Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2021-05-25
12:41:17 sean-k-mooney we can desided to limit it in python to Custom_ if we want too or not
12:41:33 gibi bauzas: no worries. I'm on a call, but I will try to multitask :)
12:53:58 admin0 lyarwood, do we save any version info or migration info in the nova database .. i have a nova database which is fairly old ..and i need to figure out if its from newton, ocata or pike
12:58:34 bauzas gibi: I soft+1d again your PUT RP spec
12:58:43 bauzas gibi: with another API alternative
12:59:12 bauzas in my mind, QS parameters are just another way to pass attributes without using data
12:59:36 bauzas so, generally those params should be identical to the resource attributes
12:59:50 gibi QS are not part of the object representation, while the body is. So whatever we add to the body we make it part of the object as per REST
13:00:08 bauzas gibi: I proposed to use a sub-resource for this
13:00:37 bauzas /force is maybe a bad name
13:01:23 gibi resource_providers/{uuid}/force feels like an action API that is also not that restful
13:02:21 sean-k-mooney bauzas: ok just pushed my comments on your spec
13:02:25 gibi bauzas: so you feel that we cannot add param to the QS that is not part of the body?
13:02:59 sean-k-mooney admin0: i belive we do have a version for the schema yes
13:03:07 sean-k-mooney admin0: i think there is a nova manage command to get it
13:03:31 gibi admin0: mysql> select * from migrate_version;
13:03:34 gibi | repository_id | repository_path | version |
13:03:39 sean-k-mooney admin0: nova-manage db version
13:03:40 gibi | nova | /opt/stack/nova/nova/db/sqlalchemy/migrate_repo | 422 |
13:03:43 gibi +---------------+-------------------------------------------------+---------+
13:03:57 gibi admin0: or wha sean-k-mooney suggessts
13:04:14 sean-k-mooney ^ is teh db table its reading form
13:04:40 sean-k-mooney so if you have a backup you have not loaded gibi's way might be quicker to grep for
13:05:04 sean-k-mooney there is also nova-manage api_db version
13:05:06 bauzas gibi: as I said in my last comment, the current API is not RESTful either way
13:05:18 sean-k-mooney nova-manage db version tells you the cell db version
13:06:23 sean-k-mooney bauzas: the resful way to update the parent uuid woudl be jsut to update it
13:06:35 sean-k-mooney so a put to resource_providers/{uuid}
13:06:41 sean-k-mooney with the uuid updated in the body
13:07:04 bauzas sean-k-mooney: technically, put would be for updating the whole resource
13:07:22 sean-k-mooney if you wanted to do a partial update we woudl use patch
13:07:24 bauzas but we're nitpicking
13:07:29 sean-k-mooney but we dont use patch in openstack
13:07:44 bauzas my biggest concern comes to the fact I feel a qs param is just hacky
13:07:49 sean-k-mooney it often comes up but we keep saying no to patch
13:08:06 bauzas just for the idea of "oh shit, we need to do *something*"
13:08:15 sean-k-mooney bauzas: well i was orginaly suggesting just allowing this by defult
13:08:49 bauzas sean-k-mooney: I proposed three alternatives : one be PATCH, one be adding a new HTTP header
13:09:01 bauzas and the last being to have a specific subresource
13:09:02 sean-k-mooney im not conviced we need to have a flag of any kind to opt into it but im not against having one
13:09:20 bauzas sean-k-mooney: see my comments on PS2, I was thinking like you
13:09:33 bauzas but looks like we agreed at the PTG to sanity check this
13:09:46 sean-k-mooney ya the header occured to me but not sure we have precendent for that
13:09:52 sean-k-mooney we did
13:10:33 sean-k-mooney i did not have stong feels for or against added a sanity check with a slight one for not doign that
13:10:46 sean-k-mooney ill go read your comment and catch up
13:11:32 gibi I'm on the side of either not add a check (but we agreed on the PTG to add it) or if we add then for me the extra parameters is better fits to the QS than to header or body or subresource
13:11:59 bauzas gibi: hence my +1 :)
