| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2021-05-06 | |||
| 10:15:29 | sean-k-mooney | by default socket.gethostname() will match the hostname returned by libvirt | |
| 10:15:29 | sean-k-mooney | by default socket.gethostname() will match the hostname returned by libvirt | |
| 10:15:45 | gibi | if the neutron agent sees a different hostname than the nova-compute then how can the nova hypervisor API help in that | |
| 10:15:45 | gibi | if the neutron agent sees a different hostname than the nova-compute then how can the nova hypervisor API help in that | |
| 10:16:46 | sean-k-mooney | gibi: we reqire the [DEFAULT]/host config option to be the same in both the nova and neutron config | |
| 10:16:46 | sean-k-mooney | gibi: we reqire the [DEFAULT]/host config option to be the same in both the nova and neutron config | |
| 10:16:59 | sean-k-mooney | so you can find the compute service assocatied with the host | |
| 10:16:59 | sean-k-mooney | so you can find the compute service assocatied with the host | |
| 10:17:14 | sean-k-mooney | then get its uuid | |
| 10:17:14 | sean-k-mooney | then get its uuid | |
| 10:17:35 | gibi | so the problematic case is when the hypervisor_hostname != hostname? But why that happen in case of libvirt? | |
| 10:17:35 | gibi | so the problematic case is when the hypervisor_hostname != hostname? But why that happen in case of libvirt? | |
| 10:17:58 | sean-k-mooney | because of https://github.com/libvirt/libvirt/blob/master/src/util/virutil.c#L454-L473 | |
| 10:17:59 | sean-k-mooney | because of https://github.com/libvirt/libvirt/blob/master/src/util/virutil.c#L454-L473 | |
| 10:18:52 | sean-k-mooney | if you have the fqdn as the canonical host name in /etc/hosts then libvirt will return an fqdn but socket.gethostname() will be the short hostname | |
| 10:18:52 | sean-k-mooney | if you have the fqdn as the canonical host name in /etc/hosts then libvirt will return an fqdn but socket.gethostname() will be the short hostname | |
| 10:19:44 | gibi | and don't we require CONF.host to be set to fqdn is enough in this case? | |
| 10:19:44 | gibi | and don't we require CONF.host to be set to fqdn is enough in this case? | |
| 10:20:08 | gibi | sorry my english is breaking up :0 | |
| 10:20:08 | gibi | sorry my english is breaking up :0 | |
| 10:20:29 | gibi | so if in this case CONF.host is set to fqdn then everything is in synch isn't it? | |
| 10:20:29 | gibi | so if in this case CONF.host is set to fqdn then everything is in synch isn't it? | |
| 10:20:34 | sean-k-mooney | nope ooo already hardcodes that to the fqdn | |
| 10:20:34 | sean-k-mooney | nope ooo already hardcodes that to the fqdn | |
| 10:21:22 | gibi | don't we use that fqdn as the RP name ? | |
| 10:21:22 | gibi | don't we use that fqdn as the RP name ? | |
| 10:21:33 | sean-k-mooney | no | |
| 10:21:39 | sean-k-mooney | well somethimes | |
| 10:21:39 | sean-k-mooney | well somethimes | |
| 10:21:45 | sean-k-mooney | we do not use the config value | |
| 10:21:45 | sean-k-mooney | we do not use the config value | |
| 10:21:51 | sean-k-mooney | so we use what libvirt returns | |
| 10:21:51 | sean-k-mooney | so we use what libvirt returns | |
| 10:22:03 | sean-k-mooney | socket.gethostname will never be the fqdn | |
| 10:22:03 | sean-k-mooney | socket.gethostname will never be the fqdn | |
| 10:22:54 | sean-k-mooney | libvirt will return the fqdn if you have "172.17.1.17 compute-0.redhat.local compute-0" and short name if you have "172.17.1.17 compute-0 compute-0.redhat.locaL" | |
| 10:22:54 | sean-k-mooney | libvirt will return the fqdn if you have "172.17.1.17 compute-0.redhat.local compute-0" and short name if you have "172.17.1.17 compute-0 compute-0.redhat.locaL" | |
| 10:23:45 | sean-k-mooney | gibi: one way to fix this is to try using the hostname and then the fqdn to lookup the rp | |
| 10:23:45 | sean-k-mooney | gibi: one way to fix this is to try using the hostname and then the fqdn to lookup the rp | |
| 10:24:05 | sean-k-mooney | gibi: we could also start using socket.gethostname in nova in the libvirt driver | |
| 10:24:05 | sean-k-mooney | gibi: we could also start using socket.gethostname in nova in the libvirt driver | |
| 10:24:15 | sean-k-mooney | for hypervior hostname | |
| 10:24:15 | sean-k-mooney | for hypervior hostname | |
| 10:27:01 | sean-k-mooney | gibi: we can look at fixing this in ooo | |
| 10:27:01 | sean-k-mooney | gibi: we can look at fixing this in ooo | |
| 10:27:29 | sean-k-mooney | by having it generate the name that libvirt will return and put it in the neutorn config or we can do https://bugzilla.redhat.com/show_bug.cgi?id=1957363 | |
| 10:27:30 | sean-k-mooney | by having it generate the name that libvirt will return and put it in the neutorn config or we can do https://bugzilla.redhat.com/show_bug.cgi?id=1957363 | |
| 10:27:31 | openstack | bugzilla.redhat.com bug 1957363 in tripleo-ansible "Canonical name should be a short name instead of FQDN" [High,New] - Assigned to rhos-maint | |
