| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-sdks - 2022-10-26 | |||
| 14:02:20 | opendevreview | Denys Mishchenko proposed openstack/ansible-collections-openstack master: Existing images update name, visibility etc https://review.opendev.org/c/openstack/ansible-collections-openstack/+/862565 | |
| 14:02:26 | fungi | what would compute:get_all normally return? is that the list of server instances? | |
| 14:02:33 | gtema | yes | |
| 14:02:46 | fungi | so essentially `openstack server list` | |
| 14:02:54 | gtema | correct | |
| 14:03:01 | fungi | thanks | |
| 14:03:06 | opendevreview | Jiri Podivin proposed openstack/cliff master: Removing helper functions providing Python < 3.3 compatibility https://review.opendev.org/c/openstack/cliff/+/862689 | |
| 14:03:11 | gtema | thank you fungi | |
| 14:03:20 | gtema | I learned something new about sdk today | |
| 14:04:26 | fungi | i'll see what more we can get from my contact there, i just wanted to make sure i had a more complete picture of what/where the problem is before reaching out | |
| 14:05:30 | gtema | sure. This rackspaceauth is a thing that we would also need to install into sdk jobs, but it should be not an issue | |
| 14:06:20 | fungi | well, since it's been an advertised option (going back at least 5 years in the published sdk docs), double-checking that it can be made to work would be good anyway | |
| 14:06:38 | fungi | and if it can't, then a docs update is in order | |
| 14:07:19 | gtema | at least now I eventually see where clarkb's problem with cinder may lay - in this type of auth project_id is not being set and we try to concat None | |
| 14:07:28 | fungi | but i'll see if he's willing to give us a password for the account too, so that both paths can be tried | |
| 14:07:45 | gtema | awesome, thanks | |
| 14:09:13 | fungi | gtema: fwiw, with opendev's tenants we don't use api_key, we set password and project_id in our clouds.yaml files | |
| 14:09:54 | gtema | yes, I have seen that, but clark mentioned he was trying in venv so I assumed he may be making few tweaks | |
| 14:10:45 | fungi | nah, the venvs we've been talking about still rely on the same /etc/openstack/clouds.yaml file we install on the server where we're running the client | |
| 14:11:31 | fungi | though we do end up needing to use an old version of the sdk and we pass --os-volume-api=1 on the client command line in order to make it work | |
| 14:11:47 | gtema | ok, then if "cinderclient.exceptions.UnsupportedVersion: Invalid client version '2.0'. Major part should be '3'" is the broken thing it is due to cinderclient having dropped support | |
| 14:12:20 | gtema | same is for v1 - it was dropped in cinderclient | |
| 14:12:48 | fungi | i have python-cinderclient==4.0.1 in both a working and broken venv | |
| 14:13:03 | gtema | we can try to implement here raw API requests (or hope SDK v1 is still working) | |
| 14:13:15 | gtema | but is that the exception you get? | |
| 14:13:27 | fungi | the problem i'm seeing with the non-working (newer sdk) venv is a "Not Found (HTTP 404)" response from the cloud | |
| 14:13:31 | opendevreview | Stephen Finucane proposed openstack/python-openstackclient master: compute: Add '--no-network', '--auto-network' flags https://review.opendev.org/c/openstack/python-openstackclient/+/862680 | |
| 14:14:19 | gtema | for something like volume list? | |
| 14:14:36 | fungi | yes | |
| 14:15:07 | fungi | if i downgrade openstacksdk from 0.48.0 to 0.41.0 then volume list works with the same command-line options and configuration | |
| 14:15:26 | fungi | i haven't finished bisecting between those yet | |
| 14:15:28 | gtema | wow, but that is very very old | |
| 14:15:36 | fungi | yes, we had to go back that far to make it work | |
| 14:15:58 | gtema | anyway, in order to reproduce I need to pass through "Policy doesn't allow volume:get_all to be performed" which I get now | |
| 14:15:59 | fungi | rather, we started pinning when it broke and haven't been able to upgrade | |
| 14:16:45 | fungi | and yes, trying to use cinderclient as a workaround is problematic in the way you noted | |
| 14:16:58 | fungi | (newer cinderclient dropped old api versions) | |
| 14:17:53 | gtema | that sucks quite heavily - supporting such old things is terribly complex | |
| 14:19:00 | fungi | thankfully most of the problem impact is to volume management, which we only do manually. it hasn't adversely impacted stuff like nodepool since it doesn't need to care about the volume api | |
| 14:20:00 | fungi | so we've been able to get by with custom venvs containing really old libs for the times when we need to add/remove/replace volumes for our persistent control plane servers | |
| 14:20:17 | fungi | it's not ideal though, of course | |
| 14:20:29 | gtema | which osc-lib and osc ver are installed? I can't get matching combo | |
| 14:21:56 | fungi | i'm going to make fresh venvs first just to make sure i have a consistent reproducer | |
| 14:22:14 | gtema | I mean the one which is working | |
| 14:22:29 | gtema | with older versions, cause sdk==0.41.0 is not working with "any" OSC | |
