Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2021-05-06
10:44:48 gibi so again, I think we first agree on what config fqdn/sortname combination OpenStack wants to supprot
10:44:57 gibi then fix the system to work in these scenarios
10:44:57 gibi then fix the system to work in these scenarios
10:45:48 sean-k-mooney ya
10:45:48 sean-k-mooney ya
10:47:25 sean-k-mooney ya personally i would prefer to either onlyu use hostname or only use FQDNs
10:47:25 sean-k-mooney ya personally i would prefer to either onlyu use hostname or only use FQDNs
10:48:03 gibi yepp, that would be clean and self consistent all the way
10:48:03 gibi yepp, that would be clean and self consistent all the way
10:48:05 sean-k-mooney i prefer hostnames but that is just a preference
10:48:05 sean-k-mooney i prefer hostnames but that is just a preference
10:48:52 sean-k-mooney the delta in function ality is fqdns allows host-1.rack-1.mydomain and host-1.rack-2.mydomain to work
10:48:52 sean-k-mooney the delta in function ality is fqdns allows host-1.rack-1.mydomain and host-1.rack-2.mydomain to work
10:49:13 sean-k-mooney where as hostnames only requires tehm all to be unique
10:49:13 sean-k-mooney where as hostnames only requires tehm all to be unique
10:49:32 gibi yepp, I see the need for FQDNs in big deployments
10:49:32 gibi yepp, I see the need for FQDNs in big deployments
10:49:42 gibi or at least accept the need
10:49:42 gibi or at least accept the need
10:50:05 sean-k-mooney the only issue if we use FQDNs only is sometime we will still end up with hostnames
10:50:05 sean-k-mooney the only issue if we use FQDNs only is sometime we will still end up with hostnames
10:50:25 sean-k-mooney i.e. if you dont have the FQDN in /etc/hosts
10:50:25 sean-k-mooney i.e. if you dont have the FQDN in /etc/hosts
10:51:03 sean-k-mooney well that and we told all the other service to use socket.gethostname()
10:51:03 sean-k-mooney well that and we told all the other service to use socket.gethostname()
10:51:46 sean-k-mooney i even changed cyborg to use gethostnaem form getfqdn https://github.com/openstack/cyborg/commit/ec0eab0ffb4fc18574dbd937f29545832d317e4c
10:51:46 sean-k-mooney i even changed cyborg to use gethostnaem form getfqdn https://github.com/openstack/cyborg/commit/ec0eab0ffb4fc18574dbd937f29545832d317e4c
10:51:53 sean-k-mooney after that intial email thread
10:51:53 sean-k-mooney after that intial email thread
10:53:08 sean-k-mooney gibi: part of me feels like we should add a fucntion to oslo.utils
10:53:08 sean-k-mooney gibi: part of me feels like we should add a fucntion to oslo.utils
10:53:14 sean-k-mooney and then just have everyone use that
10:53:15 sean-k-mooney and then just have everyone use that
10:53:50 sean-k-mooney and then in nova we can perhaps think about have a reshape or soemthing to normalise the RPs
10:53:50 sean-k-mooney and then in nova we can perhaps think about have a reshape or soemthing to normalise the RPs
10:54:15 gibi I think you can rename without reshape
10:54:15 sean-k-mooney the fact this currently depend on the virt driver you use is less then ideal.
10:54:15 gibi I think you can rename without reshape
10:54:16 sean-k-mooney the fact this currently depend on the virt driver you use is less then ideal.
10:54:26 sean-k-mooney gibi: ya i think you are right
10:54:26 sean-k-mooney gibi: ya i think you are right
10:54:31 sean-k-mooney the allocation dont need to move
10:54:31 sean-k-mooney the allocation dont need to move
10:54:35 gibi yepp
10:55:14 sean-k-mooney what we can do its encode the libvirt logic which is in the neutron patch in an oslo patch
10:55:14 sean-k-mooney what we can do its encode the libvirt logic which is in the neutron patch in an oslo patch
10:55:20 sean-k-mooney and then use that
10:55:20 sean-k-mooney and then use that
10:55:42 sean-k-mooney for libvirt at least it woudl not change behaivor and for other driver i suspect it wont either
10:55:42 sean-k-mooney for libvirt at least it woudl not change behaivor and for other driver i suspect it wont either
10:55:56 gibi I cannot comment on other drivers cases
10:55:56 gibi I cannot comment on other drivers cases
10:56:11 sean-k-mooney well ironic they use the ironic node uuid as the name right?
