Posted	Nick	Remark
#openstack-nova - 2018-06-06
21:45:18	melwitt	alright
21:45:49	dansmith	mriedem: reading the docs for the other weights we have makes me realize we have some that accept negative values
21:46:14	dansmith	so maybe I should make the default -BIGNUMBER and require them to use negative values to properly weigh failures
21:46:15	artom	mriedem, drive by review on https://review.openstack.org/#/c/572790/, will go to sleep soon (travelling back to my time zone)
21:46:25	dansmith	I thought they were all zero/positive
21:46:37	mriedem	dansmith: i know there was an iops one that i saw
21:46:41	mriedem	that was the only one i saw i guess
21:47:49	dansmith	mriedem: well, what do you think?
21:48:02	openstackgerrit	Brianna Poulos proposed openstack/nova master: Add trusted_image_certificates to REST API https://review.openstack.org/486204
21:48:03	openstackgerrit	Brianna Poulos proposed openstack/nova master: Add notification support for trusted_certs https://review.openstack.org/563269
21:48:04	openstackgerrit	Brianna Poulos proposed openstack/nova master: Add certificate validation docs https://review.openstack.org/560158
21:48:26	openstackgerrit	Merged openstack/nova master: Trivial: let internal use only func has _ prefix https://review.openstack.org/572304
21:50:19	mriedem	sorry i got artom'ed
21:50:41	artom	im in ur patch, artoming ur patch
21:50:42	dansmith	ouch
21:50:54	mriedem	dansmith: if negative makes more sense then that's fine by me
21:51:00	melwitt	dansmith: if none of the other weighers auto-negate things, I would think it's more intuitive if it's specified as a negative number. with the thinking that if the normal pattern is "bigger numbers win" then specifying a larger negative number means weight a failure as far less likely to be selected
21:51:01	mriedem	honestly the weight multipliers break my head
21:51:14	dansmith	well, I dunno
21:51:14	melwitt	am I thinking about that right?
21:51:32	dansmith	melwitt: the way I think of all our weighers, the bigger the number, the more significance it has
21:51:43	melwitt	I'd want to pick something that fits in best with what's there so that people who already use it don't have to reverse their thinking for that one option
21:51:50	dansmith	however, there area few where you may want affinity or the opposite, so you can do negative or positive
21:51:52	melwitt	*already use weights
21:52:00	dansmith	you likely never want to prefer hosts that suck,
21:52:19	dansmith	so I dunno that it would ever be positive (if we didn't auto-negate)
21:52:32	mriedem	https://github.com/openstack/nova/blob/master/nova/conf/scheduler.py#L428 is the one i noticed
21:52:44	mriedem	oh and this https://github.com/openstack/nova/blob/master/nova/conf/scheduler.py#L737
21:52:49	mriedem	weight_of_unavailable...
21:53:25	dansmith	that's a bit different
21:54:02	dansmith	and the iops one can be positive or negative with useful meaning
21:54:10	melwitt	do the other negative ones ever make sense as positive values? k
21:54:52	dansmith	yes, ops defaults negative, but changing it to positive means something useful
21:55:07	dansmith	soft_anti_affinity and soft_affinity are both positive, but have opposite meanings
21:55:38	dansmith	https://github.com/openstack/nova/blob/master/nova/scheduler/weights/affinity.py#L93-L96
21:55:39	melwitt	so if it auto negates ... to make build fails even less likely to schedule toward, you would increase the positive value. that is, make build failures heavier
21:55:42	dansmith	Just like I do ^
21:56:19	dansmith	melwitt: right, if it auto-negates (as it does now) then you raise the number to make build failures more significant (i.e. less likely to be chosen)
21:56:31	dansmith	I think that anti-affinity one is my proof that I'm okay
21:56:33	melwitt	yeah, that makes sense to me
21:56:45	dansmith	it auto-negates because it makes more sense to have a positive value for significance
21:56:55	melwitt	yeah it's like a coefficient I guess
21:57:10	mriedem	i left a question in there, but should we put a min value on the multiplier?
