| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2019-07-23 | |||
| 12:37:02 | openstackgerrit | Josephine Seifert proposed openstack/nova-specs master: Spec for the Nova part of Image Encryption https://review.opendev.org/608696 | |
| 12:55:13 | openstackgerrit | Merged openstack/python-novaclient master: Update api-ref location https://review.opendev.org/672135 | |
| 12:56:52 | efried | sean-k-mooney1: alex_xu: Care to re-ack https://review.opendev.org/#/c/651681/ please? | |
| 12:57:38 | efried | alex_xu: Please don't +W -- we need a non-Intel core to do that. (mriedem, bauzas, and johnthetubaguy have all had hands on this previously) | |
| 12:59:24 | efried | mriedem: Since gibi_off is _off for 2w, I'm going to propose the change to consts for https://review.opendev.org/#/c/668925/ as a fup so we can merge it, cool? | |
| 12:59:44 | efried | Unless you think it's okay to make the change there and fast approve? | |
| 13:01:46 | mriedem | efried: wfm | |
| 13:01:51 | mriedem | the fup i mean | |
| 13:03:11 | bauzas | efried: I can look | |
| 13:03:56 | bauzas | like I said, I already reviewed it | |
| 13:04:40 | cdent | I can never remember this: what, if anything, is the special magic to force a migration to a specific host/az? | |
| 13:05:13 | openstackgerrit | Eric Fried proposed openstack/nova master: Consts for need_healing https://review.opendev.org/672284 | |
| 13:05:16 | efried | mriedem: ^ | |
| 13:05:42 | efried | cdent: the old way or the new way? | |
| 13:06:02 | efried | API or CLI? | |
| 13:06:07 | jroll | when adding a new compute host, is there a way to bring it on disabled for scheduling, until we can go set up aggregates and such? | |
| 13:06:20 | cdent | efried: CLI, ocata-ish | |
| 13:06:32 | bauzas | cdent: live-migrate --host ? | |
| 13:06:50 | efried | jroll: for new code or old? | |
| 13:07:00 | jroll | efried: ocata and rocky :| | |
| 13:07:03 | bauzas | cdent: for a specific AZ, you need to first asking for an AZ when creating the instance | |
| 13:07:34 | bauzas | (or using the default_az option) | |
| 13:07:52 | cdent | bauzas: yeah, in this case the issue is that by some unknown series of events the instance is in an AZ that no longer exists, so forcing it out of that one is desired | |
| 13:07:59 | cdent | s/in/started in/ | |
| 13:08:46 | mriedem | https://docs.openstack.org/nova/latest/user/aggregates.html#implications-for-moving-servers | |
| 13:09:38 | mriedem | https://docs.openstack.org/python-openstackclient/latest/cli/command-objects/server.html#server-migrate | |
| 13:09:42 | mriedem | openstack server migrate --live | |
| 13:10:00 | mriedem | *openstack server migrate --live <host> will by default forcefully live migrate the server to the specified host and bypass the scheduler | |
| 13:10:01 | efried | jroll: mriedem was just in this code. I suspect with his new feature you could create the provider and tagg it with COMPUTE_STATUS_DISABLED before adding it to the service. But prior to that... I'm not sure. | |
| 13:10:38 | cdent | mriedem: thanks, i was in the first link (been there many times, it seems) | |
| 13:10:41 | mriedem | cdent: what i'm not sure about is if, on ocata, the forced live migration will update the az in the request spec (probably not) and it might not update the az on the instance record either | |
| 13:10:48 | mriedem | cdent: i wrote it for you :) | |
| 13:11:16 | mriedem | if the az is gone you're kind of in manual db surgery area, at least on ocata | |
| 13:11:28 | cdent | yeah, I figured that and already passed that along | |
| 13:11:35 | mriedem | there was a newer bug fix to not allow deleting an az that had instances in it, but that's not on ocata | |
| 13:11:48 | mriedem | there is also a blueprint in train to allow changing the az on unshelve | |
| 13:12:05 | mriedem | but that won't help you in ocata either | |
| 13:12:19 | jroll | efried: cool, thanks | |
| 13:13:14 | cdent | mriedem: and what about queens? | |
| 13:13:17 | cdent | any diff? | |
| 13:16:40 | efried | bauzas: Thanks. If it's acceptable, could you please +A? | |
| 13:16:57 | stephenfin | dansmith: You're going to hate this, but can we do this? http://paste.openstack.org/show/754754/ | |
| 13:17:07 | stephenfin | And I mean just that | |
| 13:17:09 | tssurya | jroll: for the old code if you have the ComputeFilter enabled it will automatically filter out disabled computes | |
| 13:17:19 | openstackgerrit | sahid proposed openstack/nova master: cellv2: fix update_cell to support cell0 https://review.opendev.org/672045 | |
| 13:17:42 | tssurya | as long as you don't have the "max_placement_results" set | |
| 13:18:02 | jroll | tssurya: right, but I want the computes to join the cluster disabled, to remove a race condition where they could get a VM scheduled before we can disable them or configure them properly | |
| 13:18:49 | tssurya | jroll: ah ok understood | |
| 13:19:16 | jroll | :) | |
| 13:19:28 | stephenfin | dansmith: I _think_ that with that change, we'll still be storing the same JSON-ified representation of the object in any database fields, only now we get the automatic conversion back to the object instead of having to remember to do it manually | |
| 13:21:14 | mriedem | cdent: the bug fix to prevent deleting an az with servers in it might have been backported to queens, i can check | |
| 13:21:34 | mriedem | but the unshelve to a new az thing is a new feature so wouldn't be backported (and isn't merged yet anyway) | |
