Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2018-10-01
19:43:51 openstackgerrit Jack Ding proposed openstack/nova master: Add HPET timer support for x86 guests https://review.openstack.org/605902
19:44:08 mriedem gryf: ask artom
19:44:17 gryf also I get the reading of etherpad from ptg, and there was an agreement for modeling numa nodes as RPs
19:44:35 dansmith gryf: artom would love to talk to you about it at length over a candle-lit dinner
19:44:39 sean-k-mooney cdent: im a little confused by that bug. what spec are you refering to that is violated pep 3333?
19:45:13 gryf as far as I can tell, bauzas was the one who is dealing with that modeling, right?
19:45:27 openstackgerrit Jack Ding proposed openstack/nova master: Correct instance port binding for rebuilds/reboots https://review.openstack.org/603844
19:45:28 cdent sean-k-mooney: initially I thought the code is violating pep3333
19:45:32 gryf dansmith, I'd like to skip romantic dinners ;)
19:45:51 cdent sean-k-mooney: so when I "the spec" I'm referring to pep3333
19:45:55 openstackgerrit Jack Ding proposed openstack/nova master: Handle missing marker during online data migration https://review.openstack.org/605164
19:45:57 cdent but the problem is more messy than just that
19:46:21 melwitt mriedem: two patches to un-deprecate nova-consoleauth start here https://review.openstack.org/605250
19:46:40 gryf so, artom, you around?
19:46:50 sean-k-mooney cdent: so " the content of native strings must still be translatable to bytes via the Latin-1 encoding" woudl seam to be in conflict with the http spec which allow unicode in URL such as the location header
19:49:00 mriedem melwitt: so you plan on undeprecating in stein, and backporting that change to stable/rocky, then re-deprecating in stein?
19:49:01 cdent it's the source of much argument
19:49:06 mriedem or only re-deprecate once cells v1 is gone?
19:49:08 melwitt mriedem: yes
19:49:51 cdent sean-k-mooney: my understanding, however, was that you couldn't put unicode codepoints in headers, but you could have urls that encode unicode, which is not quite the same thing
19:50:16 melwitt mriedem: and I'll stack the patches that remove use of nova-consoleauth by non-cells v1 code on top of the re-deprecation patches
19:51:01 sean-k-mooney cdent: that definetly used to be the case but i was under the impression that restiction was relaxed in the last few years
19:51:35 cdent sean-k-mooney: that may be, but wsgi itself has not caught up/changed and has its own contract
19:56:13 mriedem melwitt: have you cherry picked those changes to stable/rocky and built the release notes to see if they are going to show up weird at all?
19:56:14 sean-k-mooney cdent: ok so https://www.rfc-editor.org/rfc/rfc3986.txt relaxes the constraitn but not entirely. "on-ASCII
19:56:17 sean-k-mooney characters must first be encoded according to UTF-8 [STD63], and then
19:56:19 sean-k-mooney each octet of the corresponding UTF-8 sequence must be percent-
19:56:21 sean-k-mooney encoded to be represented as URI characters
19:56:23 sean-k-mooney "
19:56:24 mriedem rewriting release note history makes me uneasy
19:58:24 sean-k-mooney cdent: so basically if you want to transport utf-8/unicode charaters in a url you must htmlencode the underlying utf-8 byte syequence rather then use the raw utf-8 charater stream
20:00:48 cdent sean-k-mooney: urlencode, not htmlencode, but yes, pretty much
20:00:50 melwitt mriedem: no -- but let me do that and verify it looks ok
20:01:52 sean-k-mooney cdent: sorry yes, i could not remember the normal name for the % encoding
20:02:34 melwitt mriedem: I guess, did you have something else in mind when you were thinking of un-deprecation? i.e. just modifying the code side and not touching the release note side?
20:04:52 sean-k-mooney melwitt: if its going to "remain" deprecated in stein after the modification is there any reason to un deprecate it?
20:05:47 melwitt sean-k-mooney: just a bad signal to folks moving to rocky. it will make them think they can stop running nova-consoleauth in rocky, but I found the code still requires it
20:08:02 sean-k-mooney melwitt: oh ok i was missing that context. the console auth service is technically optional so i dont neccicarly think that is a problem
20:08:15 openstackgerrit melanie witt proposed openstack/nova stable/rocky: Revert "Deprecate the nova-consoleauth service" https://review.openstack.org/607037
20:08:15 openstackgerrit melanie witt proposed openstack/nova stable/rocky: Un-deprecate the nova-consoleauth service https://review.openstack.org/607038
20:09:15 sean-k-mooney for example i belive you can configure nova/libvirt dirver to have no console correct?
20:09:32 melwitt sean-k-mooney: a couple of users already tripped over it, caused confusion and wasted time
20:10:57 sean-k-mooney melwitt: right, then other than the possible weirdness with the release notes i guess undeprecating and deprecating in stein is proablly a better user experice.
