Earlier  
Posted Nick Remark
#openstack-nova - 2018-02-15
17:22:58 bauzas oops, jaypipes too ^
17:23:57 dansmith my only concern would be the dependent specs, if people want those landed first
17:24:01 dansmith but I don't mind either way really
17:24:55 stephenfin mriedem: I think you identified what he was looking for, which was different to what I thought. Given that the caching scheduler is deprecated (I didn't know that), we don't need to worry about IMO
17:25:42 mriedem bauzas: it is certainly possible to know if we're doing a first schedule or a reschedule,
17:25:48 mriedem RequestSpec.retry has that information
17:26:00 bauzas mriedem: for a reschedule yes
17:26:05 bauzas mriedem: for a move operation, nope
17:26:08 edleafe bauzas: that's cool. At least there has been some discussion since my comment
17:26:11 mriedem bauzas: that's because https://review.openstack.org/#/c/505771/
17:26:49 mriedem well, the issue in ^ confuses the fact that you're doing a reschedule or a move
17:26:57 bauzas shit
17:27:26 bauzas how many times will I regret to not have thought more on persisted fields for the RequestSpec object ?
17:27:47 openstackgerrit Claudiu Belu proposed openstack/nova master: compute: Cleans up allocations after failed resize https://review.openstack.org/543971
17:27:57 jaypipes bauzas: hold please.
17:27:58 cdent bauzas: feel free to +w the spec if there's been enough feedback
17:28:00 jaypipes still reviewing
17:28:01 mriedem i thought that would be an easy fix but then second guessed myself, and need to write a functional test to be sure
17:28:07 cdent bauzas: except for jay :)
17:28:26 bauzas jaypipes: ack
17:29:52 efried Could rebase it on top of the other spec
17:31:29 dansmith or someone could slam it in: https://review.openstack.org/#/c/544694/
17:31:42 dansmith it's really simple, has a lot of +1s from stakeholders and only needs a +W
17:31:50 mriedem looking
17:32:04 mriedem i will ram this down your throat like obamacare to the republicans
17:32:18 mriedem lemme call nancy quick
17:32:25 dansmith lolol
17:33:01 openstackgerrit Jay Pipes proposed openstack/nova-specs master: mirror nova host aggregates to placement API https://review.openstack.org/545057
17:34:21 dansmith apparently by "lots" I meant "two" but.. you know
17:34:53 mriedem consider it rammed
17:35:03 mriedem thanks obama
17:35:19 mriedem the two that matter
17:38:51 jaypipes dansmith: cern doesn't use the cachingscheduler does it?
17:39:08 mriedem tssurya: ^
17:39:24 dansmith jaypipes: presumably not because they're working through placement issues right now
17:39:34 jaypipes ak
17:40:50 openstackgerrit Merged openstack/nova-specs master: Support member_of param for allocation candidates https://review.openstack.org/544694
17:42:05 tssurya mriedem, jaypipes, dansmith : no we don't
17:42:25 mriedem excellente
17:47:19 gibi dansmith: I also left some comments / questions on https://review.openstack.org/#/c/544585 just now
17:51:39 openstackgerrit Surya Seetharaman proposed openstack/nova master: Extending delete_cell --force to delete instance_mappings https://review.openstack.org/540073
17:52:54 bauzas gibi: dansmith: I got the name. Let's call it "scheduler transformers" and name the first one "autobot"
17:53:07 dansmith gibi: replying
17:53:08 mriedem gibi: i cleaned up the commit message on this and added a simple unit test for detach failing, see if you still like it ^
17:53:09 mriedem oops
17:53:13 openstackgerrit Matt Riedemann proposed openstack/nova master: Detach volumes when VM creation fails https://review.openstack.org/528385
17:53:15 mriedem gibi: ^
17:53:28 bauzas please folks, don't tell me about Shia LaBeouf or anything stupid like the movies
17:53:39 gibi mriedem: looking
17:53:55 bauzas they killed my childhood
17:58:58 mriedem gibi: and left a comment
17:59:09 mriedem deleting the volume before we reschedule is kind of dumb
17:59:24 mriedem let's say i boot from volume with a pre-existing volume,
17:59:32 gibi mriedem: oops, +Wd it
17:59:36 mriedem spawn on the first host fails after i've attached the instance to the volume,
17:59:47 mriedem we then detach and delete the volume, and reschedule,
17:59:55 mriedem then we get to the 2nd host, and attach will fail b/c the volume is gone
18:00:05 mriedem well i was +2,
18:00:08 mriedem and might still be,
18:00:27 mriedem because this is a latent thing that's always been in here, but i just don't think it's ever worked because of the bug he's fixing
18:00:51 gibi mriedem: I see. I guess this is a new bug report then
18:02:40 gibi mriedem: we fixed something that allowed another latent bug to manifest itself
18:03:02 mriedem at least delete_on_termination defaults to False...
