14:00:30 <abhishekk> #startmeeting glance 14:00:31 <openstack> Meeting started Thu Feb 27 14:00:30 2020 UTC and is due to finish in 60 minutes. The chair is abhishekk. Information about MeetBot at http://wiki.debian.org/MeetBot. 14:00:32 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 14:00:32 <abhishekk> #topic roll call 14:00:35 <openstack> The meeting name has been set to 'glance' 14:00:41 <abhishekk> #link https://etherpad.openstack.org/p/glance-team-meeting-agenda 14:00:46 <abhishekk> o/ 14:01:58 <rosmaita> o/ 14:02:07 <nao-shark> o/ 14:02:32 <rosmaita> abhishekk: i haven't had time to look at the glanceclient patches this morning, did my comments make sense? 14:02:32 <abhishekk> lets wait 2-3 minutes for jokke_ 14:02:52 <abhishekk> rosmaita, Yes, I have added inline answer to those 14:03:02 <rosmaita> ok, will take a look 14:03:14 <abhishekk> If you still insist to remove those check then I will do it in a separate patch 14:04:26 <abhishekk> lets start 14:04:38 <abhishekk> #topic release/periodic jobs update 14:04:56 <abhishekk> We need to release glanceclient ASAP 14:05:14 <abhishekk> two important patches needs to be merged 14:05:16 <jokke_> o/ 14:05:22 <abhishekk> #link https://review.opendev.org/699656 14:05:26 <whoami-rajat> Hi 14:05:28 <abhishekk> jokke_, o/ 14:05:37 <abhishekk> #link https://review.opendev.org/709086 14:05:41 <abhishekk> whoami-rajat, o/ 14:06:02 <abhishekk> Kindly review these patches and get it going 14:06:39 <abhishekk> Periodic jobs couple of timeouts due to parser error 14:06:47 <abhishekk> which is our next topic 14:06:59 <abhishekk> #topic the stestr situation 14:07:04 <abhishekk> rosmaita, floor is yours 14:07:18 <rosmaita> yeah, i was looking into this real quickly 14:07:30 <rosmaita> was wondering if we could force the issue by going back to testr 14:07:38 <rosmaita> but, it looks like that is no longer maintained 14:07:50 <rosmaita> plus, the real problem seems to be in subunit 14:08:03 <jokke_> yeah iirc that's why we moved to stestr 14:08:10 <rosmaita> and it looks like stephinfin has a patch up for that 14:08:15 <abhishekk> rosmaita, https://github.com/testing-cabal/subunit/pull/40 this seems to be solved the parser error 14:08:24 <rosmaita> looks like the subunit maintainer is not too keen on that patch 14:08:45 <abhishekk> I have applied that in my local environment, and seems to be working 14:08:53 <rosmaita> so i wonder whether it would help if abhishekk and efried for nova added comments to that pull request 14:08:54 <jokke_> is it still the same issue that we're too heavy with logging and overwhelm subunit? 14:09:11 <abhishekk> jokke_, yes 14:09:33 <rosmaita> i don't know 14:09:46 <abhishekk> stephinfin has reproduce for this, https://review.opendev.org/#/c/700522/1/nova/tests/unit/test_utils.py 14:09:47 <rosmaita> the consensus seemed to be that there's some kind of mishandling of a file object somewhere 14:09:50 * jokke_ can see os_subunit fork coming soon to your neighbourhood 14:10:26 <rosmaita> anyway, i put this on the agenda wondering if there was an alternative testrunner we could use 14:10:36 <abhishekk> I have applied this patch in local environment without subunit pull changes it is failing with parser error 14:10:40 <rosmaita> but it looks like not if we want to have highly parallel tests running 14:11:05 <rosmaita> yeah, so the situation is that the patch works, but the subunit maintainer thinks it is a hack 14:11:10 <abhishekk> and with his pull changes it is working 14:11:57 <jokke_> rosmaita: thanks for looking into this and I truly hope we as community can come together and fix the issues rather than a) forking yet another project or b) start fiddling around and patching yet another test runner. Also like you said the options list is pretty short 14:12:11 <abhishekk> I can comment on that, but will it help? 14:12:39 <rosmaita> yeah, i think maybe abhishekk report on the pull request, and maybe we can circle up with stephen for some strategy 14:12:40 <jokke_> does he has better solution to offer? 14:12:53 <rosmaita> yeah, fix our tests so that it doesn't happen 14:12:57 <rosmaita> :P 14:13:04 <jokke_> oh lol, so no 14:13:11 <jokke_> yeah lets keep the pressure on 14:13:33 <abhishekk> rosmaita, cyril is going to drop a mail to him 14:13:47 <rosmaita> ok, sounds good ... that's all from me 14:14:06 <jokke_> one can always start flame war on reddit :P 14:14:12 <jokke_> That helps every time 14:14:17 <abhishekk> :D 14:15:04 <abhishekk> we had lengthy discussion on this today, lets see how this will progress 14:15:05 <abhishekk> moving ahead 14:15:10 <abhishekk> #topic Delete image from single store 14:15:12 <jokke_> ++ 14:15:32 <abhishekk> jokke_, do you mind if I take this ahead? 