| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2019-11-07 | |||
| 19:20:54 | efried | they make more sense read the other way around. | |
| 19:21:40 | efried | I thought there was an option to make it delete the volume when the instance goes away. But I guess that's still just in novaclient, not in osc yet? | |
| 19:22:02 | sean-k-mooney | efried: you can set it in the block device mappings | |
| 19:22:08 | mriedem | you can do it in osc if using the --block-device-mapping option, | |
| 19:22:14 | sean-k-mooney | but i dont know if you can do that with osc | |
| 19:22:17 | mriedem | which used to not allow passing id as a source image, | |
| 19:22:21 | sean-k-mooney | oh ok | |
| 19:22:28 | mriedem | but when i added --boot-from-volume i think i fixed that as well | |
| 19:22:46 | mriedem | yeah so i just need to change to use --block-device-mapping | |
| 19:23:09 | mriedem | updating | |
| 19:23:10 | efried | mmkay | |
| 19:24:52 | efried | those docs for --block-device-mapping could stand to be formatted a bit better. Now that we're generating them, you only have to update one place in osc... | |
| 19:24:53 | efried | :P | |
| 19:26:29 | mriedem | god now i have to look at the osc source code to see what it passes for boot_index | |
| 19:26:46 | mriedem | because you can't specify a boot index and if it doesn't pass one it defaults to None (not root) | |
| 19:27:27 | efried | so add that to the docs while you're in there reformatting | |
| 19:27:33 | efried | cause I know you were already doing that | |
| 19:28:02 | mriedem | i'm about to abandon this gd turd | |
| 19:29:45 | artom | *snerk* | |
| 19:29:59 | artom | I mean, the novaclient isn't getting EOL'ed anytime soon, what's wrong with keeping it? | |
| 19:31:04 | mriedem | ok so if you use --block-device-mapping you have to specify one of --image|--image-property|--volume and if you use --image, that's an ephemeral root disk and the bdm is just a non-root volume attachment | |
| 19:32:02 | mriedem | artom: i try as much as possible to use osc for new things so we can stop using nova cli | |
| 19:32:04 | efried | so much for "I can just *" | |
| 19:32:22 | sean-k-mooney | artom: the main blockers to killing the nova client is we dont have a good way to update osc and lack of people working on closing the gaps | |
| 19:32:32 | mriedem | --boot-from-volume preserves the volume by default b/c that's the default in the api for bdm.delete_on_termination (false) | |
| 19:32:50 | artom | sean-k-mooney, well, osc is a moving target as well | |
| 19:33:04 | artom | Now it's moving to openstacksdk, right? So we should be updating that as well? | |
| 19:33:11 | openstackgerrit | Dustin Cowles proposed openstack/nova-specs master: Spec: Provider config YAML file https://review.opendev.org/693414 | |
| 19:33:30 | artom | And it was/is missing basic microversion support | |
| 19:33:38 | sean-k-mooney | artom: the intent is we woudl jsut update the sdk for the api changes and then have a thin wrapper in osc | |
| 19:33:55 | sean-k-mooney | but yes its moving slowly to the sdk | |
| 19:34:30 | artom | mriedem, I agree with the sentiment :) It often comes up downstream - clients or whoever will say "I can't do this with osc [which is the recommended client for interaction with our OSP product]" and we'll come back with "use novaclient instead" | |
| 19:35:37 | sean-k-mooney | artom: we only do that if there is no other choice | |
| 19:35:39 | sean-k-mooney | but yes | |
| 19:35:47 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Remove the TODO about using OSC for BFV in test_evacuate.sh https://review.opendev.org/693406 | |
| 19:36:01 | efried | hahaha | |
| 19:36:04 | sean-k-mooney | although we really should file a bug downstream to fix the osc gap everytime we do that | |
| 19:37:04 | efried | wasn't there a spec to allow you to flip the 'remove on instance deletion' flag after the instance was created? | |
| 19:37:19 | mriedem | yes | |
| 19:37:30 | sean-k-mooney | efried: i think they wanted to set it as part of the attachment or somthing | |
| 19:37:32 | mriedem | that would be a pretty crappy workaround here | |
| 19:37:43 | efried | https://review.opendev.org/#/c/580336/ | |
| 19:37:55 | artom | *snerk* | |