18:03:15 mriedem this is also kind of why i didn't like the spec proposed for doing delete_on_termination with ports
18:03:22 mriedem since nova is not good at orchestrating this
18:04:28 gibi I also think that nova should only delete these if nova was the one that created them in the first place. But it would be even better if nova would never create these things
18:04:43 mriedem yeah that's the exact thing i was thinking
18:04:51 mriedem and we handle that with ports by storing a flag if nova created the port or not
18:05:00 mriedem in the nw info cache
18:05:33 mriedem because if nova did create this volume and attach it, then we fail and reschedule but don't delete it, we'll create another volume on the next host we try
18:05:38 mriedem and never cleanup the old one
18:06:04 mriedem and given enough retries, you could also hit quota limits on creating volumes
18:06:32 mriedem oh, well we do know if nova created it actually,
18:06:38 mriedem if source_type=volume, it's pre-created
18:08:26 mriedem ok i left some notes in the review so we can remember this conversation 2 years from now
18:08:30 mriedem when it's still a bug :)
18:09:23 imacdonn was something fixed recently-ish w.r.t. ports on reschedule .... seems I've seen issues with an exception, something like PortInUse, when an instance creation fails and gets rescheduled on another compute node, but that probably was with Ocata
18:09:25 openstackgerrit Merged openstack/nova-specs master: Add placement-req-filter spec https://review.openstack.org/544585
18:10:42 mriedem imacdonn: that's pretty vague
18:10:43 openstackgerrit Lee Yarwood proposed openstack/nova master: DNM: Test LM with encrypted volumes https://review.openstack.org/536350
18:10:47 mriedem https://github.com/openstack/nova/commit/08d24b733ee9f4da44bfbb8d6d3914924a41ccdc went into newton
18:10:52 gibi mriedem: thanks for that note
18:10:52 mriedem *mitaka
18:11:15 mriedem imacdonn: i have this https://review.openstack.org/#/c/520248/
18:11:21 mriedem but might not be what you're asking about
18:11:29 imacdonn mriedem: yeah, I know ... I'd have to try to come up with a way to reproduce it ... but it's likely that the port was not created by nova, in my case .. because I use heat a lot, and usually create the port separately there
18:11:31 openstackgerrit Lee Yarwood proposed openstack/nova stable/pike: DNM: Test LM with encrypted volumes https://review.openstack.org/545074
18:12:36 mriedem imacdonn: in that case nova shouldn't try to delete the port
18:12:40 mriedem just unbind it
18:12:52 gibi dansmith: thanks for the reply on the request filter spec. I have a better view now, so I'm +1
18:12:56 mriedem if nova didn't unbind it and a reschedule tried to bind that host to another host, that would fail
18:13:03 dansmith gibi: np, thanks for those comments
18:13:35 openstackgerrit Matt Riedemann proposed openstack/nova stable/queens: Detach volumes when VM creation fails https://review.openstack.org/544144
18:14:06 imacdonn mriedem: yeah. I'm fairly sure I've seen this with Ocata .. but it's one of those things that only happens when something else goes wrong (to cause the initial creation failure), so it never got hot enough for me to really dig into ... you just reminded me of if when you mentioned ports above
18:15:34 mriedem at the risk of angering the gods, i think we actually do a relatively decent job of managing ports and cleaning up after ourselves
18:15:39 mriedem with volumes, not so much
18:16:16 imacdonn heh. I'll try to find time to attempt to reproduce it ... but my "round tuit" supply is low
18:16:38 mriedem i understand
18:18:22 imacdonn I guess you really can by ANYTHING on Amazon! https://www.amazon.com/Round-TUIT-Tokens-Multi-Pack-Encouragement/dp/B00J8KVHLY
18:18:45 mriedem heh, find me a hang in there poster with a kitten that doesn't suck

Earlier   Later