14:16:03 <abhishekk> IMO this will be very helpful feature and don't want to waste our efforts 14:16:29 <jokke_> I don't care, haven't had time for it and my request for adding hours to a day has not been approved yet :P 14:18:30 <rosmaita> abhishek has been disconnected 14:18:52 <rosmaita> ok, so jokke_ you are ok with abhishekk taking over the patch? 14:18:54 <abhishekk> sorry, I have been disconnected from the network 14:18:54 <abhishekk> did I miss something? 14:18:58 <rosmaita> nope 14:19:07 <jokke_> I don't care, haven't had time for it and my request for adding hours to a day has not been approved yet :P 14:19:12 <jokke_> ^^ that possibly 14:19:12 <abhishekk> cool 14:19:49 <abhishekk> ok, moving ahead 14:20:03 <abhishekk> #topic glance cinder store nfs mount issue 14:20:10 <abhishekk> whoami-rajat, stage is yours 14:20:14 <whoami-rajat> Hi 14:20:37 <whoami-rajat> So i wanted to discuss regarding my patch that mounts the image-volume in the glance directory rather than the cinder one 14:21:09 <whoami-rajat> i've faced error regarding that in my environment (Permission Denied) and i think it can cause much bigger issues too 14:21:28 <whoami-rajat> Nova has the same approach of mounting nfs volumes to their own custom path 14:21:36 <whoami-rajat> and similarly i would like to propose for glance 14:22:08 <rosmaita> that seems to make sense 14:22:37 <jokke_> whoami-rajat: I hope my stance did not come too hars. Happy that you're looking into the problems there. This is just really something os_brick should handle rather than each consumer reinventing the wheel on their own, and that's why os_brick was pushed to us to use in that driver 14:23:22 <whoami-rajat> jokke_, not at all harsh :) 14:23:30 <rosmaita> well, i think that glance_store is supposed to be providing the abstraction layer to glance 14:23:32 <eharney> i'd like to understand the objection better... i think there's some misunderstanding about what os-brick does and doesn't do 14:23:45 <rosmaita> so the place to do the os-brick mediation would seem to be in the glance_store driver 14:25:20 <rosmaita> it has suddenly gone quiet in here? or am i disconnected? 14:25:37 <jokke_> nope, just quiet 14:25:39 <abhishekk> nope 14:25:46 <rosmaita> ok, thanks 14:26:04 <rosmaita> eharney: can you explain why nova takes the approach it does with os-brick? 14:26:43 <eharney> os-brick provides an interface to connect to things, but the consumer (glance_store here, or nova) still has to have code that handles block devices (iscsi/fc), nfs volumes, rbd volumes, etc 14:27:15 <jokke_> So iirc the whole idea of cinder+os-brick was that we get block device from cinder and os-brick provides us access to it so that we do not need to take care of the special sauce of different back-ends in glance 14:27:29 <eharney> os-brick will not provide a block device when cinder is serving volumes over nfs 14:27:51 <jokke_> and yes, we already have special sauce for rbd, but lets be honest they are just special in every way. 14:28:32 <eharney> i assume that the glance_store driver for cinder needs similar work for rbd, but that's a whole different project 14:28:51 <abhishekk> I just have one question, if cinder has configured multiple nfs backends then how this new config option will help? 14:29:26 <eharney> nfs exports are mounted to different directories under $cinder_nfs_mount_point_base 14:29:50 <whoami-rajat> jokke_, the code we're trying to implement isn't for each and every driver but the types of driver i.e. iscsi, fc, nfs, rbd so we just need to have code handling these scenarios rather than all the storage drivers 14:31:02 <whoami-rajat> eharney++ 14:31:36 <whoami-rajat> the mount point is a directory with the volume id (IIRC) inside the mount_point_base 14:31:52 <eharney> i don't think i answered the why nova does what it does question... some of that predates the existence of cinder 14:32:18 <whoami-rajat> /var/lib/glance/<vol-id>/ 14:32:23 <eharney> but the "how" of what it does, is that it determines where it wants to mount nfs exports to consume cinder volumes, and chooses a path to mount them to. this particular item is one of things missing from glance_store currently 14:32:47 <jokke_> whoami-rajat: well, that's the thing ... when the cinder driver was introduced the selling point was that we don't need any of that as cinder abstracts