| abhishekk_ | dansmith, sorry, I just saw your ping now related to glance_store patch, but I don't think this is related to what we currently witnessing, @pranali please confirm | 08:09 |
|---|---|---|
| abhishekk_ | https://review.opendev.org/c/openstack/glance_store/+/884524 this is the patch dansmith is referenced yesterday | 08:10 |
| pranali | abhishekk_, dansmith no the this glance_store fix is not related to the deletion issue we noticed, rbd throws ImageBusy exception not the ImageHashSnapshot, I have also confirmed that by testing the image download and new loaation api without this glance_store patch as well | 09:27 |
| pranali | dansmith, this ceph issue is not only with new location api but also with image download api which i've reported already https://bugs.launchpad.net/glance/+bug/2045769 | 09:29 |
| dansmith | abhishekk: I guess pranali is already gone, but.. if this is not just related to the location API, surely this must be causing failures in the gate without the location api changes right? | 14:32 |
| dansmith | if so, then fine, but if not, including it in the location api set means we're changing two things at once... | 14:33 |
| abhishekk | what kind of failure you are talking about? | 14:33 |
| dansmith | okay re-thinking what you said yesterday... you think we literally don't have anything that tests the sequence of steps listed at the top of the bug currently right? | 14:35 |
| abhishekk | yes | 14:36 |
| dansmith | okay, and why do you think we're hitting that procedure with no changes other than the location api? | 14:36 |
| abhishekk | I just created large size image, then simultaneously while downloading deleted that image, it failed image was active but it was deleted from ceph | 14:38 |
| abhishekk | created image with image-create/imgae-create-via-import command not using location API | 14:39 |
| dansmith | yeah I understand, like in the bug right? | 14:39 |
| abhishekk | yes | 14:39 |
| dansmith | I'm asking about why you think moving to the location api (usage by nova) means we now hit the problem with the existing tests | 14:40 |
| abhishekk | because earlier we were not calculating hash in old location API | 14:40 |
| abhishekk | what changed with new location API is we set image to active and then download it in background to calculate the hash | 14:41 |
| dansmith | right, and the concern is that now we can't delete that image until the hash is done right? | 14:42 |
| abhishekk | ceph deletes it but imgae state is still active in glance | 14:43 |
| dansmith | or, the problem as nova/tempest see it | 14:43 |
| abhishekk | ceph deletes the image even if IsInUse raised from store side | 14:43 |
| dansmith | oh, is it? that's not what I was expecting, but I also haven't seen anyone lay out *why*, which is what I'm trying to get at | 14:43 |
| abhishekk | yes | 14:43 |
| abhishekk | there is one ceph link in the bug I guess which caused this regression I think | 14:44 |
| dansmith | okay, and that's a problem for the location api stuff why? because glance leaves the image as active? | 14:44 |
| abhishekk | yeah | 14:44 |
| abhishekk | and as an workaround in the patch we changing the state to deleted in the location API patch | 14:45 |
| dansmith | wait, what? | 14:45 |
| dansmith | then why do we need to switch the ceph release in my nova patch? | 14:45 |
| abhishekk | https://review.opendev.org/c/openstack/glance/+/886749/32/glance/async_/flows/location_import.py#91 | 14:45 |
| abhishekk | I was not aware about that, let me check | 14:46 |
| dansmith | okay, *that* is my whole concern here :) | 14:46 |
| abhishekk | I think she left it there by mistake | 14:47 |
| dansmith | that except clause in the import seems reasonable to me | 14:47 |
| abhishekk | I asked her to test it with pacific as well | 14:47 |
| dansmith | abhishekk: okay that's what I've been trying to dig at this whole time :) | 14:47 |
| abhishekk | really sorry | 14:47 |
| abhishekk | I think even removing that it will work | 14:47 |
| dansmith | hence this: [07:45:10] <dansmith> it seems like just moving ceph back to pacific is not a good solution if pacific doesn't match what we're actually going to run on the current openstack/distro/ceph releases | 14:48 |
| dansmith | no, I wasn't clear I guess | 14:48 |
| dansmith | I'll revert that and push it up and let it run | 14:48 |
| abhishekk | I thought you were talking about local testing after reading comment from the bug | 14:48 |
| dansmith | nope | 14:49 |
| dansmith | I thought she was asserting that we'd roll back to pacific until it gets fixed in ceph, which I think is not reasonable | 14:49 |
| abhishekk | also my irc bouncer stopped working lately so was not able to see all messages (so I missed it yesterday) | 14:50 |
| abhishekk | yep, looking at patch anyone has thought the same | 14:50 |
| dansmith | ack | 14:50 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!