| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2020-12-03 | |||
| 17:05:12 | bauzas | the first case seems weirdo to me | |
| 17:05:16 | gibi | hm, test-server.1 and test-server.2 becomes test-server as a hostname in the geuest | |
| 17:05:20 | sean-k-mooney | so as an exampel we are proposign if servername=example.fqdn then instance.hostname==example and instance.display_name==example.fqdn | |
| 17:05:22 | bauzas | the second case is indeed somehow hard to convince | |
| 17:05:39 | bauzas | gibi: and I'm cool with this | |
| 17:05:52 | bauzas | (to have duplicate hostnames) | |
| 17:05:53 | stephenfin | sean-k-mooney: yes, and the alternative is instance.hostname=example-fqdn | |
| 17:06:05 | sean-k-mooney | yes | |
| 17:06:31 | sean-k-mooney | so if we maintain dispaly name that means that nova show example.fqdn will work in both cases | |
| 17:06:36 | sean-k-mooney | so ti wont affect server show | |
| 17:06:41 | sean-k-mooney | or list | |
| 17:06:54 | gibi | yeah, I agree not to affect server show and list | |
| 17:06:55 | stephenfin | Yes, it'll only affect people relying on cloud-init, which you're saying is already broken for FQDN? | |
| 17:07:00 | bauzas | just being clear, don't touch display_name ! | |
| 17:07:08 | stephenfin | *cloud-init for DNS integration | |
| 17:07:09 | sean-k-mooney | so it will only effect what is in the metadata and passed to designate | |
| 17:07:27 | bauzas | hangouts, folks ? | |
| 17:07:48 | stephenfin | isn't a bit late for you Europeans? | |
| 17:07:49 | bauzas | actually, my wife will kill me | |
| 17:07:52 | stephenfin | yeah :) | |
| 17:07:56 | gibi | mine too | |
| 17:07:58 | sean-k-mooney | yes with designate you get test-dns.invalid.dns.cloud.seanmooney.info for a servername=test-dns.invalid.dns via dns | |
| 17:08:01 | bauzas | stephenfin: dude, you're part of Europe | |
| 17:08:11 | stephenfin | *continental Europeans | |
| 17:08:12 | stephenfin | ;) | |
| 17:08:17 | bauzas | on a small isolated island tho | |
| 17:08:18 | gibi | :D | |
| 17:08:30 | sean-k-mooney | bauzas: its the brit in him showing :P | |
| 17:08:47 | bauzas | well, I actually feel isolated too | |
| 17:08:49 | gibi | push this two option ack to the ML: a) truncate b) replace with - | |
| 17:08:50 | bauzas | anyway | |
| 17:08:53 | bauzas | gibi: +1 | |
| 17:08:59 | stephenfin | gibi: good call; me does that | |
| 17:09:06 | gibi | sleep on it | |
| 17:09:11 | sean-k-mooney | cool that work for me too | |
| 17:09:14 | gibi | then try to close this out next week | |
| 17:09:21 | gibi | when stephenfin is back | |
| 17:09:22 | bauzas | stephenfin: clarify there are only two options and that we're about to backport the chosen | |
| 17:09:37 | gibi | ^^ +1 | |
| 17:10:02 | sean-k-mooney | do we need a workaround flag for this | |
| 17:10:05 | sean-k-mooney | for the backports | |
| 17:10:05 | artom | Random idea | |
| 17:10:13 | bauzas | (but again, from an ops pov (well from the last decade), I just feel that naming your server with 'x.y' is either very smart or totally crazy | |
| 17:10:18 | artom | You can set the port's dns_name "manually" via the Neutron API | |
| 17:10:23 | bauzas | sean-k-mooney: NO | |
| 17:10:34 | artom | So... why not just stop doing it automagically when Nova creates/updates the port? | |
| 17:10:37 | sean-k-mooney | bauzas: ok | |
| 17:10:50 | artom | And let people pick their own hostname that way | |
| 17:10:53 | sean-k-mooney | artom: only with designate | |
| 17:10:59 | bauzas | yeah | |
| 17:11:07 | sean-k-mooney | well or a vey new extenion i think | |
| 17:11:07 | artom | sean-k-mooney, not based on https://docs.openstack.org/api-ref/network/v2/index.html?expanded=update-port-detail,show-port-details-detail#update-port | |
| 17:11:09 | sean-k-mooney | ill check | |
| 17:11:21 | bauzas | and that's exactly why I think that nova shouldn't deal with TLDs in their server hostnames | |
| 17:11:35 | artom | Like, the original BZ we have is because Neutron exploded | |
| 17:11:38 | artom | Not Designate | |
| 17:11:53 | bauzas | but I can recognize that the dash conversion seems kinda more acceptable to users | |
| 17:12:48 | bauzas | anyway, /me needs to bail out | |
| 17:12:51 | bauzas | ++ | |
| 17:13:07 | gibi | o/ | |
| 17:13:19 | stephenfin | o/ | |