13:11:59 sean-k-mooney bauzas: if we were to have a subresouce i actully woudl go with PUT /resource-providers/{uuid}/uuid/{new uuid}
13:12:22 bauzas gibi: I don't wanna hold on this, but I don't appreciate the QS param outcome
13:13:01 sean-k-mooney sorry
13:13:26 sean-k-mooney put /resource-providers/{uuid}/parent_uuid/{parent}
13:13:34 sean-k-mooney with an empty body
13:13:57 sean-k-mooney that basically woudl work like patch without using it
13:14:42 bauzas this sounds quite good to me
13:14:51 bauzas but again, I'm just one
13:14:57 sean-k-mooney you also have to expressly opt into that url
13:16:09 gibi bauzas: sorry I don't get why the current PUT is not RESTful
13:16:16 gibi you have to PUT all the attributes
13:16:26 gibi that you POSTed before
13:16:35 bauzas I guess the universe entropy is probably smaller than a discussion of 3 engineers about specing an API endpoint
13:17:01 bauzas count a 4th and you'll wait for the big crunch
13:17:53 bauzas gibi: it's not RESTful in the sense you can't update all the attributes as of now
13:18:22 gibi bauzas: you have restriction of the value of a field yes, but you have to still list all the fields in the PUT
13:18:24 sean-k-mooney bauzas: i think gibi was saying you would get teh current state update teh parent field and then put back the updated state
13:18:32 gibi sean-k-mooney: yes
13:18:39 gibi for me this is the restful put
13:18:52 sean-k-mooney if you do a update of the full resouce yes
13:18:56 bauzas this
13:19:18 bauzas if you just PUT a name, this isn't RESTful
13:19:37 gibi but the current PUT is not partial and I'm not suggesting any partial PUT
13:19:40 gibi either
13:19:46 bauzas and if you PUT a parent uuid which is not None while the resource parent is not None, then it's not RESFul either
13:20:06 bauzas because you depend on the state of the resource
13:20:23 bauzas but we're bikeshedding I guess
13:20:28 sean-k-mooney gibi: right i think bauzas was assuming you wanted to change that
13:20:43 sean-k-mooney bauzas: were you assuming gibi was suggeting using put for a partial update
13:20:57 gibi so today, PUT takes 3 paramters, name, uuid, and parent_uuid. After my change PUT takes the same parameters
13:21:07 bauzas actually, I was wrong
13:21:11 gibi just the value restriciotn of parent_uuid is relaxed
13:21:12 sean-k-mooney gibi: maybe we coudl put json exmaple into the spec to make that clear
13:21:14 bauzas name and uuid aren't optional
13:21:21 bauzas but parent_uuid is
13:21:37 gibi I assume parent_uuid field is not optional, it just allow taking the value of None
13:21:57 bauzas gibi: tbc, I'd be OK with a parent_uuid QS param name
13:22:05 sean-k-mooney gibi: right that is how i think of that too
13:22:37 gibi the API ref says optional, but I'm not sure that it is just due to the microversion dependency
13:22:39 bauzas but I'm not OK with some allow_reparenting name
13:22:40 sean-k-mooney the parent is an intrinic part of the resouce but None is a valid and default value
13:23:06 sean-k-mooney bauzas: you can change the name whenever you like
13:23:16 sean-k-mooney oh sorry
13:23:19 sean-k-mooney misread that
13:23:37 sean-k-mooney bauzas: you ment you are not ok with teh allow_reparenting query arg name
13:24:24 sean-k-mooney bauzas: i would prefer allow_reparenting over force
13:33:38 openstackgerrit Merged openstack/nova-specs master: Set minversion of tox to 3.18.0 https://review.opendev.org/c/openstack/nova-specs/+/791975
13:33:39 sean-k-mooney gibi: ill get back to the reparenting spec later today. left my toughts on this as comments but i have no stong feeling agaisnt the current query arg approch nesssarly.
13:33:57 gibi bauzas, sean-k-mooney: thanks
13:46:22 masterpe I have placed indexes on the deleted_at column on all the tables on the databases nova, nova_api and nova_cell0, this speeds this nova-manage db archive_deleted_rows with multiple minutes per run. First it was 19 minutes now it is 19 seconds.
13:46:38 masterpe So I think this is a bug.
13:47:01 gibi masterpe: please file a bug. and feel free to propose a patch that adds the index

Earlier   Later