| 10:27:31 | openstack | bugzilla.redhat.com bug 1957363 in tripleo-ansible "Canonical name should be a short name instead of FQDN" [High,New] - Assigned to rhos-maint | |
| 10:28:42 | gibi | I see that the name returned from libvirt is highly host configuration dependent. But I don't think we have to automate all the possible combinations to work, if there is at least one set of configuration that make the system work for every libvirt case | |
| 10:28:42 | gibi | I see that the name returned from libvirt is highly host configuration dependent. But I don't think we have to automate all the possible combinations to work, if there is at least one set of configuration that make the system work for every libvirt case | |
| 10:29:41 | sean-k-mooney | well since this is not a problem with devstack as we see form the ci jobs it pretty clear that its actully caused by ooo's configuration of the host | |
| 10:29:41 | sean-k-mooney | well since this is not a problem with devstack as we see form the ci jobs it pretty clear that its actully caused by ooo's configuration of the host | |
| 10:29:59 | sean-k-mooney | gibi: the defualt we have in code do work out of the box just not as deployed by ooo | |
| 10:29:59 | sean-k-mooney | gibi: the defualt we have in code do work out of the box just not as deployed by ooo | |
| 10:30:06 | gibi | does tripleo configures the hostname on the compute host ? | |
| 10:30:06 | gibi | does tripleo configures the hostname on the compute host ? | |
| 10:30:24 | gibi | I mean the /etc/hosts and /etc/hostname | |
| 10:30:24 | gibi | I mean the /etc/hosts and /etc/hostname | |
| 10:30:28 | sean-k-mooney | yes | |
| 10:30:30 | sean-k-mooney | both | |
| 10:30:30 | sean-k-mooney | both | |
| 10:30:44 | sean-k-mooney | it templates out /etc/host with the fqdn first | |
| 10:30:44 | sean-k-mooney | it templates out /etc/host with the fqdn first | |
| 10:30:54 | sean-k-mooney | it alsso sets /etc/hostname | |
| 10:30:54 | sean-k-mooney | it alsso sets /etc/hostname | |
| 10:31:08 | sean-k-mooney | one way to "fix" it woudl be to put the fqdn in /etc/hostname | |
| 10:31:08 | sean-k-mooney | one way to "fix" it woudl be to put the fqdn in /etc/hostname | |
| 10:31:22 | sean-k-mooney | but that is kind of a hack | |
| 10:31:22 | sean-k-mooney | but that is kind of a hack | |
| 10:31:25 | gibi | so then tripleo can make actually configures (indirectly) what hypervisor hostname libvirt returns | |
| 10:31:25 | gibi | so then tripleo can make actually configures (indirectly) what hypervisor hostname libvirt returns | |
| 10:31:39 | gibi | s/can make// | |
| 10:31:39 | gibi | s/can make// | |
| 10:31:40 | sean-k-mooney | yep | |
| 10:31:57 | gibi | and also tripleo configures CONF.host for each services | |
| 10:31:57 | gibi | and also tripleo configures CONF.host for each services | |
| 10:32:05 | sean-k-mooney | yep it does | |
| 10:32:05 | sean-k-mooney | yep it does | |
| 10:32:10 | gibi | so then tripleo needs to make these configrations consistent | |
| 10:32:10 | gibi | so then tripleo needs to make these configrations consistent | |
| 10:32:17 | gibi | at least in my eyes | |
| 10:32:17 | gibi | at least in my eyes | |
| 10:32:19 | sean-k-mooney | yep idealy | |
| 10:32:20 | sean-k-mooney | yep idealy | |
| 10:32:45 | sean-k-mooney | but alos in a way that does not change hypervior hostname | |
| 10:32:45 | sean-k-mooney | but alos in a way that does not change hypervior hostname | |
| 10:32:54 | sean-k-mooney | on upragde | |
| 10:32:54 | sean-k-mooney | on upragde | |
| 10:32:59 | sean-k-mooney | because that would break nova | |
| 10:32:59 | sean-k-mooney | because that would break nova | |
| 10:33:53 | sean-k-mooney | i think the current problem is i think that chnage may have already happened which i might need to check | |
| 10:33:53 | sean-k-mooney | i think the current problem is i think that chnage may have already happened which i might need to check | |
| 10:33:54 | gibi | so the problem is that the current config is not fully consistent and blocks the bandwidth feature, but making the config consistent would result in breaking nova as the we don't tolerate hostname changes | |
| 10:33:54 | gibi | so the problem is that the current config is not fully consistent and blocks the bandwidth feature, but making the config consistent would result in breaking nova as the we don't tolerate hostname changes | |
| 10:35:06 | sean-k-mooney | well osp 13(queens) on rhel7 used the same config as as we used in osp 16(train) on rhel 8 | |
| 10:35:06 | sean-k-mooney | well osp 13(queens) on rhel7 used the same config as as we used in osp 16(train) on rhel 8 | |
| 10:35:24 | sean-k-mooney | so in queens we already had placement RPs but only nova used them | |
| 10:35:24 | sean-k-mooney | so in queens we already had placement RPs but only nova used them | |
| 10:36:00 | sean-k-mooney | i need to alk one of our downstream peopel what the hypervior hostname is on 13 and 16 and see if it has changed | |