| 17:13:46 | sean-k-mooney | artom: im pretty sure nova used to explode at one point too with some of the shared nova net code | |
| 17:14:02 | sean-k-mooney | artom: the orginal bz was in my dev cloud in intel | |
| 17:14:18 | artom | sean-k-mooney, bz or launchpad bug? | |
| 17:14:20 | sean-k-mooney | it was filed by someone on my team after i told them it was never supported | |
| 17:14:27 | sean-k-mooney | sorry the launch pad bug | |
| 17:14:44 | artom | Well, nova-network doesn't exist anymore, so... :) | |
| 17:16:09 | sean-k-mooney | i think it used to explode somewhere in https://github.com/openstack/nova/blob/stable/train/nova/network/minidns.py | |
| 17:17:27 | sean-k-mooney | this si where we are current failing right https://github.com/openstack/nova/blob/stable/train/nova/network/neutronv2/api.py#L678-L686 | |
| 17:20:03 | artom | sean-k-mooney, no, based on https://bugzilla.redhat.com/show_bug.cgi?id=1872314 we actually make the request to Neutron | |
| 17:20:03 | openstack | bugzilla.redhat.com bug 1872314 in openstack-nova "[OSP 16.1] Can't launch instance if name ends with a number" [Medium,New] - Assigned to nova-maint | |
| 17:20:14 | artom | And bubble up the BadRequest it returns | |
| 17:20:36 | sean-k-mooney | artom: oh i know | |
| 17:21:03 | sean-k-mooney | but back in like havana i think it broken in nova it also broke in neutorn | |
| 17:21:16 | sean-k-mooney | at somepoint neutron started allowing the fqdn | |
| 17:21:35 | sean-k-mooney | but it was never done intentually at least form a nova point of view | |
| 17:22:35 | artom | So https://github.com/openstack/nova/blob/stable/train/nova/network/neutronv2/api.py#L1138 would be where our request to Neutron explodes | |
| 17:23:02 | artom | And https://github.com/openstack/nova/blob/stable/train/nova/network/neutronv2/api.py#L1494 is where we populate the dns_name | |
| 17:23:18 | artom | I'm saying don't actually do that | |
| 17:23:40 | artom | And let the users do a port update manually with the hostname of their chosing, not one that Nova has derived from the VM name | |
| 17:25:05 | artom | And *then* if Neutron isn't happy, it really is user error and WONTFIX :) | |
| 17:27:57 | sean-k-mooney | artom: that wont fix the fact the contend ot the metadata api is not what you expect | |
| 17:28:12 | artom | sean-k-mooney, eh? | |
| 17:28:32 | sean-k-mooney | look at http://lists.openstack.org/pipermail/openstack-discuss/2020-November/019137.html | |
| 17:29:05 | sean-k-mooney | with a default designate domain of cloud.seanmooney.info adn default dhcp_domain of novalocal | |
| 17:29:22 | sean-k-mooney | fi you create a server called test-dns.invalid.dns | |
| 17:30:49 | sean-k-mooney | you will have "test-dns.invalid.dns.novalocal" "test-dns.invalid.dns.cloud.seanmooney.info" "test-dns.cloud.seanmooney.info" and test-dns for hostnames/fqdns | |
| 17:31:14 | sean-k-mooney | but you will never have test-dns.invalid.dns which was the fqdn you uses as the servername | |
| 17:31:17 | artom | sean-k-mooney, I feel like we're conflating many problems into 1 | |
| 17:31:38 | sean-k-mooney | not really | |
| 17:31:58 | sean-k-mooney | if we allow FQDNs we shoudl support them consitently | |
| 17:32:17 | sean-k-mooney | right now you will get 4 possible name for the server none of them will be the one you passed in unless | |
| 17:32:59 | sean-k-mooney | you explcity configure [DEFAULT]/dhcp_domain="" and do not use a default designate domain | |
| 17:33:55 | artom | sean-k-mooney, right, this falls in the wider "our DNS interoperation is crap" problem | |
| 17:34:02 | artom | On which my grip is not great | |
| 17:34:21 | artom | But feels like fixing that would involve a new microversion with a separate hostname and/or fqdn parameter | |
| 17:35:18 | artom | At the tightest possible scope, this bug is caused by Nova deriving a `dns_name` from the display name, and passing it to Neutron, which can then refuse it, despite the display name being perfectly valid | |
| 17:36:03 | sean-k-mooney | if you want real fun read https://review.opendev.org/c/openstack/neutron-specs/+/88623/16/specs/liberty/internal-dns-resolution.rst | |
| 17:36:19 | artom | If I want real fun I'll go do some drugs | |
| 17:36:50 | sean-k-mooney | https://bugs.launchpad.net/neutron/+bug/1459030 | |