21:57:14	melwitt	I think either way is find, both make sense in their own way
21:57:17	mriedem	otherwise anything negative weighs failed hosts higher right?
21:57:19	melwitt	*fine
21:57:45	dansmith	mriedem: I dunno, if somene wanted to actually select recent fails, they could set it to -BIG, which ... maybe?
21:57:50	dansmith	if they do auto-disable on their own,
21:57:59	dansmith	they may want recently-fixed hosts to be super important or something?
21:58:00	mriedem	via notifications or something
21:58:04	melwitt	although allowing negative and positive values could make you favor build failures if you mess up, so maybe that is worse
21:58:10	dansmith	I wouldn't document that, but if you remove the ability, then they can't
21:58:19	mriedem	sure
21:58:24	mriedem	anywho
21:58:24	dansmith	melwitt: right but if you do ^ then maybe you want that
21:58:41	mriedem	watcher could have it's own super awesome auto-disabler
21:58:43	dansmith	I think I'm going to leave it as-is, I convinced myself with the anti-affinity precedent
21:59:30	openstackgerrit	Matt Riedemann proposed openstack/nova master: Block swap volume on volumes with >1 rw attachment https://review.openstack.org/572790
22:01:28	openstackgerrit	Dan Smith proposed openstack/nova master: Change consecutive build failure limit to a weigher https://review.openstack.org/572195
22:14:39	openstackgerrit	Matt Riedemann proposed openstack/nova master: Block swap volume on volumes with >1 rw attachment https://review.openstack.org/572790
22:15:58	dansmith	mriedem: oh, those were all in default previously I guess? we should update that doc
22:17:47	mriedem	i'm sure they are old yeah
22:17:52	mriedem	but we can cleanup the rest in a follow up
22:17:57	mriedem	they likely have a rename alias
22:18:19	dansmith	yeah they do
22:18:28	dansmith	I didn't even process that as the config section,
22:18:37	dansmith	I just copied a block and changed the names
22:19:12	openstackgerrit	Dan Smith proposed openstack/nova master: Change consecutive build failure limit to a weigher https://review.openstack.org/572195
22:25:49	mriedem	in trumps america, i process it for you
22:26:20	mriedem	i think i'll just fast approve given sylvain was +2
22:26:22	mriedem	is that ok?
22:26:42	melwitt	I was going to +2 it once you were okay with it
22:27:05	mriedem	ok all yours
22:28:20	melwitt	en route to the sarlacc put
22:28:29	melwitt	*pit
22:28:46	mriedem	i hope you get that reference
22:28:49	mriedem	otherwise, for shame
22:29:11	mriedem	oh btw,
22:29:13	melwitt	of course I do! I'm hip like that
22:29:19	mriedem	dansmith didn't like evil dead 2
22:29:22	mriedem	as expected
22:29:30	dansmith	correction
22:29:33	melwitt	lol
22:29:37	dansmith	I didn't like the first 45 minutes of it
22:29:44	dansmith	I have no idea about the second 45 minutes
22:30:04	mriedem	did you at least see where he fights his own hand?
22:30:12	dansmith	yeah
22:30:17	mriedem	well alright then
22:30:20	mriedem	also,
22:30:29	mriedem	efried sorta looks like bruce from evil dead
22:30:40	mriedem	efried: you should take advantage of that on halloween
22:30:42	melwitt	*mind blown*
22:30:43	mriedem	or really any day
22:30:44	efried	I've actually heard that before.
22:31:10	mriedem	do you frequently say to your wife, "gimme some sugar, baby"?
22:31:31	mriedem	if so, you might be ash
22:31:42	efried	I may have seen those movies once, 20 years ago. They're not auto-memes that just spring forth from my head, I'm afraid.
22:32:06	mriedem	i saw the first one between slitted fingers when i was probably <10
22:32:10	mriedem	so it's burned in there
22:32:12	mriedem	the granny in the basement

Earlier   Later