Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2020-09-04
15:37:57 sean-k-mooney dansmith: im not against adding it
15:38:09 sean-k-mooney just the idea that voting = in gate and check
15:38:16 bauzas sean-k-mooney: okay, I think I found the problem
15:38:32 bauzas sean-k-mooney: for create, we don't need to verify the segments
15:38:39 bauzas for a network
15:38:43 bauzas but for a port, we do
15:38:47 sean-k-mooney bauzas: correct
15:38:49 bauzas and then, for a move op, too
15:38:55 sean-k-mooney well for a port only if it has an ip
15:39:00 bauzas we need to look at the port to know the segment
15:39:16 bauzas it has a port when moving, right?
15:39:30 sean-k-mooney yes
15:39:31 bauzas I mean, it does have an ip address
15:39:37 sean-k-mooney yes
15:39:40 bauzas okay, so for create, meh
15:39:49 gmann sean-k-mooney: there is no voting job which is not running on gate pipeline except the ceph one - https://review.opendev.org/#/c/747443/
15:39:55 bauzas unless if it has a specific address
15:40:00 sean-k-mooney gmann: nova-lvm
15:40:05 gmann it is n-v
15:40:08 sean-k-mooney gmann: that is only in check
15:40:29 gmann https://github.com/openstack/nova/blob/master/.zuul.yaml#L136
15:40:36 sean-k-mooney also the linux bridge one
15:40:59 sean-k-mooney oh we shoudl stop setting that there
15:41:03 sean-k-mooney and move it down
15:41:04 bauzas sean-k-mooney: fwiw, the spec is then invalid for the pseudo-code
15:41:13 bauzas sean-k-mooney: https://specs.openstack.org/openstack/nova-specs/specs/victoria/approved/routed-networks-scheduling.html#proposed-change
15:41:15 bauzas but meh
15:41:27 bauzas i'll look at the requested network
15:41:38 bauzas if it has a ip address, I'll look at the segment
15:41:56 bauzas if it doesn't have an ip address, I'll just look at all the segments from the network
15:42:02 bauzas sean-k-mooney: lgty ? ^
15:42:25 sean-k-mooney am ya that sound viable
15:42:32 bauzas cool
15:42:45 bauzas we're getting the fixed IPs from the VIF
15:42:54 bauzas (in the instance infocache)
15:43:00 sean-k-mooney yes
15:43:07 bauzas so I can look at them and ask for the related segments
15:43:11 bauzas amirite ?
15:43:22 gmann sean-k-mooney: yeha only neutron-tempest-linuxbridge is 2nd one not in gate, i did not notice this as it is defined in neutron side
15:43:37 bauzas now, the big question is : how can I get a segment from an IP address, but I'll figure this out
15:43:54 sean-k-mooney bauzas: you need to get the subnet with is in the vif too
15:44:01 bauzas ah, right
15:44:03 sean-k-mooney so instead of looking at the ips
15:44:08 sean-k-mooney you cna look at teh subnet
15:44:08 bauzas then this is better
15:44:32 sean-k-mooney so i would like to update the subnet object ot have the segment id in it optionally at some point
15:44:43 bauzas okay, lemme look
15:44:55 sean-k-mooney well proably beter to put the segment in the vif object
15:45:05 bauzas eek
15:45:06 sean-k-mooney we should not have a singel port on multipel segment
15:45:07 bauzas can't do it
15:45:21 sean-k-mooney that is the correct place to put it
15:46:43 sean-k-mooney we can put it in the subnet but even though a port can have multipel ips form multiple subnets in general that is not true for routed networks
15:47:07 sean-k-mooney well...
15:47:08 gibi bauzas: I think looking at the ip_allocation of the port is a better source of information than checking if it has an ip or not
15:47:25 bauzas sean-k-mooney: I can try to lookup directly in the pre-filter
15:47:36 bauzas gibi: good point
15:47:37 sean-k-mooney no it cant because of limitation in the neutron implemation
15:47:45 bauzas but grrrr, I need to stop working
15:48:09 bauzas parents-teacher meeting at 6pm, since kids are back at school
15:48:10 sean-k-mooney gibi: we could but i think caching it in the VIF is better
15:48:32 sean-k-mooney to avoid needign to call neutron during the prefileter
15:48:35 gibi sean-k-mooney: for move operations I'm OK with that
15:48:49 sean-k-mooney gibi: ya for create since we dont have the vif objects
15:48:50 gibi bauzas: no worries, enjoy your weekend
15:48:56 bauzas thinking out loud, this is unrealistic to merge it before victoria-3
15:48:58 sean-k-mooney we need to look it up if we are passed a port
15:49:27 sean-k-mooney proably altough viable for early m1
15:49:31 bauzas so, I'll continue working on it, but let's all agree this will be a Wallaby merge
15:49:43 bauzas yeah, can continue investigating during RC1
15:49:55 bauzas we're close to the job done, I think
15:50:02 gibi bauzas: OK. let's merge this in W
15:50:21 sean-k-mooney bauzas: for the limited usecases in the spec maybe :)
15:50:29 sean-k-mooney enjoy o/
15:51:11 gibi bauzas: also if you boot an instance with a network then I think nothing guarantees that every compute host will is attached to one of the segments of that networks. So we still have to filter hosts based on segments even if the instance is booted with network
15:51:59 bauzas gibi: we do it already in my change
15:52:29 gibi bauzas: yes, you just said something about it above and I wanted to make sure this check will not be removed
15:52:32 gibi :)
15:52:46 gibi bauzas: also I read through your patches and I like them
15:52:59 gibi thanks for picking up this feature
15:53:27 sean-k-mooney gibi: ya so bauzas is going to limit to the full set of segment on a network if we boot to a network
15:53:46 sean-k-mooney or just the segment related to the ip if we have a port with an ip
15:54:11 gibi sean-k-mooney: yeah that feels correct
15:54:13 sean-k-mooney gibi: so that will covert the case fo only looking at the subset of host with the correct segment reacblitiy
15:54:21 gibi aagree
15:56:07 sean-k-mooney gibi: did you see https://review.opendev.org/#/c/749175/1 by the way
15:56:32 sean-k-mooney that might fix the PF edgecase i notes while testing your code
15:56:39 gibi no I haven't
15:56:49 gibi I will check that patch on Monday
15:56:51 sean-k-mooney ill try to test that on monday and ill retest your code
15:56:56 gibi cool
15:57:14 sean-k-mooney they are proposing it for a different edgecase
15:57:31 sean-k-mooney where a deivce did not support sriov and after some operation it now does
15:57:42 sean-k-mooney changing it form type-pci to type-PF
15:58:39 sean-k-mooney that only happens if 1 you change the device phsyically, 2 you updated the frimeare to add sriov supprot, 3 you reconfitre the firmware/bios to supprot it by for exampel chanige from datacenter bridging mode to sriov mode
15:59:03 sean-k-mooney so there usecase is rare but the fix i think might fix both
15:59:30 sean-k-mooney although i might also need to be update to hanel the removed vf case
16:13:52 openstackgerrit Merged openstack/nova master: doc: Add IPv6 metadata address https://review.opendev.org/744492
16:23:06 gibi o/
16:23:49 lyarwood \o

Earlier   Later