Earlier  
Later
New Search
Posted Nick Remark
#openstack-sdks - 2019-08-15
20:56:50 mordred efried: cool thanks
20:56:50 efried because allow_version_hack=True we're actually returning an adapter rather than raising an exception.
20:56:50 efried BUT
20:56:50 efried mordred: From what I can tell, we're (correctly) getting a conn error while the service is still coming up
20:56:59 mordred oh
20:57:02 mordred that's ...
20:57:42 efried I noticed there's a few places we explicitly set allow_version_hack=True, but afaict that's the default in ksa anyway.
20:57:46 mordred this is really a behavior design issue that I think we're going to need to fix - and is an actual valid usecase for wanting the other behavior
20:58:10 mordred sdk is assuming that an end-user is wanting to talk to a cloud, and that the cloud exists, so it does what it can to get them _something_
20:58:28 mordred but for nova - you want the inverse - you *want* an error when there is an issue with the remote servie
20:58:44 mordred beause you're more assured that you know what you're doing so if you get an error it's because there is an error
20:58:49 efried I don't think this is one of the sdk paths where we explicitly set allow_version_hack=True. I think we're not setting that, but getting True because that's the default.
20:59:11 efried that being the case, a possible solution is to default allow_version_hack=False *except* in those cases where we set it explicitly.
20:59:14 efried Three of them, I believe.
20:59:34 efried openstack/service_description.py:229: allow_version_hack=True,
20:59:34 efried openstack/service_description.py:205: allow_version_hack=True,
20:59:34 efried openstack/service_description.py:109: allow_version_hack=True,
20:59:34 efried efried@efried-ThinkPad-W520:~/openstack/openstacksdk$ git grep -n allow_version_hack
20:59:51 mordred yeah - but regardless of that setting, sdk is going to, as it currently stands, try REALLY hard to not throw an exception for you
21:00:13 efried mm. Then yeah, I need a way to pass an option that says "throw an exception unless what you give me is really working"
21:00:17 mordred I kind of think we need a connection flag "raise_on_missing_service" or something named better than that
21:00:19 mordred yeah
21:01:19 mordred because contrary to an end-user who may be trying to use a weird cloud, or one that's configured weirdly and may not know all the ins and outs - it is *expected* that an admin will have told nova about anyhting weird in their cloud
21:01:35 mordred and it is much better to get a hard logged error so the admin can investigate the misconfig
21:01:45 efried for sure.
21:01:56 mordred efried: I have a thought ... let me see if I can make a patch
21:02:15 efried fwiw, the same sequence with ironicclient looks like this: https://zuul.opendev.org/t/openstack/build/402a73a9238643c2b893d53b37a6ce27/log/controller/logs/screen-n-cpu.txt.gz?severity=3
21:02:50 efried wait, sorry, that's not right.
21:02:58 mordred efried: you get "virtdriver not ready" when the service pointed to by the catalog is not there, yes?
21:04:24 efried VirtDriverNotReady is raised by the Ironic driver itself when it encounters an error trying to get the client (in whatever form)
21:04:58 efried You can see it's getting the same error, but eventually it succeeds.
21:04:58 efried https://zuul.opendev.org/t/openstack/build/402a73a9238643c2b893d53b37a6ce27/log/controller/logs/screen-n-cpu.txt.gz?severity=3
21:04:58 efried This is from the previous patch, which still uses sdk, but doesn't *rely* on it really early on, for node_list:
21:06:09 efried actually, I think we're seeing that error for baremetal while we're building conn for a different service
21:06:32 efried Here's a "normal" one, from before attempting to use sdk for ironic: https://zuul.opendev.org/t/openstack/build/cbe668cd674747168ac05759e9e147f3/log/controller/logs/screen-n-cpu.txt.gz?severity=3
21:07:01 efried That StrictVersion exception, which I think comes from ironicclient when the service isn't ready, is the one we're translating to VirtDriverNotReady.
21:08:26 mordred yeah - makes sense - you're asking what version the service is and you're getting nothing because there is no service
21:09:42 efried Okay, so that gels - that's raising an exception and nova is set up to deal with that and retry; but the sdk path isn't raising, so we thing things are fine and actually (probably) bounce trying to access the real thing.
21:10:36 mordred yeah- and what's worse, we're never going to re-try version discovery beacuse teh proxy will already be constructed
21:11:14 mordred I think I've got a good handle on what we want the behavior to be
21:11:44 mordred now if I can just translate that into python words
21:31:05 efried mordred: FWIW, I think I'm going to want to use this option universally from nova.
21:31:14 efried I just did some local twiddling with a placement service
21:31:25 efried and (as expected, I think) I get a Proxy both times.
21:32:13 efried here's the diff in the logs (< is service down) http://paste.openstack.org/raw/757499/
21:36:12 efried I tried setting allow_version_hack=False but it still returns the proxy
21:36:45 openstackgerrit Monty Taylor proposed openstack/openstacksdk master: WIP Add a strict_proxies option for services https://review.opendev.org/676829
21:36:57 mordred efried: ^^ how about something liek this
21:37:22 efried mordred: does it need to be on top of those other patches?