20:11:56 melwitt yeah. I didn't consider the possibility of un-depcrecating it in the code only, and leaving the release notes alone. I had been thinking I should back out every mention of it
20:14:08 artom gryf, hey, so, I can't really talk now, but I'll be around later tonight, or tomorrow
20:14:56 openstackgerrit Balazs Gibizer proposed openstack/nova master: Ignore forcing of evacuation for nested instance https://review.openstack.org/606111
20:14:56 openstackgerrit Balazs Gibizer proposed openstack/nova master: Run negative server moving tests with nested RPs https://review.openstack.org/604125
20:15:49 gryf artom, tomorrow will be fine
20:19:16 mriedem melwitt: i didn't really have something specific in mind, because it's a weird issue
20:19:33 mriedem if nova-compute in stein is expecting stuff in the consoleauth service, it shouldn't be deprecated
20:19:55 mriedem but is that only true if the token isn't in the cell db?
20:20:04 mriedem or you're using cells v1 - which is itself deprecated
20:20:48 mriedem as an aside, it seems we should have something in grenade for consoles....create a console on a server on the old side and verify the console token is still valid on the new side...
20:21:07 mriedem i'm not sure how easy that is though
20:26:22 melwitt well, in rocky the console proxy by default will _not_ fall back to nova-consoleauth -- you have to use the [workarounds]enable_consoleauth to make it look there. the snag was in the fact that compute/api stores the token in both nova-consoleauth and the database, and that's an artifact that I didn't understand from the original patches from paulmurray
20:26:46 melwitt that's the non-cells v1 case. for cells v1, it will always need nova-consoleauth
20:38:44 sean-k-mooney melwitt: so the only case where you always need console auth is cells v1 which is deprecated and for non cells v1 you only need it if you enable a workaround which is off by default
20:39:43 melwitt sean-k-mooney: non cells v1 you need it in rocky even if you didn't enable the workaround because compute/api is still storing a token auth in nova-consoleauth regardless. the console proxy doesn't have to have nova-consoleauth
20:40:20 sean-k-mooney melwitt: oh ok i see thank i missed that bit
20:40:37 mriedem i'm about to get kicked out of this coffee shop
20:40:44 melwitt it's confusing
20:41:10 mriedem maybe write something up to the ML to help digest and sort it out?
20:41:13 mriedem so others are on the same page?
20:41:20 dansmith melwitt: would changing the default just be a better route?
20:41:25 mriedem and give operators a heads up before they get to rocky
20:41:30 dansmith we _want_ it deprecated and you shouldn't need it right?
20:42:06 dansmith or is there something we actually missed in the conversion to service-less?
20:42:16 dansmith if the latter, why not just fix and backport that?
20:42:25 melwitt dansmith: yeah, but to "fix" that, we'd have to remove the calls in compute/api that regardlessly store the token auth in both the consoleauth service _and_ the database. me missing that fact is why it got deprecated in the first place
20:43:05 melwitt the proxy code was optionally using nova-consoleauth, the compute/api code was unconditionally using nova-consoleauth
20:43:35 dansmith is it just not making an rpc call unless the workaround is on?
20:44:09 melwitt yeah, the proxy won't fall back on nova-consoleauth if the workaround option isn't enabled
20:44:18 dansmith what I mean is,
20:44:33 dansmith can we just not make the rpc calls from the compute api code unless the workaround is enabled?
20:44:49 dansmith will that fix the problem such that they don't have to run the service?
20:44:51 melwitt yeah, I think we could do that
20:45:11 dansmith that seems massively better to me than confusing everyone (including some of us here) with a backported un-deprecation dance
20:45:27 dansmith since that was the original goal, we just missed some of it (right?)
20:45:43 melwitt I think that should do it, obviously after this fiasco I will have to test it and make 100% sure it would work
20:46:02 melwitt before I claim that's all we need
20:46:06 dansmith yeah, that'd be good :)
20:46:30 dansmith did we deprecate and not also change devstack to not deploy?
20:46:37 dansmith deploy consoleauth I mean
20:46:43 melwitt yeah
20:46:46 dansmith I would think that would have caught this
20:46:48 dansmith ack
20:46:50 melwitt aye
20:47:14 melwitt yeah, it'd have to do a "if cells v1" before deploy/not deploy but I think that would be fine
20:47:56 melwitt and maybe I could rig up a DNM test for the workaround being enabled to test that in the gate too
20:48:44 dansmith we have cellsv1 conditions
20:52:35 sean-k-mooney melwitt: you could make the rpc conditional and then make a devstack change that depends on it that disables deploying the consoleauth service to test it
20:53:12 melwitt yeah, that's what we're talking about doing
20:54:05 sean-k-mooney cool and then the DNM test change is to reassert that if consoleauth is deploy the workaround still well works
20:56:15 melwitt yeah, that's what I was thinking
21:39:02 openstackgerrit Chris Dent proposed openstack/nova master: Clean up header encoding handling in compute API https://review.openstack.org/606967
21:44:13 openstackgerrit Merged openstack/nova stable/queens: Revert "Make host_aggregate_map dictionary case-insensitive" https://review.openstack.org/605266
21:44:19 openstackgerrit Merged openstack/nova stable/queens: Enforce case-sensitive hostnames in aggregate host add https://review.openstack.org/605267
21:59:15 openstackgerrit Merged openstack/nova stable/pike: Fix instance evacuation with PCI devices https://review.openstack.org/590059
22:04:16 openstackgerrit Merged openstack/nova stable/pike: Update nova network info when doing rebuild for evacuate operation https://review.openstack.org/590070
22:37:26 mriedem so has the great consoleauth undeprecafacation been sorted out?

Earlier   Later