the backend from us, now you're saying that we need special sauce for every type of connectivity as os-brick can't handle that and I guess next step is that we need spaghetti for HP, NetAPP, DellEMC, PureStore etc. cause they will need their 14:32:53 <jokke_> own special treatment 14:32:58 <eharney> no, it's not a per-driver issue 14:33:00 <eharney> it's a per-protocol issue 14:33:05 <abhishekk> current glance_store cinder driver has many loop holes 14:33:22 <eharney> so, iscsi, fc (which are mostly the same in brick), nfs, ceph, and... other rare ones if someone wants to 14:33:49 <jokke_> so can we give os-brick "base mount path" and it does the right thing there or do we need to bring special setting for each protocol? 14:34:23 <eharney> base_mount_path is only relevant for FS protocols (and at this point only NFS is really of interest there, i think) 14:35:06 <jokke_> cause _that_ is the problem I have with this proposal. If os-brick needs a folder it has permission to do it's thing I'm fine with that. If we need to specify that for each and every thing separately I'm having problem with it 14:35:54 <abhishekk> +! 14:35:57 <abhishekk> +1 14:35:57 <rosmaita> well, it's either fix it inside the glance_store cinder driver, or have a glance_store cinder_nfs, cinder_iscis etc 14:36:41 <jokke_> eharney: so can we call it "cinder_base_mount_path" and never have this discussion again when there is next special sayce that needs some node local FS trickstery, been that loop mounts, nfs, iscsi fs with volume images etc. 14:36:45 <jokke_> ? 14:37:27 <eharney> i guess it can be called that since it's in the glance_store cinder driver, but both nova and cinder have the option named "nfs_mount_point_base" 14:37:34 <jokke_> cause I know that day will come if we call that nfs specific that next guy doing their thing doesn't want to use it as it's nfs 14:38:04 <rosmaita> i think it's better to be consistent with cinder and nova 14:38:13 <rosmaita> since people from those teams may wind up working on it 14:38:33 <eharney> the problem with making it generic is that later when you add support for, i dunno, yzfs, then you need a different option named something else for yzfs mounts 14:38:36 <whoami-rajat> jokke_, i think the glance cinder store has a generic code that requires some restructuring for some specific cases. 14:38:48 <eharney> so i'm not sure it's a good idea to avoid putting "nfs" in the name 14:39:33 <abhishekk> rosmaita, as per our glance store standard we define each option with store_prefix_option_name 14:39:35 <whoami-rajat> eharney, nova has different mount_point_base names for different drivers like quobyte 14:39:41 <eharney> whoami-rajat: right 14:39:43 <jokke_> eharney: that's exactly what I wan to avoid 14:39:51 <rosmaita> ok, so i guess if we're arguing over the name of the config opt, then the issue is basically settled? 14:40:08 <eharney> i'm not sure it's a good idea to try to avoid that, i think that's a requirement 14:40:19 <eharney> but, what rosmaita said 14:40:38 <jokke_> eharney: so if brick is given mount path it can then do /put/mount/craps/here/[nfs,yfs,psnpfs] and the driver side should never need to know about it 14:41:00 <eharney> true, we could just make it "cinder_mount_point_base" and all nfs mounts go under $cinder_mount_point_base/nfs/<asdf> 14:41:11 <jokke_> it's just gives brick a path that it has permissions to do what it needs to do 14:41:11 <eharney> and yzfs goes under $cinder_mount_point_base/yzfs etc 14:41:13 <eharney> not a bad idea 14:42:22 <rosmaita> whoami-rajat: that sound OK to you? 14:42:29 <jokke_> cause there is no reason the consumer should be touching those or know about them, I'm assuming brick gives the consumer just fd anyways 14:42:35 <abhishekk> sounds reasonable to me 14:42:54 <whoami-rajat> rosmaita, i'm not sure there are 2 discussions going on 14:43:09 <abhishekk> :D 14:43:19 <whoami-rajat> rosmaita, if you're talking about making cinder_mount_point_base the generic and all FS driver files go into it then i'm ok with it 14:43:21 <jokke_> and that would avoid us having extra 300 lines of config options and comments for them in our already mile long config files just because 14:43:35 <rosmaita> whoami-rajat: ok, and what issue is still open? 14:43:41 <abhishekk> +1 14:44:04 <whoami-rajat> rosmaita, i think jokke_ is still suggesting it inside os-brick ? 