20:05:28 <markwash> #startmeeting glance
20:05:28 <openstack> Meeting started Thu Mar  6 20:05:28 2014 UTC and is due to finish in 60 minutes.  The chair is markwash. Information about MeetBot at http://wiki.debian.org/MeetBot.
20:05:29 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
20:05:31 <openstack> The meeting name has been set to 'glance'
20:05:52 <arnaud__> o/
20:05:55 <nikhil___> o/
20:06:04 <ameade> o/
20:06:39 <rosmaita> o/
20:07:01 <markwash> o/
20:07:12 <iccha_> o/
20:07:14 <markwash> so, we are in feature freeze
20:07:25 <markwash> there are some blueprints and bugs targeted to rc1
20:07:34 <markwash> #link https://launchpad.net/glance/+milestone/icehouse-rc1
20:08:49 <markwash> the idea is to get these in before tuesday so that I don't have to beg or explain the case for more time :-)
20:08:55 <markwash> and so we can cut rc1 sooner
20:09:01 <iccha_> will do mark
20:09:28 <markwash> also, landing fixes to important bugs is high priority
20:09:49 <markwash> the sooner we can do that, the sooner we can cut rc1, and the sooner we can get it in some more people's hands
20:10:33 <ameade> people should feel free to slam me with review requests, i may be able to pick up a bug here and there
20:10:49 <markwash> I don't have any more immediate items for today's meeting, should we check in on the status of the artifacts work?
20:11:00 <arnaud__> that would be nice
20:11:12 <ameade> markwash: yeah whats the word?
20:12:24 <ameade> are there any blueprints?
20:12:30 <markwash> seems we're still in design phases. . but I don't see alex or jbernard around to discuss with
20:12:35 <arnaud__> at the end of the meeting I would like to discuss a behavior that I am seeing: I want to confirm that this is not a bug
20:12:42 <markwash> ameade: there are blueprints but they're still being defined IIRC
20:13:44 <rosmaita> https://blueprints.launchpad.net/glance/+spec/artifact-repository-api
20:13:58 <rosmaita> also, https://wiki.openstack.org/wiki/Glance-artifacts-api-faq
20:14:14 <ameade> rosmaita: ty sir
20:14:22 <rosmaita> np
20:14:48 <arnaud__> Mark do we have randall working on the heat side?
20:14:48 <markwash> seems there's not much new information however
20:15:07 <markwash> arnaud__: I haven't heard from him in a bit, but I think he's still interested
20:15:19 <arnaud__> okay
20:16:01 <markwash> arnaud__: you and I had some good conversation about directions the artifacts structures could go
20:16:18 <markwash> iirc, we had hoped to share that info with alex, have you had a chance to follow up with him?
20:16:34 <arnaud__> Iwas busy on some other stuff this week
20:16:41 <markwash> its been one of those weeks for me as well
20:16:45 <arnaud__> I have our conversation
20:16:55 <arnaud__> I will talk to him before next meeting
20:17:13 <markwash> all right, looks like we don't have much more in terms of status updates
20:17:23 <ameade> markwash: do you know if anyone is working on the cinder driver?
20:17:33 <markwash> ameade: not that I'm aware of
20:17:45 <ameade> it's basically just a stub atm
20:17:56 <arnaud__> I know zhiyan was interested in it
20:17:58 <zhiyan> ameade: cinder store driver? what do you think?