| 19:37:58 | efried | mriedem: but eventually we'll want osc support for setting that option initially, right? | |
| 19:38:25 | mriedem | i just loudly audibly sighed | |
| 19:38:27 | efried | so like, instead of removing the TODO, you should *expand* it to explain all of that. | |
| 19:39:10 | mriedem | making the --boot-from-volume optional more customizable like the --block-device-mapping option is not what we want, it defeats the purpose of a simple boot from volume option, | |
| 19:39:18 | efried | right | |
| 19:39:22 | mriedem | so that likely means adding bootindex= to --block-device-mapping | |
| 19:39:36 | sean-k-mooney | honestly the default shoudl proably be delete on instnace terminate | |
| 19:39:40 | mriedem | but that has implications for how the command currently requires you to specify --image|--image-property|--volume | |
| 19:39:45 | efried | you need a new option, --remove-volume-on-instance-deletion, which is only allowed with --boot-from-volume | |
| 19:40:01 | mriedem | i don't think that's much better ux | |
| 19:40:07 | efried | I'm not serious. | |
| 19:40:13 | mriedem | heh, ok | |
| 19:40:35 | efried | I know it's tough to tell with my rbf and deadpan humor style. | |
| 19:40:41 | artom | --boot-from-volume size=1,cleanup=plz | |
| 19:40:50 | mriedem | sean-k-mooney: if you have a dumb ass old app which stores data in the root volume, then deleting that by default is likely not what you want | |
| 19:40:57 | mriedem | artom: f and no | |
| 19:41:00 | artom | Conversly, --boot-from-volume size=1,cleanup=nowai | |
| 19:41:00 | mriedem | see what i just said above | |
| 19:41:04 | mriedem | no | |
| 19:41:05 | mriedem | no | |
| 19:41:11 | artom | Hahaha | |
| 19:41:13 | mriedem | that's what --block-device-mapping is for | |
| 19:41:14 | sean-k-mooney | mriedem: proably but we have had customer complain about this before | |
| 19:41:27 | sean-k-mooney | mainly because they expected it to delete by default | |
| 19:41:34 | mriedem | well they were wrong | |
| 19:41:37 | efried | OS_DELETE_VOLUME_WITH_INSTANCE=true openstack server create .... | |
| 19:41:37 | mriedem | bug report closed | |
| 19:41:50 | sean-k-mooney | yep and they acepted that | |
| 19:41:59 | efried | (the env var solution is just for dtroyer) | |
| 19:42:00 | mriedem | "please delete my data by default thanks!" | |
| 19:42:09 | artom | efried, we need to find a way to use /proc and or /sys for this | |
| 19:42:14 | efried | lol | |
| 19:42:26 | mriedem | or, | |
| 19:42:35 | mriedem | read in yaml support for the osc commands like kubectrl | |
| 19:42:35 | efried | I'm sure we could develop a hardware module | |
| 19:42:39 | mriedem | *kubectl | |
| 19:42:47 | artom | efried, oh oh oh! Cyborg with FPGAs | |
| 19:42:50 | mriedem | then you could define your block_device_mapping_v2 in yaml! | |
| 19:42:53 | efried | you mean a yaml front end to the REST API? | |
| 19:43:03 | mriedem | efried: just osc to start | |
| 19:43:11 | mriedem | i'm kidding | |
| 19:43:12 | mriedem | again | |
| 19:43:21 | sean-k-mooney | have we intoduced them to this thing call curl | |
| 19:43:39 | artom | Ya man, it's what I do in the squat rack every Friday | |
| 19:43:42 | sean-k-mooney | they can pass a json payload and be as declaritvie as they like | |
| 19:43:51 | eandersson | btw the Rescheduling Exceptions we saw were most likely due to missing records in placement. | |
| 19:44:08 | mriedem | hear me out, bring back xml support in the rest api | |
| 19:44:12 | mriedem | who's with me?! | |
| 19:44:13 | eandersson | We just didn't know until we deployed https://github.com/openstack/nova/commit/a5269012a3b442a9e4055a7d523faff45f105f2b | |
| 19:44:32 | sean-k-mooney | eandersson: missing compute node resouce providers? | |
| 19:45:01 | eandersson | Since we did a major jump (Mitaka -> Rocky) something went wrong when we first ran heal_allocations | |
| 19:45:05 | eandersson | So a lot of VMs were missing | |
| 19:45:24 | sean-k-mooney | oh ok | |
| 19:45:37 | sean-k-mooney | so it was missing the allocations not the RPs | |
| 19:45:41 | eandersson | Yea | |
| 19:45:51 | eandersson | We ran it, but not sure why we didn't notice that it failed. | |
| 19:45:56 | eandersson | --dry-run will help A LOT with this | |