| 14:22:36 | gtema | and older OSC falls on osc-lib | |
| 14:23:07 | gtema | and finally falling on "module 'collections' has no attribute 'MutableMapping'" | |
| 14:25:16 | fungi | yeah, that's why i'm trying to reproduce with clean venvs first, because i'd like to know how to create an equivalent of the working reference venv i have from scratch | |
| 14:25:28 | gtema | ok, thks | |
| 14:26:44 | fungi | also note that the venvs in question are built for python 3.6 because they're on an ubuntu 18.04 lts (bionic) server | |
| 14:26:57 | fungi | looks like i need to force downgrade cryptography because of that | |
| 14:34:14 | fungi | python3.6 -m venv foo | |
| 14:34:18 | fungi | foo/bin/pip install -U pip setuptools wheel | |
| 14:34:21 | fungi | foo/bin/pip install openstackclient python-openstackclient==3.17.0 openstacksdk==0.41.0 python-cinderclient==4.0.1 | |
| 14:34:24 | fungi | sudo foo/bin/openstack --os-cloud=openstackci-rax --os-region-name=DFW --os-volume-api=1 volume list | |
| 14:34:35 | fungi | gtema: that ^ reproduces a working one for me | |
| 14:35:00 | gtema | ugh, now I need to find somewhere py36 for me ;-) | |
| 14:35:01 | gtema | thanks | |
| 14:36:03 | fungi | i'm winding forward with openstacksdk versions to find the one where the first breakage occurs | |
| 14:37:20 | fungi | huh, so i was able to wind forward to openstacksdk==0.48.0 and it continues working | |
| 14:37:58 | fungi | also it may be reproducible on newer python interpreters, i haven't tried yet | |
| 14:38:58 | gtema | okay, in any way now I nearly always fall on policies | |
| 14:39:19 | gtema | I mean either some other py exception of policy | |
| 14:40:31 | fungi | i'm able to make i work with 0.102.0 in fact | |
| 14:40:54 | gtema | lol - so something different in playing a role | |
| 14:41:06 | gtema | welcome in the dependency hell | |
| 14:42:27 | fungi | yes, i'm trying to play with other dependency variables now | |
| 14:43:07 | gtema | what is the issue with v2 itself? | |
| 14:43:19 | gtema | maybe fixing this will cost less then making v1 work | |
| 14:43:20 | fungi | not completely implemented in rackspace | |
| 14:43:24 | gtema | ah, ok | |
| 14:43:45 | fungi | at least that's my understanding, but i can get a more specific error | |
| 14:45:25 | fungi | what was the cinderclient version where v2 api support was dropped? | |
| 14:45:47 | gtema | roughly a year ago | |
| 14:46:05 | gtema | https://opendev.org/openstack/python-cinderclient/commit/6ebee33bf219d42fcd0d4bf2a8c0819bddb24f3c | |
| 14:47:05 | fungi | yeah, if i pin python-cinderclient<8 i only get the deprecation warning for v2 | |
| 14:48:04 | fungi | adding --debug, it looks like it's trying to get /v2/610275/610275/volum | |
| 14:48:31 | fungi | er, it's trying to get /v2/$id/$id/volume/detail | |
| 14:48:42 | fungi | from the block storage api endpoint | |
| 14:48:43 | gtema | ok, seems like a bug | |
| 14:48:50 | fungi | and that returns a 404 not found | |
| 14:50:18 | fungi | server list uses /v2/$id/servers/detail so i guess the doubled project id in the url is a problem | |
| 14:51:11 | gtema | yeah, on the other side v2 cinder is not present in the "service_catalog" - explicitly in the quotes since this is a thing on its own | |
| 14:51:56 | gtema | and using normal OSC it fails to find volumev2 endpoint for region | |
| 14:54:41 | fungi | right, i got that as well once i pinned back a bit | |
| 14:56:34 | frickler | fungi: checking on bridge we set block_storage_endpoint_override to .../v2/610275. maybe drop the last part of that | |
| 14:56:45 | fungi | ooh, good eye | |
| 14:56:56 | opendevreview | Dmitriy Rabotyagov proposed openstack/python-openstackclient master: Implement server and host evacuate commands https://review.opendev.org/c/openstack/python-openstackclient/+/828708 | |
| 14:57:56 | frickler | I haven't found a cli option to override that override | |
| 14:58:14 | gtema | in cli no way | |
| 14:58:40 | gtema | the problem is that this override is only influencing SDK, but not cinderclient itself | |
| 14:59:03 | fungi | frickler: good call. with that edit i'm able to get it work with latest sdk and cli installable on python 3.6, as long as i pin python-cinderclient<5 | |
| 14:59:30 | fungi | i'll push up a change to our clouds.yaml since that simplifies what we need to do to make the cli work | |
| 14:59:50 | gtema | that all sounds like a very dirty hack ;-) | |
| 15:03:45 | frickler | with the same cap it also works on the new bridge with py3.10 | |
| 15:04:02 | opendevreview | HaeIYoon proposed openstack/openstacksdk master: Add image metadef object operations https://review.opendev.org/c/openstack/openstacksdk/+/858350 | |
| 15:04:12 | frickler | hack for maybe-a-non-openstack cloud, but progress anyway | |
| 15:04:41 | gtema | this is exactly what I always want to say - this doesn't feel like openstack cloud | |
| 15:05:28 | frickler | though admittedly there may be an old private cloud running icehouse or similar that might alse need the capped cinderclient | |
| 15:06:03 | frickler | *also | |