| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-sdks - 2020-11-18 | |||
| 10:29:48 | gtema | I think OSC currently doesn't support those | |
| 10:30:26 | stephenfin | The SDK is supposed to be complete, but that API (os-hosts) is deprecated and the things we said to use instead (os-hypervisors, os-services) are present in 2.1 iirc | |
| 10:30:32 | gtema | investing time in deprecated/dropped functionality is hard | |
| 10:30:49 | stephenfin | So I think it's exceptionally low priority, IMO. I wouldn't bother, personally | |
| 10:31:05 | gtema | pretty much same for me | |
| 10:31:22 | gtema | my plan was to get basic host ops covered in SDK to be able to use them from OSC | |
| 10:32:26 | stephenfin | we already have e.g. 'hypervisors list' | |
| 10:32:39 | stephenfin | so I think you're covered there? | |
| 10:32:48 | stephenfin | that'll give you host information | |
| 10:33:04 | gtema | that's true, but if you want to fill 'openstack aggregate add host' with host check you need that | |
| 10:33:19 | gtema | or you mean hypervisors are really giving same info back? | |
| 10:33:42 | stephenfin | the latter; it's the same information exposed by both afaik | |
| 10:33:55 | gtema | hm, interesting | |
| 10:35:12 | stephenfin | To be clear though, that idea was also a nice-to-have UX improvement. Converting more of the other commands to SDK like you've been doing is maybe more helpful? | |
| 10:35:27 | stephenfin | I should start prefixing those comments with 'idea:' or something to distinguish them | |
| 10:36:11 | gtema | perfectly fine with that. I would be perhaps better if I work on SDK switch changes (SDK and OSC) and you improve UX | |
| 10:36:24 | stephenfin | wfm :) | |
| 10:36:48 | gtema | hypervisor vs host are 2 totally different set of APIs (facepalm) | |
| 10:37:01 | gtema | making smooth transition is not easy | |
| 10:37:33 | gtema | especially getting host details is so bad, while getting hypervisor details is ok | |
| 10:39:03 | gtema | you will participate tomorrow in a meeting, right? | |
| 10:39:11 | stephenfin | gtema: We could just deprecate the 'host *' commands | |
| 10:39:19 | gtema | I wanted to agree on the "order" of next switches | |
| 10:39:20 | stephenfin | rather than providing a shim | |
| 10:39:54 | gtema | interesting idea. This would however mean we can't drop novaclient from deps in a long run | |
| 10:40:00 | stephenfin | it's not like we'll break older clouds; those have os-hypervisors and os-services too | |
| 10:40:02 | gtema | we did that with glance | |
| 10:40:20 | stephenfin | we could if deprecate them for removal | |
| 10:40:31 | stephenfin | i.e. we'll drop novaclient in the next major version bump | |
| 10:40:47 | stephenfin | fwiw, we have the same issue with commands like 'server add floating ip' | |
| 10:41:18 | gtema | I was personally thinking to do a major bump with nova switch, so we can take it with us | |
| 10:41:40 | stephenfin | those use deprecated "orchestration" APIs; I've seen patches to replace some of those with client side equivalents but I'm not sure if that's better than asking people to do this stuff manually | |
| 10:41:56 | stephenfin | as for the meeting, that's a 14:00 UTC, yeah? I can be there | |
| 10:42:13 | gtema | so you suggest not to deal with host (SDK/OSC) as of now at all? | |
| 10:42:44 | stephenfin | unless you really want to | |
| 10:42:52 | stephenfin | just deprecate the commands and remove them in a future major version bump | |
| 10:43:06 | gtema | surely not. As I said - my hears were flying around yesterday | |
| 10:43:33 | stephenfin | then let's do that; if someone complains, they can do the SDK work :) | |
| 10:43:39 | gtema | perhaps I will then finish SDK patch as is to get some sort of support | |
| 10:44:54 | gtema | oki, then I will start with hypervisors next | |