21:39:12 mordred efried: no - that's just my local working state - that's a very unclean patch
21:39:22 efried mordred: Okay.
21:39:40 efried So right off the bat: I just tried allow_version_hack=False locally and it still returns the proxy.
21:39:44 mordred efried: (wanted to get the first stab down so I didn't lose thinking context during dinner)
21:40:04 efried so I think that path should still have a check and raise, maybe on that get_endpoint()
21:40:05 mordred efried: yeah - I think there are some other code paths in there that will return you something
21:40:40 efried no, this was for placement, which doesn't have a shim, so it should be hitting that code path.
21:40:55 efried which is actually confirmed by the logging - I don't see that "Fallback" message, which is only emitted when allow_version_hack=True.
21:41:26 mordred ah - gotit. so maybe check that get_endpoint() actually returns something?
21:42:14 efried yeah, exactly
21:42:21 mordred what if ...
21:42:52 mordred if self._strict_proxies - we do temp_client.get_endpoint_data() instead of get_endpoint() - and check to make sure endpoint_data has a version?
21:43:10 efried do all services have a version?
21:45:21 mordred everything except swift I think
21:45:46 mordred but I'm not 100% sure if that's the right thing to test
21:46:22 openstackgerrit Monty Taylor proposed openstack/openstacksdk master: WIP Add a strict_proxies option for services https://review.opendev.org/676829
21:46:33 mordred efried: ^^ shotgun to see if we can get one of the things to trigger :)
21:46:56 efried mordred: What's weird is, I'm making sure that get_endpoint[_data] should fail, but the proxy is still returning.
21:46:56 mordred and then figure out which one is the RIGHT one to trigger on when the brain has more available brainjuice
21:47:10 efried is there a try/except somewhere up above?
21:47:16 mordred there shouldn't be
21:47:27 mordred and that is weird
21:47:41 efried I'm not getting an actual exception until I try to use the thing.
21:47:46 mordred you don't have that patch that added a placement service_description in do you?
21:48:34 mordred efried: (I'm probably going to want to construct a bunch of weird tests for this one so we can be sure we've got this covered)
21:49:20 efried mordred: Well, yes, I have the commit that added placement service_description, but that only happens in a test fixture, when you setUp the fixture, which I'm not doing locally.
21:50:46 mordred weird
21:51:46 mordred efried: I unfortunately have to adjourn to the dinner - if you find anything, let me know - otherwise I'll pick it up first thing in the morning and see if I can't construct some catalogs/request_mock tests we can use
21:52:11 efried mordred: Thank you sir. I'll noodle with it for a few more minutes before I, too, have to run (and choke some people)
21:59:51 efried mordred: a couple of typos notwithstanding, strict_proxies is a success.
22:00:56 mordred efried: yes? SWEET
22:01:09 mordred efried: then in the morning I'll work on tests and cleaning it up
22:01:28 efried mordred: Sweet, then we'll need 0.35.0 pretty quick.
22:01:48 efried I'll mark up the typos (at least the ones I saw), least I can do :P
22:03:04 efried mordred: It works for placement... I'm not sure it'll work for ironic. I need to dig more.
22:54:54 adriant mordred: how/when do we want to do the MFA work?
23:01:36 openstackgerrit Eric Fried proposed openstack/openstacksdk master: WIP: Connection(strict_proxies) https://review.opendev.org/676837
23:01:56 efried mordred: there's my spin on it ^
23:02:02 adriant I don't know your timetable, but wanna organise a meeting time and I can stay up late so we can discuss and plan?
23:02:05 adriant mordred: ^
23:09:59 openstackgerrit Eric Fried proposed openstack/openstacksdk master: WIP: Connection(strict_proxies) https://review.opendev.org/676837
23:16:36 openstackgerrit Eric Fried proposed openstack/openstacksdk master: WIP: Connection(strict_proxies) https://review.opendev.org/676837
#openstack-sdks - 2019-08-16
02:45:24 openstackgerrit Brin Zhang proposed openstack/python-openstackclient master: Microversion 2.77: Support Specifying AZ to unshelve https://review.opendev.org/665336
11:27:02 openstackgerrit Monty Taylor proposed openstack/openstacksdk master: WIP: Connection(strict_proxies) https://review.opendev.org/676837
11:27:31 mordred efried: I like your version. that ^^ fixes it so that the existing tests work - I'm gonna add new tests now
12:31:35 efried mordred: awesome. I will try to spin up a nova patch that Depends-On it, but I've had trouble making that work recently.
12:32:01 openstackgerrit Monty Taylor proposed openstack/openstacksdk master: Add strict_proxies option for Connection https://review.opendev.org/676837
12:32:20 mordred efried: that now has tests (that showed some issues, so they're good tests :) )
12:32:36 efried good stuff
12:33:09 mordred efried: I turned on strict_proxies in the from_conf tests - since the overlap of the two is gonna be pretty common
12:33:38 efried mordred: Do you think there's value in getting some real cloud to play around with before we unleash this on the world?

Earlier   Later