| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2018-03-08 | |||
| 18:10:06 | mriedem | iow you can run nova-scheduler in a separate host, vm, venv, container whatever | |
| 18:10:24 | stephenfin | Coolness | |
| 18:10:32 | mriedem | just not a function in lambda yet | |
| 18:10:45 | mriedem | but intel is working on it | |
| 18:11:03 | stephenfin | (I'd defined a host-specific option on a nova-compute node and was wondering why the NUMATopologyFilter couldn't see it) | |
| 18:11:23 | mriedem | you'd have to pump that information up to the scheduler via the compute node | |
| 18:11:33 | mriedem | the HostState object in the scheduler is really a wrapper over a compute node | |
| 18:11:45 | mriedem | and the compute node record has all sorts of hook points for crap, like the stats dict | |
| 18:12:03 | mriedem | or, via traits or something in placement | |
| 18:12:03 | stephenfin | Yup. I'm targeting the HostState object now | |
| 18:12:08 | mriedem | if you want the new hotness | |
| 18:12:22 | stephenfin | jaypipes is going to _hate_ v1 of this NUMA-aware vSwitch code | |
| 18:12:26 | mriedem | but since we don't model numa nodes yet... | |
| 18:12:44 | stephenfin | Nor network-NUMA affinity (from neutron) | |
| 18:12:44 | mriedem | s/jaypipes/everyone/ | |
| 18:12:51 | stephenfin | Hence that static config option | |
| 18:12:52 | stephenfin | :) | |
| 18:13:19 | tssurya | dansmith, mriedem : fix for the regression I caused - https://review.openstack.org/#/c/550967/ , hoping this was what was expected... | |
| 18:14:02 | dansmith | tssurya: I was just commenting on that | |
| 18:14:26 | tssurya | dansmith: ack | |
| 18:15:34 | dansmith | tssurya: in short, it is mostly what I expected, but the last couple days of working on purge and archive --all-cells has me thinking of a few other things | |
| 18:16:15 | dansmith | oh I see you also have WIP patches up for the cell disabling stuff, cool | |
| 18:16:31 | tssurya | dansmith: I totally agree that the mappings and specs can get orphaned | |
| 18:18:08 | tssurya | dansmith: yea just WIPs for now since the spec is not merged yet | |
| 18:18:14 | dansmith | awesome | |
| 18:18:22 | tssurya | and thanks for you review and time! | |
| 18:18:32 | dansmith | I +2d that spec earlier today | |
| 18:19:33 | tssurya | Yes , thanks for that | |
| 18:19:42 | tssurya | I am guessing a second core also needs to review right ? | |
| 18:19:56 | mriedem | dansmith: did you want this in ocata? https://review.openstack.org/#/c/542767/ | |
| 18:20:01 | mriedem | it's a rhetorical questoin | |
| 18:20:08 | mriedem | tssurya: yes i've got it starred | |
| 18:20:34 | tssurya | mriedem : thank you | |
| 18:21:31 | dansmith | tssurya: fair warning: mriedem will try to find some spelling mistake in that spec to lodge a -1 against my +2 | |
| 18:21:39 | dansmith | tssurya: but that's about me, not you, so don't take it personally :) | |
| 18:21:54 | mriedem | not for spelling, | |
| 18:21:59 | mriedem | but likely for something about tests for docs | |
| 18:22:14 | dansmith | tests for docs? | |
| 18:22:34 | tssurya | now I am confused | |
| 18:22:36 | mriedem | *or | |
| 18:22:39 | mriedem | heh | |
| 18:22:41 | dansmith | mriedem: oohhh :) | |
| 18:23:53 | tssurya | mriedem : I will wait for it then | |
| 18:42:38 | dansmith | I believe I have voted on all non-WIP tssurya patches as of this moment | |
| 18:43:34 | dansmith | two cookies! | |
| 18:45:49 | mriedem | what about your love for pizza too? | |
| 18:46:07 | dansmith | there it is. | |
| 18:46:08 | mriedem | and proper. punctuation. | |
| 18:54:58 | cfriesen | anyone aware of issues with the metadata service in devstack? does it need to be explicitly enabled? | |
| 19:48:29 | mriedem | cfriesen: add n-api-meta to ENABLED_SERVICES | |
| 19:48:38 | mriedem | although i think you can also just run it under n-api | |