| 10:47:04 | openstackgerrit | Artem Goncharov proposed openstack/openstacksdk master: Add support for deprecated nova hosts https://review.opendev.org/763081 | |
| 11:15:07 | openstackgerrit | Ananya proposed openstack/ansible-collections-openstack master: Migrating routers_info from AnsibleModule to OpenStackModule https://review.opendev.org/763162 | |
| 11:28:18 | openstackgerrit | Ananya proposed openstack/ansible-collections-openstack master: Migrating routers_info from AnsibleModule to OpenStackModule https://review.opendev.org/763162 | |
| 11:36:51 | openstackgerrit | Stephen Finucane proposed openstack/python-openstackclient master: Add 'server unshelve --wait' option https://review.opendev.org/763166 | |
| 11:36:51 | openstackgerrit | Stephen Finucane proposed openstack/python-openstackclient master: Add 'server shelve --offload', 'server shelve --wait' options https://review.opendev.org/762396 | |
| 11:53:35 | openstackgerrit | Stephen Finucane proposed openstack/python-openstackclient master: compute: Add 'server show --topology' option https://review.opendev.org/680928 | |
| 12:07:06 | openstackgerrit | Merged openstack/openstacksdk master: Add update_flavor method https://review.opendev.org/762989 | |
| 12:08:03 | openstackgerrit | Merged openstack/openstacksdk master: Complete compute aggregate functions https://review.opendev.org/762131 | |
| 12:08:04 | openstackgerrit | Merged openstack/openstacksdk stable/victoria: Don't set list_type to dict for server groups. https://review.opendev.org/763119 | |
| 13:26:51 | stephenfin | gtema: FYI https://review.opendev.org/#/c/763178/ | |
| 13:27:08 | stephenfin | that should unblock your aggregate and flavor OSC changes | |
| 13:27:41 | gtema | hmm, I wanted to go with hypervisors further, but ok | |
| 13:28:18 | gtema | do not really want to make release each week ;-), but is still required | |
| 13:28:44 | stephenfin | meh, releases are cheap :) | |
| 13:28:56 | openstackgerrit | Merged openstack/python-openstackclient master: Add a few selectable fields to the "openstack server list" output https://review.opendev.org/741181 | |
| 13:28:57 | openstackgerrit | Merged openstack/openstacksdk master: Update TOX_CONSTRAINTS_FILE https://review.opendev.org/761055 | |
| 13:28:59 | openstackgerrit | Merged openstack/python-openstackclient master: update lower-constraints.txt https://review.opendev.org/760270 | |
| 13:29:03 | stephenfin | but we can wait of course, if you'd prefer | |
| 13:29:06 | gtema | yes, they are | |
| 13:29:14 | gtema | already +1-ed | |
| 13:30:54 | stephenfin | sweet | |
| 13:32:09 | gtema | hypervisors API is also not so straight forward | |
| 13:32:16 | gtema | but not that bad as host | |
| 14:16:48 | openstackgerrit | Gregory Thiemonge proposed openstack/openstacksdk master: Add tls_enabled param for Octavia Pools https://review.opendev.org/763189 | |
| 14:17:36 | openstackgerrit | Gregory Thiemonge proposed openstack/openstacksdk master: Add tls_enabled param for Octavia Pools https://review.opendev.org/763189 | |
| 15:26:17 | openstackgerrit | Artem Goncharov proposed openstack/openstacksdk master: Complete compute.hypervisor functions https://review.opendev.org/763202 | |
| 17:42:15 | openstackgerrit | Rodolfo Alonso Hernandez proposed openstack/python-openstackclient master: Add "fields" parameter to ListPort query https://review.opendev.org/754117 | |
| 18:06:07 | openstackgerrit | Stephen Finucane proposed openstack/cliff master: trivial: Remove references to Python 2.7 https://review.opendev.org/760334 | |
| 19:56:57 | openstackgerrit | Merged openstack/cliff master: Remove six https://review.opendev.org/754683 | |
| 21:08:27 | openstackgerrit | Merged openstack/cliff master: trivial: Remove references to Python 2.7 https://review.opendev.org/760334 | |
