yes 14:32:22 bummer 14:32:25 abhishekk: ok do you have tomorrow free? 14:32:27 how bad a change is it? 14:32:36 Its a mess 14:32:42 jokke_, I have 14:32:43 :( 14:33:01 abhishekk: ok, lets get on bluejeans tomorrow and have a look on this 14:33:13 Fixing cache-manage was a easy task, but cache-prefetcher will be tough one 14:33:23 jokke_, ping me once you are online 14:33:45 abhishekk: will do ... I'll try to make it as early as possible to not burn your fri night 14:34:08 moving on 14:34:15 thank you ;) 14:34:37 #topic Feature branch for edge image handling (namely metadata caching) 14:34:56 :-) 14:34:56 So GregWaines and his guys have been working on this metadata caching spec 14:35:19 Is this possible to move forward on at least a POC wrt the Glance Edge work ? 14:35:22 and their code is based on old glance which makes all kind of life around this difficult 14:35:25 in the stein timeframe 14:35:59 yeah we would .... minimally port our existing code to stein 14:36:14 So I was thinking of kicking up feature branch for it so that the code could be brought up to master while we keep working on the spec 14:36:29 I like that idea 14:36:33 +1 14:36:36 I wanted to bring this up to confirm that everyone would be on board with this 14:37:16 i agree, it's a good idea, it will allow work to continue 14:37:48 I'm not expecting this to be released with Stein, but I do want to see the work moving on and us having common place to experiment and try this out 14:38:06 sounds good to me 14:38:12 checking out and carrying gerrit reviews for long period of time is not really sustainable 14:38:22 yeah understand that this will miss Stein wrt official glance branch 14:38:50 but gets our forked code closer to real glance and somewhere where you guys can look at it more easily 14:39:00 exactly 14:39:01 ++ 14:39:03 exactly 14:39:24 ++ 14:39:43 ++ 14:39:52 So what I'm expecting from our cores is, please lets not nitpick on the commits to the feature branch, lets do the experientation and reverse if we need to 14:40:04 GregWaines, I will be happy to contribute as well 14:40:09 ack 14:40:26 excellent 14:40:32 and once we get the branch to the point we can agree it being sustainable we can then work on the merge between the branches 14:41:53 I do not want us wasting huge amount of time to try to finetune the code up to the highest standards while we are still working to even agree on the approach in the spec 14:42:02 but this allows the work to continue 14:42:11 Agreed 14:42:45 +1 14:43:35 GregWaines: this might mean that we need to spend quite some time to work out patch from the feature branch we can submit to master if we agree on the approach but I think it would be worth of the effort to keep this moving 14:43:58 I am ok with that 14:44:08 as I expect the git history being a mess we don't want to just merge into master at the end :P 14:44:48 cool 14:45:32 #startvote Everyone ok with the very experimental feature branch for metadata caching? yes, no 14:45:34 Begin voting on: Everyone ok with the very experimental feature branch for metadata caching? Valid vote options are yes, no. 14:45:35 Vote using '#vote OPTION'. Only your last vote counts. 14:45:39 #vote yes 14:45:47 #vote yes 14:45:51 #vote yes 14:45:57 #vote yes 14:46:01 #vote yes 14:46:04 #vote yes 14:46:14 I am not a core :) 14:46:18 I think that's all of us 14:46:25 LiangFang: this wasn't just for cores 14:46:44 Ok Ok 14:46:48 I really want us all aboard as a team 14:46:54 #endvote 14:46:55 Voted on "Everyone ok with the very experimental feature branch for metadata caching?" Results are 14:46:56 yes (6): LiangFang, smcginnis, GregWaines, abhishekk, jokke_, rosmaita 14:47:17 thanks all ... smcginnis I'll put the release patch for that up ;) 14:47:29 #topic Open Discussion 14:47:43 i was just looking at https://review.openstack.org/#/c/628430/ ... maybe you should tell ken'ichi that this restores visibility to its pre-ocata default, and since most people are on newton, they won't notice a difference when the upgrade to stein if we get back to viz==private on image creation 14:47:46 Anything else. We have generous 12 min ... 14:47:59 hi, so since we've reached the open discussion, there's something that I'd like to bring up (sorry for not adding a topic to the meeting agenda). 