Earlier  
Posted Nick Remark
#openstack-sdks - 2021-02-18
14:22:02 gtema right
14:23:11 gtema I suggest gouthamr you make it working and then we figure out where to squeeze it in
14:23:44 gouthamr ack will do, failing currently because i'm missing an important devstack local.conf parameter :)
14:24:18 gtema great
14:24:37 gtema do we have student related (or from students) questions?
14:24:56 gouthamr lemme formally introduce ashrod98[m] (Ashley Rodriguez), NicoleBU (Nicole Chen) and markypharaoh[m] (Mark Tony)
14:25:17 diablo_rojo I for one am STOKED you are all on IRC.
14:25:21 NicoleBU Hello! Its nice to meet you guys :)
14:25:24 ashrod98[m] hello! Nice to meet you
14:25:31 markypharaoh[m] Hi Team !!!
14:25:36 gtema awesome, welcome
14:26:01 gtema I guess you are all also half sleeping?
14:26:08 gouthamr they've started working on adding follow up patches..
14:26:11 markypharaoh[m] No questions from my end.. nicole, ashley?
14:26:31 gouthamr #link https://tree.taiga.io/project/ashrod98-openstacksdk-manila-support/kanban (the manila openstacksdk backlog)
14:26:53 ashrod98[m] nah its 9 am around here so I'm awake. and no questions from me. Just working through some unit testing right now
14:27:18 gtema US is too wide to keep all TZ in the head :)
14:27:27 NicoleBU Same here, I don't currently have any questions
14:27:28 stephenfin o/
14:27:30 gtema but still better than RUS
14:27:31 diablo_rojo Its the west coast that gets you. Three hours is a lot lol.
14:27:52 ashrod98[m] haha yeah 6 am is wayyyy early
14:28:29 gtema what is the timeline for BU to participate?
14:28:56 gouthamr gtema: they're graduating in may
14:29:06 gtema ok, similar to NDSU
14:29:14 gtema just 2 month to go
14:29:17 diablo_rojo Correct.
14:29:22 gtema we are nearly in spring already
14:29:28 diablo_rojo Time flies.
14:29:36 gtema sure
14:29:40 diablo_rojo Except it does more so when we can lol.
14:30:06 gouthamr deep, diablo_rojo
14:30:07 ashrod98[m] Yeah our class presentations are in April. We hope to have few more patches in by then
14:30:18 gtema we should perhaps speed up then, not to waste precious time
14:30:45 gouthamr i've got a couple of questions wrt the base patch that's out there:
14:30:52 diablo_rojo gtema, +1
14:30:54 gouthamr #link https://review.opendev.org/c/openstack/openstacksdk/+/773556/ (Add shared file systems support)
14:30:54 gtema sure
14:31:23 gouthamr so this Version resource question, gtema - https://review.opendev.org/c/openstack/openstacksdk/+/773556/4/openstack/shared_file_system/version.py
14:31:57 gouthamr can we allow this to merge, and address it from all services at once?
14:32:16 gtema I am not sure we can make it properly working that easy
14:32:17 gouthamr or, can we add a versions proxy method?
14:32:28 gtema we do not talk with unversioned endpoint at all
14:32:34 gtema this is all delegated to KSA
14:32:48 gtema I guess this exists historically, but I have never seen it working
14:33:23 gtema versions method on proxy will not work, since you really need to send request to unversioned point
14:33:50 gtema of course we can hack it, but this is not something I like
14:34:12 amotoki I think the current sdk depends on KSA regarding version discovery and we assume clouds.yaml or other info provides the major version info, right?
14:34:22 gtema right
14:34:28 amotoki I am not sure we really query the unversioned URL
14:34:39 gtema exactly - we do not
14:34:51 gtema this is all done in KSA, we get back the versioned one already
14:35:05 gtema we can, however, get list of supported versions from KSA
14:35:55 gouthamr ^ i think my usecase of "what api versions are supported" is met with the get_all_version_data() method from teh connection object
14:36:00 gtema and for that we would not need individual service version resource, I would vote for making it general (if doing it at all)
14:36:06 gouthamr but, i thought we could prettify it a bit
14:36:16 gouthamr can i take a stab at this, post this merging?