20:18:23 <ameade> zhiyan: i'd like it to get finished :)
20:18:26 <zhiyan> ameade: IMO the issue for it is Brick
20:19:09 <markwash> there are a few other issues as well that have come up
20:19:20 <ameade> zhiyan: can you summarize brick and why it's a blocker? i know brick is part of cinder
20:19:28 <markwash> 1 is that we don't have support for multi-format images, so its a bit hard to use cinder drivers for anything other than raw images
20:19:34 <zhiyan> ameade: without it we/Glance (and other project who outside Cinder) can't access volumes
20:19:47 <ameade> ah i see
20:20:06 <markwash> 2) is that its not clear that the deployment pattern for glance and cinder would allow glance to own the cinder volumes, so they could get deleted, leaving the location a dangling reference
20:20:18 <ameade> that makes sense too
20:20:26 <ameade> i really dont want dead code sitting there then
20:20:27 <zhiyan> right, currently it is *part of* cinder, but a standard lib
20:21:01 <ameade> what i really wanna see is the glance.store code turn into a transfer service
20:21:09 <nikhil___> +1
20:21:15 <ameade> but it gets weird i think when you mix block and object storage
20:21:44 <markwash> there was a really interesting idea put forward by lifeless at the triple O summit that could help us kick off the transfer service
20:21:48 <zhiyan> ameade: is is not a dead code, via image_handler within nova, we can consume that volume acted image as we needed (since nova side has volume accessing code)
20:22:19 <lifeless> \o/
20:22:27 <markwash> I captured it in a blueprint (which links to lifeless ideas in an etherpad) https://blueprints.launchpad.net/glance/+spec/multicast-image-transfer
20:24:03 <ameade> hmm ok, i'll try to digest this stuff and form some opinions
20:24:26 <markwash> the whole idea of the transfer service isn't entirely clear in my mind
20:24:39 <nikhil___> that seems to overlap a bit with the bittorrent protocol within the xfer service
20:24:43 <markwash> so a usecase like that was interesting in that it seemed like it *could* help clarify some things
20:24:54 <markwash> nikhil___: yeah, I think there is a lot of similarity
20:24:59 <nikhil___> we have a few use cases markwash
20:25:03 <zhiyan> imo multicast still is a thing for "copy", my idea is zero-copy
20:25:05 <markwash> I kind of think they would be side-by-side protocols
20:25:15 <nikhil___> would like to provide feedback
20:25:18 <nikhil___> markwash: yeah
20:25:34 <nikhil___> zhiyan: ?
20:25:47 <markwash> zhiyan: you're right, but I think there are requirements for good copy support as well
20:26:12 <markwash> anyway, I'd like to avoid ratholing too far down at this point :-)
20:26:13 <ameade> yeah i think both are important though
20:26:43 <markwash> so, arnaud__ you had a quick bug you wanted some discussion about?
20:26:45 <zhiyan> markwash: ok, but tbh prevent copy is my finial target.
20:27:18 <ameade> zhiyan: you mean related to multi-attach volumes?
20:27:18 <arnaud__> markwash: yes
20:27:29 <ameade> zhiyan: are you the one working on that in cinder?
20:27:45 <zhiyan> ameade: they are different topic, iiuc.
20:27:50 <arnaud__> +1
20:27:51 <zhiyan> yes, i am
20:28:21 <ameade> kk, sorry we can move on now
20:28:43 <arnaud__> ok :) so here is the story
20:30:39 <arnaud__> in nova, I create an image - upload content of size 0 (StringIO('')) - then get the location for the image - then replace the bits at this location with something much bigger than 0 - then update the image with the new size of this big thing
20:30:43 <arnaud__> this works
20:30:55 <arnaud__> after that I am able to boot the new image from nova
20:31:15 <arnaud__> I feel like something is messed up :) (i.e. the checksum)
20:31:24 <arnaud__> is it clear?
20:31:32 <nikhil___> hmm
20:32:11 <ameade> arnaud__: ah yeah, like nova isn't checking the checksum?
20:32:29 <zhiyan> arnaud__: did you check the checksum value after you update the image?
20:32:35 <gokrokve_> Hi. Sorry I am late.
20:32:58 <nikhil___> may be we need to add verificaion of the etag (checksum returned from store) with the one store in glance?
20:33:09 <markwash> arnaud__: so you just update the size directly in glance?