| 21:42:40 | openstackgerrit | Merged openstack/openstacksdk master: Add tls_enabled param for Octavia Pools https://review.opendev.org/763189 | |
| #openstack-sdks - 2020-11-19 | |||
| 01:31:19 | iokiwi | Hi I am working on a fix for this, https://storyboard.openstack.org/#!/story/2007672. According to the docs, the default behavior for `openstack image save` should go to stdout, but recent changes / current implimentation makes it go to memory. | |
| 01:52:52 | iokiwi | Is it still desireable that output go to stdout? Should I restore this as the default output? | |
| 05:08:44 | openstackgerrit | Merged openstack/python-openstackclient master: Add "fields" parameter to ListPort query https://review.opendev.org/754117 | |
| 06:35:45 | frickler | iokiwi: it seems there are actually two issues here, both related to the switch from glanceclient to sdk: | |
| 06:36:02 | frickler | a) the default output to stdout is broken | |
| 06:37:11 | frickler | b) osc now tries to buffer the complete image in memory before saving it to a file or stdout, leading to an OOM when the image size is larger than available memory | |
| 06:37:34 | frickler | waiting for gtema to be back online for further discussion | |
| 06:38:24 | frickler | for me, while a) certainly is a regression that should be fixed, b) is the more severe issue | |
| 07:11:41 | openstackgerrit | Simon Merrick proposed openstack/python-openstackclient master: stop image downloads to memory https://review.opendev.org/763317 | |
| 07:32:37 | iokiwi | frickler thanks my patch addresses both. Certainly agree that b) is the bigger issue (especially when downloading a 100gb image) | |
| 07:34:51 | openstackgerrit | Artem Goncharov proposed openstack/openstacksdk master: Complete compute.hypervisor functions https://review.opendev.org/763202 | |
| 07:36:38 | frickler | iokiwi: thanks, that looks much simpler than I had expected. will test later today. it might be good to add a releasenote, though. | |
| 07:37:02 | gtema | that is definitely needed | |
| 07:37:15 | iokiwi | fickler sure I'll add one now | |
| 07:37:21 | gtema | things go easy when you use SDK ;-) | |
| 07:38:06 | iokiwi | Based on the docs, I am not sure sdk will do md5 checksum when stream=True | |
| 07:38:13 | iokiwi | Do you think this will be a problem? | |
| 07:38:39 | gtema | well, the checksum validation is anyway doomed, since many clouds do not do this properly | |
| 07:39:01 | gtema | checking | |
| 07:40:55 | iokiwi | I think the sdk just wont/can't try to do md5 sum so to be more specific, is the md5 checksum important enough to us to implement it? Based on your comment above, maybe not. | |
| 07:41:38 | gtema | it is important that this possibility remains in SDK as is, since there are other users (except OSC) depending on it | |
| 07:43:26 | gtema | well, looking to the code I don't see it would change the behaviour really. | |
| 07:43:45 | gtema | it was initially designed to be either output or stream, not both together | |
| 07:48:23 | gtema | so as long as you download image into file I do not think there is effect of this change | |
| 07:48:54 | gtema | (since my cloud disabled image download I can't really verify anything) | |
| 07:50:28 | frickler | gtema: saving a 50g image to a file gives OOM for me with latest, works fine with 5.1.0. setting test up with the patch now | |
| 07:54:43 | frickler | the patch fixes that and also the stdout issue. does get some not so nice output when pipe fails, but that might be fixed in a followup http://paste.openstack.org/show/800191/ | |
| 07:58:08 | openstackgerrit | Simon Merrick proposed openstack/python-openstackclient master: stop image downloads to memory https://review.opendev.org/763317 | |
| 07:59:32 | gtema | that's weird. I can't understand why it should change the behavior | |