10:56:11 sean-k-mooney well ironic they use the ironic node uuid as the name right?
10:56:32 sean-k-mooney not sure what hyperv or vmware does
10:56:32 sean-k-mooney not sure what hyperv or vmware does
10:56:38 gibi yes. but I don't know if neutron ever needs that
10:56:38 gibi yes. but I don't know if neutron ever needs that
10:56:49 gibi that = ironic node name
10:56:49 gibi that = ironic node name
10:56:57 sean-k-mooney ya the neutron agent dont run on the ironic nodes
10:56:57 sean-k-mooney ya the neutron agent dont run on the ironic nodes
10:57:35 gibi I have to jump to a call from the top of the hour
10:57:35 gibi I have to jump to a call from the top of the hour
10:57:50 gibi I linked this discussion to the neutron patch for reference
10:57:50 gibi I linked this discussion to the neutron patch for reference
10:57:54 sean-k-mooney no worries thanks for litening
10:57:54 sean-k-mooney no worries thanks for litening
10:58:51 sean-k-mooney ill try an follow up downstream and see if we can come to can concreate conclutions
10:58:51 sean-k-mooney ill try an follow up downstream and see if we can come to can concreate conclutions
10:59:09 gibi OK
10:59:34 gibi sorry for being dense
10:59:34 gibi sorry for being dense
10:59:51 sean-k-mooney you werent being dense
10:59:51 sean-k-mooney you werent being dense
10:59:58 sean-k-mooney it helped
10:59:58 sean-k-mooney it helped
11:00:55 gibi then I'm glad :)
11:00:55 gibi then I'm glad :)
12:18:18 sean-k-mooney gibi: would adding a new weigher need a spec. or could that be a specless blueprint
12:18:18 sean-k-mooney gibi: would adding a new weigher need a spec. or could that be a specless blueprint
12:18:49 gibi if the config of the weigher is trivial then I'm OK with a specless bp
12:18:49 gibi if the config of the weigher is trivial then I'm OK with a specless bp
12:19:37 sean-k-mooney i have 2 that i think might be useful. 1 is weighe based on the number of runing instnace and the other is a weigher that prefers the current host an instance is on
12:19:37 sean-k-mooney i have 2 that i think might be useful. 1 is weighe based on the number of runing instnace and the other is a weigher that prefers the current host an instance is on
12:20:59 gibi please note that today all the weigher is turned on by default so more weigher means a more complex behavior.
12:20:59 gibi please note that today all the weigher is turned on by default so more weigher means a more complex behavior.
12:21:23 gibi * all the weighers are
12:21:23 gibi * all the weighers are
12:21:59 sean-k-mooney yep
12:22:20 sean-k-mooney we may or may not want that for these
12:22:20 sean-k-mooney we may or may not want that for these
12:22:44 sean-k-mooney basically our perf team has a bunch of hosts with differnt size for ram disk and cpu
12:22:44 sean-k-mooney basically our perf team has a bunch of hosts with differnt size for ram disk and cpu
12:23:11 sean-k-mooney and they are booting a bunch of small instance and they are all landing on 2 hosts because it has much more ram,cpu and disk then the others
12:23:11 sean-k-mooney and they are booting a bunch of small instance and they are all landing on 2 hosts because it has much more ram,cpu and disk then the others
12:24:36 gibi I think it is a valid need to have weighers that are not turned on by default
12:24:36 gibi I think it is a valid need to have weighers that are not turned on by default
12:24:38 sean-k-mooney as in the have 46 hosts and booted about 520 instace at this point and they have 2 host with 237 and 247 instances
12:24:38 sean-k-mooney as in the have 46 hosts and booted about 520 instace at this point and they have 2 host with 237 and 247 instances

Earlier   Later