14:48:04 we'd like to run Glance on Windows. We have customers that are using Hyper-V and HA SMB shares on top of Storage Spaces Direct clusters. 14:48:10 they'd like to leverage those storage backends for Glance images as well, but the linux cifs client doesn't provide proper support for HA SMB shares at the moment. 14:48:14 I have a couple of small patches that will allow us to do so: https://review.openstack.org/#/c/630705/ https://review.openstack.org/#/c/630706/ 14:48:16 rosmaita: thanks for that input 14:48:18 we already have CIs voting on Nova and Cinder that test our Nova Hyper-V driver in conjunction with the Cinder SMB driver. We could easily test Glance on Windows along with it, also testing each patch that lands on Glance. 14:48:36 I have some patches up for reviews, please have a look at those when all of you have some free time 14:48:48 free time ... what's that? 14:48:56 :D 14:49:01 lpetrut: having CI on that would be amazing! 14:49:11 lpetrut: Doesn't look like the changes for that are too extensive. That's great. 14:49:55 just wanted to know what's your input on this so that we may move forward. If that's acceptable for you, I can file a blueprint and start working on a CI 14:50:14 the CI would be great, that addresses a lot of my reservations about this 14:50:17 lpetrut: I did notice your message in os-glance channel and the pathes but as they are flagged as WIP and we had some chaos here around the switf driver and good amount of discussion going on around the metadata caching I have kind of skipped those so far 14:50:50 lpetrut: Was the "WIP" just because you were waiting on the OK to move ahead with these? Or is there more work to be done? 14:51:11 sure, no worries. indeed, I wanted to have some input before removing the "WIP" tag :) that, and I need to add some tests 14:51:24 lpetrut, I doubt those patch are tested as well 14:51:47 https://review.openstack.org/#/c/630706/4/glance/common/wsgi.py@362 14:52:03 lpetrut: cool ... have you actually ran our tests on Win or are these just addressing some specific issues you stumbled upon? 14:52:12 Here CONF.pipe_handle is used while option name defined as 'pipe-handle' 14:52:40 abhishekk: that's intended 14:52:42 Eagle eyes! :) 14:53:03 :D 14:53:11 jokke: did some minimal testing for now (uploading some images, spinning up some instances). I'll run the full tempest suite ASAP 14:53:34 at this point, the only thing that I'm a bit concerned is the glance task feature 14:53:43 aren't we all 14:53:48 lpetrut, ohh, eager to know how it understands the difference 14:53:53 lpetrut: cool ... run the unit and the functional tests from the repo as well. I'm very interested to hear how it behaves 14:54:29 the functional tests may need some changes in order to run on Windows but I think that's manageable 14:54:40 IIRC last person who tried this deemed it way too much of work to keep going, so I'd be very happy to hear if we can claim it actually working 14:55:18 for now, we only support running under eventlet. I was aiming IIS as well, but had some issues because of chunked encoding and the fact that it was buffering the whole image 14:55:21 and if we get CI to back that up, I have no problems to support such efforts 14:55:38 awesome, thanks a lot for feedback. I'll keep you posted. 14:55:48 lpetrut: we really only support running under eventlet for multiple reasons 14:56:32 one being the taskflow totally breaking or the full bufferinf of the images under fully fledged web servers 14:56:58 so don't spend too much time banging your head against the wall to get it working under iss 14:56:59 jokke_: I was actually wondering if that's expected to change in the near future. Some people want to drop eventlet. 14:57:20 lpetrut: yeah they do, we have no intention to spend time on that 14:57:30 got it 14:57:35 as it seems to be massive change for us 14:57:42 2min 14:57:52 would this require a spec? 14:57:52 with minimum manpower atm 14:58:37 lpetrut: I'll need to look into the patches. At least spec-lite would be amazing just for documenting the work that is going on 14:58:57 we can see what needs to be done and if we want to expand it to full spec 14:59:03 makes sense. we'll provide some proper documentation once we're there. 14:59:27 ok last minute on its way 14:59:31 anything else? 14:59:46 nope, please review \0/ 14:59:51 ok 14:59:52 Thanks all! 14:59:57 #endmeeting