14:36:25 gtema gouthamr, generally you don't mess with microversions at all
14:36:30 gouthamr erm, why?
14:36:43 gtema as a user you should be abstracted from that
14:36:50 gtema we did lot to make this transparent
14:37:09 gtema user doesn't need to know in which MV function was added
14:37:29 gtema he just says - I want X, and SDK says, sorry, X is not supported in your case
14:37:49 gouthamr i see, yes - that'd be the case for most users
14:37:51 gtema OSC is in it's side also getting MV handling "dropped"
14:38:35 gtema on the resource level you specify which MV it corresponds to and all operations on it automatically get this MV used (unless server doesn't support it)
14:38:59 tosky talking about that (sorry for jumping in)... can we move forward with this, or an equivalent solution? https://review.opendev.org/c/openstack/python-openstackclient/+/590807
14:39:06 gtema on the functions side you implement a check (or better to say: MV x.y is required for this to work)
14:39:21 tosky (if it's unrelated I'll wait of course; that's a major pain point in "selling" OSC internally in cinder)
14:39:44 gtema we move a bit into different direction, gimme a sec
14:40:34 gtema https://opendev.org/openstack/openstacksdk/src/branch/master/openstack/compute/v2/hypervisor.py#L90
14:41:03 gtema OSC user just tells: I want this and that. SDK verifies with the server whether it is possible or not
14:41:55 gtema for cinder our NDSU students will complete SDK part and we can move OSC to use SDK
14:42:08 gtema with that MV handling will pretty much be offloaded from users completely
14:42:31 gtema we are doing this currently for nova, albeit with no big progress in last month
14:44:03 gtema another example of tricky MV implementation is https://opendev.org/openstack/openstacksdk/src/branch/master/openstack/compute/v2/_proxy.py#L1807
14:44:21 gtema this literally was dropped from OSC and handled in SDK
14:45:15 gtema further questions about MV?
14:45:35 tosky I understand that's an internal restructuring, but when it's time to go back to the cinder part of OSC, it would be better to advertise that in the cinder channels as well
14:45:35 amotoki OSC user still needs to specify MV, right?
14:45:50 gtema amotoki - user CAN, but he doesn't need
14:46:04 gtema highest possible will be used
14:46:16 amotoki gtema: so does user need to specify the major version only?
14:46:30 tosky (and see if anything can be fixed there: apart from this supposed "microversions are not supported" claims which pop up from time to time, another complain is about some choices in the parameters which may not match the expectations on the cinder side)
14:46:42 gtema it depends on the service catalog - pretty much it "picks" major version for us
14:47:22 gouthamr i don't fully understand the version negotiation aspect in https://opendev.org/openstack/openstacksdk/src/branch/master/openstack/utils.py#L98 - i'll read up more
14:47:29 gtema tosky, I know this pops up as a complain often
14:47:39 gouthamr in the meanwhile, i'll submit a patch dropping the versions resource
14:47:45 gouthamr from the manila implementation'
14:47:55 gtema and we thought it make not much sense to continue convincing people, we just need to start doing things
14:48:42 gtema I was hoping to complete together with stephenfin nova part this cycle to be a show case for other teams
14:48:54 gtema not sure, however, we will be in time (busy times)
14:49:09 amotoki gtema: thanks. I will try to follow it up.
14:49:14 gtema cool
14:49:47 gtema summary: SDK negotiates it's own max version, max version supported by server and the one optionally selected by the user
14:49:53 gtema and picks the highest possible
14:50:17 gouthamr ^ thanks for that clarification
14:50:19 gtema this gives user still possibility to fall back to the older one
14:50:44 gouthamr one last thing regarding this $topic
14:50:55 gouthamr #link https://review.opendev.org/c/openstack/keystoneauth/+/774042 (Specify manila microversion header)

Earlier   Later