| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-04-07 | |||
| 16:31:32 | johnthetubaguy | tonyb: I had a quick go at kicking this off, as I hadn't seen anything go up: ^ | |
| 17:09:53 | samarendra | what does the nova.conf on the controller actually do if it isn't also an hypervisor ? | |
| 17:10:23 | mriedem | samarendra: different options are used by different services | |
| 17:10:30 | samarendra | it doesn't seem like it configures the api according to it because i am able to have and use different capabilities on the hypervisor than what is on the hypervisor | |
| 17:11:19 | samarendra | i have disk_cachemodes different on my compute node and controller and the compute node options take precedence when spinning up the vm | |
| 17:12:05 | mriedem | that's because disk_cachemodes is only read from the nova-compute service | |
| 17:12:23 | samarendra | mriedem: ok | |
| 17:12:39 | samarendra | so i can remove the option completely from controller ? | |
| 17:12:42 | mriedem | yes | |
| 17:12:50 | samarendra | ok. | |
| 17:13:40 | samarendra | mriedem: also on the same note, is there actually a detailed doc (apart from code :)) telling how disk_cachemodes can be configured? | |
| 17:13:49 | samarendra | last i saw in the doc it was a bit confusing | |
| 17:14:26 | mriedem | besides https://docs.openstack.org/developer/nova/sample_config.html you mean? | |
| 17:14:46 | samarendra | mriedem: yes | |
| 17:15:08 | samarendra | it talks of only file and block | |
| 17:15:17 | mriedem | there has been a multi-release effort underway to provide better help text to all of the config options, but unfortunately it looks like one hasn't been done yet | |
| 17:15:40 | samarendra | ok i just found another one on https://docs.openstack.org/developer/openstack-ansible/mitaka/install-guide/configure-nova.html | |
| 17:16:34 | samarendra | mriedem: ok. i will try and contribute something on that front from our experience | |
| 17:16:40 | mriedem | thanks | |
| 17:16:59 | mriedem | samarendra: this is the blueprint for the cleanup effort by the way https://blueprints.launchpad.net/nova/+spec/centralize-config-options-pike | |
| 17:18:14 | mriedem | samarendra: fyi https://review.openstack.org/#/c/18042/ | |
| 17:18:28 | samarendra | ok. i guess we using mitaka may mean not many useful things can be contributed. | |
| 17:18:40 | mriedem | sure it odes | |
| 17:18:41 | mriedem | *does | |
| 17:18:49 | mriedem | the option is basically unchanged | |
| 17:19:03 | mriedem | the option was introduced in grizzly | |
| 17:19:05 | mriedem | without docs | |
| 17:19:09 | samarendra | oh ok | |
| 17:20:30 | mriedem | from https://libvirt.org/formatdomain.html | |
| 17:20:35 | mriedem | "unsafe" since 0.9.7 | |
| 17:20:35 | mriedem | The optional cache attribute controls the cache mechanism, possible values are "default", "none", "writethrough", "writeback", "directsync" (like "writethrough", but it bypasses the host page cache) and "unsafe" (host may cache all disk io, and sync requests from guest are ignored). Since 0.6.0, "directsync" since 0.9.5, | |
| 17:21:48 | mriedem | sfinucan: is it possible to use choices with ListOpt? | |
| 17:22:14 | samarendra | mriedem: if by choices you mean x=y, a=b, then we are already using it :) | |
| 17:22:37 | mriedem | no sorry i meant something else | |
| 17:23:07 | samarendra | mriedem: ok. btw thanks for the links! i never thought of looking at libvirt docs | |
| 17:30:55 | mriedem | also https://www.ibm.com/support/knowledgecenter/linuxonibm/liaat/liaatbpkvmguestcache.htm | |
| 17:35:25 | openstackgerrit | Tadas Ustinavičius proposed openstack/nova master: Adding unit test for the patch https://review.openstack.org/445558 | |
| 17:36:23 | samarendra | mriedem: thanks. have seen that already :) | |
| 17:36:36 | mriedem | i'm shamelessly copying it into the config option help | |
| 17:43:51 | samarendra | mriedem: where can i find bugs for openstack nova? i was looking if anyone reported any issues with attaching and detaching volumes on ppc64le arch. or would that be in the domain of openstack-cinder? | |
| 17:44:26 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Expand help text for [libvirt]/disk_cachemodes https://review.openstack.org/454823 | |
| 17:44:26 | mriedem | samarendra: https://bugs.launchpad.net/nova/ | |
| 17:44:31 | mriedem | samarendra: also ^ | |
| 17:45:02 | samarendra | review.openstack.org would be where people submit patches for the bugs on launchpad, right? | |