14:44:27 <rosmaita> no, i thought it was going to be done in the cinder driver 14:44:58 <whoami-rajat> rosmaita, then i misunderstood the words and i'm clearly ok with everything 14:45:04 <abhishekk> 15 minutes remaining 14:45:30 <jokke_> the best case scenario is that it's in os-brick, we can pass it when we initiate the connector (regardless if it's needed or not) and brick does the right thing 14:46:16 <jokke_> it just knows "Hey I have a path here if I need one I have permissions to" 14:46:45 <jokke_> that would be nice way to consume it :P 14:47:46 <whoami-rajat> can we also have the final action items so i don't miss out anything? 14:48:18 <whoami-rajat> i know one for sure to rename the mount base path config option 14:48:43 <abhishekk> cinder_mount_point_base we can have this option in glance_store cinder driver 14:49:01 <jokke_> whoami-rajat: that's a good start, I think we need to continue this discussion offline. I need to run in a minute 14:49:28 <abhishekk> and all nfs mount goes under cinder_mount_point_base/nfs and others under cinder_mount_point_base/ 14:49:29 <whoami-rajat> jokke_, ack, thanks for the discussion 14:49:34 <jokke_> can't be late twice a day and I joined couple of min late so need to take it back and run couple early :P 14:49:48 <abhishekk> :D 14:49:55 <abhishekk> moving to open discussion 14:50:03 <abhishekk> #topic Open discussion 14:50:13 <rosmaita> i have something 14:50:22 <rosmaita> #link http://lists.openstack.org/pipermail/openstack-discuss/2020-February/012883.html 14:50:44 <rosmaita> question about show_multiple_locations deprecation status 14:51:08 <rosmaita> i replied, but the author suggested revising the train release notes to say the workaround is still needed 14:51:25 <rosmaita> abhishekk: that's up to you 14:51:42 <abhishekk> I need to do one stable/train release coming week 14:51:50 <jokke_> I haven't read that but I think we need to lift the deprecation and hopes of getting rid of it due to how popular ceph became and everything is heavily relying on hetting those location uris 14:52:12 <jokke_> getting 14:52:22 <abhishekk> rosmaita, If you provide me the wording I will add one for sure 14:52:42 <abhishekk> I am pretty bad in writing release notes 14:52:50 <rosmaita> abhishekk: sure, i think we can just copy the one from stein 14:52:58 <rosmaita> i will look and see 14:53:04 <abhishekk> rosmaita, ack, 14:53:06 <jokke_> Ping me if anything is needed. Now I really need to run (today is last day when my passport is valid and I have appointment to order a new one so can't be late from that) :) 14:53:20 <jokke_> will be back online in a while 14:53:26 <abhishekk> jokke_, o/~ 14:53:28 <rosmaita> jokke_: ok, don't get deported 14:53:47 <rosmaita> abhishekk: i'll get up a patch for you in a bit 14:53:49 <abhishekk> we have new member waiting I guss 14:53:56 <abhishekk> rosmaita, thanks 14:54:09 <abhishekk> nao-shark, around? 14:54:26 <nao-shark> Thanks. 14:54:33 <nao-shark> I have two questions . 14:54:48 <nao-shark> https://review.opendev.org/#/c/687390/ 14:54:50 <nao-shark> This is my spec for revive the S3 driver. 14:55:00 <nao-shark> and this is my patch of S3 driver. https://review.opendev.org/#/c/695844/ 14:55:08 <nao-shark> I want to edit glance-api.conf to show how to configure the S3 driver. 14:55:16 <nao-shark> But it looks like generated by oslo-config-generator. 14:55:26 <nao-shark> So my first question is Can I make a separate patch for glance-api.conf before my S3 driver patch is merged? 14:55:35 <nao-shark> My concern is that modifications to the S3 driver may directly affect the content of glance-api.conf. 14:56:05 <abhishekk> nao-shark, I will have a look at those and will get back to you 14:56:14 <rosmaita> nao-shark: what you can do is regenerate the config file and submit it as part of your patch 14:56:32 <rosmaita> i think we still keep a full config sample? or did we stop doing that? 14:56:40 <abhishekk> nao-shark, could you ping me tomorrow in the morning time? 14:56:50 <abhishekk> rosmaita, I guess we have stopped doing that 14:57:00 <rosmaita> oh, ok 14:57:19 <rosmaita> nao-shark: best work it out with abhishekk tomorrow 14:57:28 <nao-shark> abhishekk OK. thanks 14:57:47 <abhishekk> cool 14:57:52 <abhishekk> anything else 14:58:08 <nao-shark> Maybe the next question should probably be tomorrow 14:58:28 <abhishekk> rosmaita, I have replied to your question on client patch 14:58:42 <abhishekk> nao-shark, yes 14:58:53 <abhishekk> or you can drop me a mail as well 14:59:25 <abhishekk> thank you all 14:59:42 <nao-shark> abhishekk OK Thanks for your support ! 14:59:50 <rosmaita> bye 14:59:53 <abhishekk> nao-shark, no worries 14:59:55 <abhishekk> #endmeeting