| 19:48:47 | mriedem | but not sure | |
| 20:13:41 | mriedem | dansmith: you might want to reply on this thread http://lists.openstack.org/pipermail/openstack-operators/2018-March/014930.html | |
| 20:13:45 | mriedem | we talked about that last week | |
| 20:14:01 | mriedem | public cloud WG people wanted to resolve that issue too | |
| 20:14:44 | dansmith | oh all right | |
| 20:14:55 | mriedem | remember, empathy | |
| 20:15:00 | mriedem | or is that only for jaypipes? | |
| 20:18:47 | dansmith | definitely not for me :) | |
| 20:27:52 | openstackgerrit | Merged openstack/nova master: Add --purge helper flag to archive_deleted_rows https://review.openstack.org/550182 | |
| 21:18:58 | mriedem | i found a typo in the disable cell spec | |
| 21:30:06 | mriedem | comments inline | |
| 21:30:48 | tssurya | mriedem: thank you | |
| 21:31:13 | sorrison | Is there anyone here able to discuss a nova security bug with me? | |
| 21:43:06 | harlowja | qq for folks, the nova-api component, should that have access to all the rabbits of all the cells (when in cells v2), or does it just need access to the top level cells rabbit? | |
| 21:44:11 | harlowja | this is in regard to console log getting when using cells; it seems like nova-api may be trying to send a message to a cell rabbit | |
| 21:46:29 | harlowja | ^ if that matters | |
| 21:52:27 | harlowja | mriedem ^ any idea :) | |
| 21:54:04 | mriedem | harlowja: cells v2 cells are not nested | |
| 21:54:08 | mriedem | there is no parent/child | |
| 21:54:17 | mriedem | the api has access to all of the cells via mappings in the nova_api db | |
| 21:54:31 | harlowja | so nova.conf needs to have all rabbit transports? | |
| 21:54:34 | harlowja | in all places? | |
| 21:54:35 | mriedem | no | |
| 21:54:46 | mriedem | nova.conf needs to be able to access the api database | |
| 21:54:58 | mriedem | the api database has the cell mappings which also has their rabbit transports | |
| 21:55:27 | harlowja | ahhhhh | |
| 21:55:30 | mriedem | when getting an instance to perform some action on it, like getting it's console log | |
| 21:55:39 | mriedem | the api looks up a mapping to see which cell the instance is in, | |
| 21:55:50 | mriedem | then targets a context to that cell so all rpc/db for that instance goes to the targeted cell | |
| 21:56:43 | harlowja | ah | |
| 21:56:43 | harlowja | so nova-api conf doesn't really even need any transport_urls | |
| 21:57:16 | harlowja | (in configuration that is) | |
| 21:57:20 | mriedem | it does to be able to rpc cast to conductor | |
| 21:57:25 | harlowja | kk | |
| 21:57:52 | mriedem | https://docs.openstack.org/nova/latest/user/cellsv2-layout.html | |
| 22:00:39 | harlowja | ya, been looking at that, may have figured it out, ha | |
| 22:08:36 | mriedem | dansmith: want to re-approve melwitt's spec for continuing the 'convert consoles to objects' work? https://review.openstack.org/#/c/543662/ | |
| 22:08:44 | mriedem | she updated it a bit after stuff that changed in queens | |
| 22:39:01 | harlowja | mriedem hey, for the https://docs.openstack.org/nova/latest/user/cellsv2-layout.html diagram, in the child cells, just want to confirm, they don't need there own nova-scheduler right? | |
| 22:39:14 | harlowja | we are seeing something where it seems to go to a cell, then go to the scheduler | |
| 22:40:34 | mriedem | harlowja: there are no child cells | |
| 22:40:38 | mriedem | there is only 1 scheduler | |
| 22:40:40 | mriedem | this isn't cells v1 | |
| 22:40:46 | mriedem | you crazy godaddy guys | |
| 22:40:47 | harlowja | sorry, cell 1 :-P | |
| 22:40:55 | harlowja | we are doing this in not cells v1 :-P | |
| 22:40:59 | harlowja | cells v2, ha | |
| 22:41:11 | mriedem | nova-api and nova-scheduler are global | |
| 22:41:27 | mriedem | and depending on how you want to handle conductor, that can be global as well | |