| 17:45:08 | mriedem | yes | |
| 17:45:17 | samarendra | ok | |
| 17:45:28 | mriedem | see if https://review.openstack.org/#/c/454823/ looks ok | |
| 17:49:46 | samarendra | mriedem: +1 | |
| 18:01:01 | mriedem | no jaypipes huh | |
| 18:01:21 | mriedem | dansmith: re: https://review.openstack.org/#/c/407228/ do you know if we changed to always pre-load instances with 'tags' loaded by default now? | |
| 18:01:42 | mriedem | i'm not really sure why we care about sending instance tags on every instance action payload | |
| 18:01:55 | dansmith | mriedem: I have a -1 on the patch to do that | |
| 18:02:05 | dansmith | https://review.openstack.org/#/c/415298/15 | |
| 18:02:10 | dansmith | haven't really gotten an answer on it | |
| 18:02:39 | mriedem | hmm, so that comes AFTER https://review.openstack.org/#/c/407228/ | |
| 18:02:53 | mriedem | and ^ will lazy-load instance tags on every instance action notification | |
| 18:02:54 | dansmith | I imagine the desire to have it in notifications is so you can do something for all instances tagged a certain way, | |
| 18:02:56 | mriedem | which is a buttload | |
| 18:02:57 | dansmith | but it does seem expensive | |
| 18:03:00 | dansmith | mriedem: yeah I know | |
| 18:03:24 | mriedem | seems we should figure out the lazy-load thing before we start blasting that everywhere | |
| 18:03:32 | mriedem | it'd be one thing if it were instance create/update | |
| 18:03:43 | dansmith | maybe I'm being silly, but it seems like we should pre-load tags anywhere we need to avoid that patch getting any more lazy-loads, then merge that patch | |
| 18:03:55 | dansmith | without making it default | |
| 18:03:56 | dansmith | agree | |
| 18:04:34 | mriedem | yeah i agree with you | |
| 18:05:23 | dansmith | you gonna slap the sad trombone on that bottom one? | |
| 18:06:24 | mriedem | wa mfing wa | |
| 18:06:45 | mriedem | sneti: are you around? | |
| 18:07:42 | mriedem | i think we basically just send tags for instance create and update | |
| 18:07:47 | mriedem | and none of the action payloads | |
| 18:08:15 | mriedem | i believe the whole intent of this is to get instance tags into searchlight for filling in the compute rest api server show response | |
| 18:08:26 | mriedem | to do that, i don't need to blast out tags for suspend/resume actions | |
| 18:08:43 | dansmith | ah, yeah | |
| 18:09:13 | mriedem | https://blueprints.launchpad.net/nova/+spec/additional-notification-fields-for-searchlight | |
| 18:09:13 | mriedem | it even says in the bp | |
| 18:09:17 | mriedem | Step1: Add simple fields to instance.create and instance.update | |
| 18:10:56 | dansmith | sonofa bee | |
| 18:11:14 | mriedem | alright cool, next thing | |
| 18:11:22 | mriedem | so on your workaround patch, | |
| 18:11:37 | mriedem | i wasn't sure if we should call that option out in the track_instance_changes option help text too or not | |
| 18:11:42 | mriedem | or if we should mention it in a reno | |
| 18:11:55 | dansmith | mriedem: or both, yeah I was going to do that | |
| 18:12:03 | dansmith | I'm currently unfscking my discover patch you conflicted with :P | |
| 18:12:04 | mriedem | i mean, a reno might raise eyebrows, which maybe you want or don't want | |
| 18:12:15 | mriedem | given it's the default behavior maybe it's better to not ruffle feathers, idk | |
| 18:12:34 | dansmith | I dunno, I think it's a good highlight for getting claims in the scheduler | |
| 18:12:38 | mriedem | sure | |
| 18:12:46 | mriedem | if you do a reno, i'd just put it in the 'other' section | |
| 18:12:52 | mriedem | not sure it's really a feature or issue or anything | |
| 18:12:56 | dansmith | okay | |
| 18:12:56 | mriedem | maybe known issue | |
| 18:13:27 | mriedem | like, hey, if you plan on having a big deployment and you don't want upcalls from all computes to the api, set this thing, which is temporary until claims in the scheduler | |
| 18:13:35 | mriedem | which is why it's a known issue and we're adding the backdoor workarsound | |
| 18:14:09 | dansmith | ack | |
| 18:15:02 | mriedem | melwitt: the first 4 changes here have my +2 https://review.openstack.org/#/c/442162/ they are super fun i swear - after the first one it's pretty easy | |
| 18:15:07 | mriedem | the first one is mostly just mechanics | |
| 18:18:35 | jbernard | melwitt: not yet, i plan to look at that next | |
| 18:23:52 | mriedem | efried: the long awaited +2 https://review.openstack.org/#/c/438119/11 | |
| 18:24:01 | efried | Wohoo!!! | |