20:33:13 <arnaud__> yes
20:33:23 <arnaud__> markwash: yes, I just update the size
20:33:28 <arnaud__> and modify what is at the location
20:33:47 <arnaud__> this is cool for me, but that make works what I want to do
20:34:02 <arnaud__> but I want to make this isn't working because of a bug
20:34:03 <arnaud__> :)
20:34:19 <arnaud__> zhiyan: the checksum of the image is populated after update
20:34:40 <zhiyan> ok, if so seems like it's a bug in nova but glance
20:35:20 <arnaud__> so this is OK behavior to have Nova replacing a location behind Glance's back? and update the size?
20:35:30 <markwash> it does seem like there is a bug in glance somewhere if you can update the size or the checksum, or have the download work with an invalid checksum
20:35:51 <arnaud__> ok
20:36:04 <arnaud__> ok, cool.
20:36:10 <zhiyan> arnaud__: humm, i think this use case is too hack to glance
20:36:18 <rosmaita> markwash: are you saying that checksum, size, and location should be linked so you must update all at the same time?
20:36:19 <markwash> this is sort of hilarious, asking you to file a bug for a functionality you really want to use :-)
20:36:40 <arnaud__> zhiyan: I agree with you, but at this point I need to get the Glance location and size the location is provided after content is uploaded
20:37:28 <zhiyan> arnaud__: why you need replace image bits via location?
20:37:44 <arnaud__> markwash: I don't want to file a bug, but I want to be sure that in the future no one else file a bug and change the behavior
20:38:07 <arnaud__> to move the bits of the snapshot instead of uploading them
20:38:15 <nikhil___> zhiyan: upload behind the scenes
20:38:20 <arnaud__> yep nikhil___
20:38:28 <nikhil___> ;)
20:39:17 <markwash> the core problem here is that clients don't know where to put data if they don't want to send it through glance
20:39:23 <arnaud__> yes
20:39:42 <markwash> arnaud__: I think its pretty likely that we will file a bug and "fix" the behavior you're seeing :-)
20:40:13 <arnaud__> ok...
20:40:14 <arnaud__> :)
20:40:55 <nikhil___> feel even more need for a xfer service which is kinda glance aware
20:40:56 <arnaud__> I would like to see a location provided before having to upload the content
20:41:16 <arnaud__> +1 nikhil___
20:41:21 <markwash> there was a suggestion at some point for glance to provide the "upload location" that clients could use
20:41:34 <zhiyan> arnaud__: iiuc, currently we support it
20:41:46 <arnaud__> zhiyan: could you develop?
20:41:57 <arnaud__> I am not seeing the direct_url just by doing image_create afaik
20:42:06 <ameade> isn't this what breaking out glance.store is for?
20:42:16 <arnaud__> not exactly
20:42:17 <zhiyan> arnaud__: create a image, it under "queued" status, and you an prepare the image bits on nova side, then use PATCH api to add that location to glance
20:42:26 <zhiyan> arnaud__: then the image will turn to "active"
20:42:36 <markwash> ameade: sort of, but its difficult to imagine all of the clients getting the right credentials for talking directly to the store
20:42:43 <arnaud__> zhiyan: the thing is that I need to have this location
20:42:44 <ameade> heh yeah
20:42:47 <arnaud__> before
20:42:54 <ameade> so this is what a third party transfer service would do
20:43:12 <arnaud__> you don't know the format that is used by glance to store the image
20:43:25 <arnaud__> the file name, the folder or whatever info contained in the URL of the image
20:44:07 <arnaud__> do you get my point zhiyan?
20:44:14 <markwash> so, can someone file a bug documenting this behavior?
20:44:24 <arnaud__> I will markwash
20:44:26 <zhiyan> arnaud__: yes, but i'm thinking if this use case on nova side is make sense
20:44:32 <markwash> okay, cool
20:44:46 <markwash> if there aren't any other items atm, I have something I need to run off to
20:45:56 <arnaud__> thank you guys
20:46:16 <ameade> bye everyone
20:46:22 <markwash> thanks!
20:46:25 <markwash> #endmeeting