| 13:22:10 | cdent | mriedem: thanks. the deleting the az bug isn't an immediate concern | |
| 13:22:43 | efried | bauzas: If you're in a spec approving mood, would you mind hitting https://review.opendev.org/#/c/608696/ (nova part of image encryption) as well please? | |
| 13:23:31 | mriedem | cdent: ok, looks like it only went back to rocky anyway https://review.opendev.org/#/q/Ic27195e46502067c87ee9c71a811a3ca3f610b73 | |
| 13:25:57 | bauzas | efried: opening another tab | |
| 13:26:03 | efried | melwitt, lyarwood, mdbooth_: Could we get a comeback on https://review.opendev.org/#/c/658903/ (RBD image up/download spec) please? | |
| 13:28:32 | openstackgerrit | Merged openstack/nova master: Update AZ admin doc to mention the new way to specify hosts https://review.opendev.org/666767 | |
| 13:29:44 | efried | stephenfin: gibi is +2 on https://review.opendev.org/#/c/641670/ (unaddressed port spec) -- looks like you're the man to send it? | |
| 13:30:05 | efried | jangutter: sean-k-mooney1: are you happy with ^ at this point? | |
| 13:31:33 | jangutter | efried: no objections from me. I'm pretty sure any security concerns can be sorted during implementation. | |
| 13:31:46 | stephenfin | efried: Yup, I'll get to that before EOD | |
| 13:31:49 | efried | donkey jangutter | |
| 13:31:56 | efried | thanks stephenfin | |
| 13:32:15 | jangutter | buy a donkey you too efried | |
| 13:35:51 | mriedem | cdent: just fyi https://review.opendev.org/#/q/I9f73c237923fdcbf4096edc5aedd2c968d4b893e was backported which updates the instance az during live migration but that still doesn't update the request spec az, which is the key thing since that's what will be used when you run the server through the scheduler to move it later (if not forced) | |
| 13:36:06 | mriedem | and if the request spec az is referencing something that's gone, and the az filter is enabled, the scheduler will novalidhost the move | |
| 13:36:06 | cdent | thanks mriedem | |
| 13:36:20 | openstackgerrit | Stephen Finucane proposed openstack/nova master: Remove 'nova.virt.driver.ComputeDriver.estimate_instance_overhead' https://review.opendev.org/672106 | |
| 13:36:40 | mriedem | so, recovery is likely: (1) force live migrate the server to some new host in a new az and then (2) manuall update the request spec az for the instance in the api db | |
| 13:36:49 | mriedem | note the request spec is a json blob in the db as well...so that's fun | |
| 13:37:25 | mriedem | i've got to run to an appointment, will be back on in a bit | |
| 13:45:46 | openstackgerrit | Merged openstack/nova-specs master: Add spec for expose-auto-converge-post-copy https://review.opendev.org/651681 | |
| 13:46:38 | openstackgerrit | Merged openstack/nova-specs master: Boot a VM with an unaddressed port https://review.opendev.org/641670 | |
| 13:48:03 | efried | stephenfin: If you agree https://review.opendev.org/#/c/672077/3/README.rst needs to be changed, I can fix it real quick and we can tag-team it? | |
| 13:48:28 | efried | uh, actually, I'm not sure what that should be changed to tbh | |
| 13:48:33 | stephenfin | efried: Drop it | |
| 13:48:38 | stephenfin | and yes | |
| 13:48:56 | dansmith | stephenfin: yeah you can't do that because it'll break rpc | |
| 13:49:29 | dansmith | stephenfin: but no, compute node won't auto serialize that to json, you have to make it do that | |
| 13:50:08 | stephenfin | Oh yeah. Duh | |
| 13:50:11 | stephenfin | Proxy field it is | |
| 13:50:29 | stephenfin | Also, what's horrible about bumping a major version again? | |
| 13:50:36 | stephenfin | *major object version | |
| 13:50:51 | stephenfin | I recall we bumped the Instance object and said never again, but I don't know why never again | |
| 13:51:20 | dansmith | stephenfin: it's not too terrible, but it is a little complicated. instance was the most complicated object, but I don't think we said never again, | |
| 13:51:28 | dansmith | it's just that it needs to be worth it | |
| 13:51:59 | efried | (stephenfin nm, there's no replacement for the landing page yet, I +Wed) | |
| 13:52:21 | dansmith | stephenfin: compute node would probably be much less annoying because of how limited its use is.. i.e. we don't send compute node objects all around to every node like we do instance, | |
| 13:52:33 | dansmith | it's mostly just the compute and conductor that have to touch them | |
| 13:52:56 | stephenfin | Yeah, that's the one I'm most interested in | |
| 13:53:16 | stephenfin | I think I'll wait til os-hypervisors is talking to placement instead of consuming the xxx_usage fields from ComputeNode first though | |
| 13:53:22 | stephenfin | So we can remove those while we're at it | |
| 13:53:35 | stephenfin | (Assuming that ends up being possible) | |
| 13:53:44 | dansmith | the point being to objectify that field, right? why is that a thing now vs. before? Because you're adding stuff to that field or something? | |
| 13:53:56 | dansmith | stephenfin: that sounds like a good idea | |
| 13:54:24 | dansmith | an object change for one field type is less good, but deprecating unused fields with that change is more meaty | |
| 13:55:04 | stephenfin | dansmith: Mostly because it wasn't initially obvious and led to me being bitten by it | |
| 13:55:23 | stephenfin | I am tackling it at in a more piecemeal fashion with https://review.opendev.org/#/c/671799/3 and the two preceding patches though | |
| 13:55:48 | stephenfin | moving the calls to convert the string to an object (and vice versa) close to where the field is actually read/set | |