Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2022-08-25
11:39:51 sean-k-mooney ack ya that can be a bit of a pain. are you going to use the falvor uuid in requester_id to correalte the allcoatiosn/devices
11:39:59 sean-k-mooney or that is not needed
11:40:02 gibi not needed
11:40:06 sean-k-mooney ack
11:40:16 gibi I needed to depend on the rt.tracked_migrations dict though
11:40:33 gibi to detect same host resize during placement update
11:41:00 sean-k-mooney ok you shoudl have access to that i guess
11:41:10 gibi yes
11:41:14 sean-k-mooney since the pci/placment stuff is happening in the compute manager
11:41:24 gibi it is happening in the rt
11:41:24 sean-k-mooney i dont think we have access to that in the driver do we?
11:41:27 sean-k-mooney the resouce tracker
11:41:32 gibi no we don't in the driver
11:41:55 sean-k-mooney ya thats what i tought ok so you need that in the move claim code?
11:42:11 sean-k-mooney i can just review the patch when you push it too
11:42:33 sean-k-mooney but i assume its needed to correctly figure out what devices to use
11:43:56 gibi it is needed to ignore devices where dev.instance_uuid is our instance but actually the dev is part of the source host allocation
11:44:20 sean-k-mooney ack ya i assumed it was somethign like that
11:44:31 sean-k-mooney do you also need it for the revert path
11:45:08 gibi yes
11:45:14 gibi in the drop migration claim
11:45:31 sean-k-mooney ya to free only the "dest" pci devices
11:45:32 gibi this is the same host resize fix without revert support https://review.opendev.org/c/openstack/nova/+/854441/2/nova/compute/resource_tracker.py
11:45:50 gibi the revert support is just passed locally
11:46:03 gibi btw, only the allocation healing needed tuning due to same host
11:46:16 gibi as it tried to heal allocation for the source devs
11:46:40 gibi the actual scheduling and claim code was OK
11:47:06 sean-k-mooney well the claim code shoudl be common for the placment/non placment workflows
11:47:15 sean-k-mooney and same hsot resize works for pci deivces in general
11:47:33 sean-k-mooney although i still suspect that bug you found with multiple alaises
11:47:39 sean-k-mooney coudl have causes some odd behaivor
11:47:57 gibi yes
11:48:16 sean-k-mooney but ya that makes sense that it woudl only be the allocations code that need tweaking
11:48:32 gibi and the allocation healing will go away after the upgrade
11:48:46 gibi as the scheduler will create proper allocations for new instances and new moves
11:49:13 gibi so all the allocation healing code is temporary until AA/BB
11:49:38 gibi so while same host needed a special case there, that special casing can be removed later
11:49:51 sean-k-mooney technially i think we could drop it in AA but we might want to kep it to BB
11:50:09 gibi it depends on how the neutron based sriov impl comes along
11:50:14 gibi as healing is needed there too
11:50:22 sean-k-mooney ya
11:50:31 sean-k-mooney so BB at the earlist then
11:50:45 sean-k-mooney we woudl have to deprecate the non placment path
11:50:58 sean-k-mooney in AA if we wanted to drop this in BB
11:51:06 sean-k-mooney so maybe even later
11:51:29 gibi yeah
11:51:43 gibi I think so
12:00:29 sean-k-mooney by the way i said i woudl reviwe the ephmeral encyption next but i started with the viommu seriese instad so ill finish that then swap
12:00:42 sean-k-mooney i +2 the new os-traits trait for config drive by the way
12:01:00 sean-k-mooney so if you can review that too we can mege it and get another release done today
12:01:06 sean-k-mooney before non client lib freeze
12:06:21 opendevreview Balazs Gibizer proposed openstack/nova master: Func test for PCI in placement scheduling https://review.opendev.org/c/openstack/nova/+/854122
12:06:22 opendevreview Balazs Gibizer proposed openstack/nova master: Support same host resize with PCI in placement https://review.opendev.org/c/openstack/nova/+/854441
12:06:22 opendevreview Balazs Gibizer proposed openstack/nova master: Support cold migrate and resize with PCI tracking in placement https://review.opendev.org/c/openstack/nova/+/854247
12:06:23 opendevreview Balazs Gibizer proposed openstack/nova master: Support unshelve with PCI in placement https://review.opendev.org/c/openstack/nova/+/854616
12:06:23 opendevreview Balazs Gibizer proposed openstack/nova master: Support evacuate with PCI in placement https://review.opendev.org/c/openstack/nova/+/854615
12:07:10 sean-k-mooney gibi: can you look at https://review.opendev.org/c/openstack/os-traits/+/854599
12:07:25 gibi yes I will today
12:23:08 opendevreview Merged openstack/nova master: Reproduce bug 1986838 https://review.opendev.org/c/openstack/nova/+/853516
12:27:25 opendevreview Merged openstack/os-traits master: add new trait for config drive regeneration https://review.opendev.org/c/openstack/os-traits/+/854599
12:39:37 sean-k-mooney ricolin: i just reviewed the iommu seriese https://review.opendev.org/q/topic:bp%252Fvirtual-iommu-support
12:39:51 sean-k-mooney ricolin: +2 on the first patch and -2 on the last two
12:39:57 sean-k-mooney *-1
12:40:14 sean-k-mooney the second patch is very close
12:40:43 sean-k-mooney the final patch need to be redone based on the libvirt version not using data that wont exsits with real libvirt
12:41:50 sean-k-mooney ricolin: ill take a look again next week if you respine them but feel free to ping me when thery are ready
12:42:16 sean-k-mooney gibi: ack as release liason i can provie the ptl ack once its proposed
12:42:25 gibi cool
12:42:57 sean-k-mooney did we relax the placement checks
12:43:06 sean-k-mooney or rather merge your patch to do that
12:43:13 sean-k-mooney or will this break placment again
12:45:56 gibi good question
12:45:58 gibi looking
12:46:44 gibi we merged https://review.opendev.org/c/openstack/placement/+/851966 that changed the testing strategy
12:47:31 gibi so the new os-traits should not break the placement tests
12:47:40 sean-k-mooney cool
12:48:06 gibi release patch is up https://review.opendev.org/c/openstack/releases/+/854617
12:50:53 sean-k-mooney hum it proably coudl have been 2.8.1 but im ok with this i guess
12:52:31 gibi it is not a bugfix
12:52:55 sean-k-mooney ya but adding a constnat is not really a feature either
12:53:01 sean-k-mooney im +1 on it in any case
12:53:10 sean-k-mooney so the bot shoudl addd the ptl approved flag soon
12:53:15 gibi thanks
12:53:32 sean-k-mooney feature bumps are what we did in the past anyway
12:53:41 sean-k-mooney so your being consitnet with that
13:13:27 gibi ... it tells a lot that our func test helper mixin does help testing server multicreate :)
13:13:43 gibi but it wont stop me
13:14:09 gibi that is the last item on the PCI bucket list (as revert resize and rescheduler seems to work)
13:38:41 sean-k-mooney ack. i think punting multi create is vaild given the know issue there if it comes to that
13:39:08 opendevreview Rajesh Tailor proposed openstack/nova master: Fix rescue volume-based instance https://review.opendev.org/c/openstack/nova/+/852737
13:47:40 opendevreview Balazs Gibizer proposed openstack/nova master: Test reschedule with PCI in placement https://review.opendev.org/c/openstack/nova/+/854626
13:47:53 gibi sean-k-mooney: ack, I give it a try but I will limit the effort
13:50:48 sean-k-mooney given its an exsiting limiation for ports with resouce request or any other usage of nested resouce providers i think its accpatble to adress this genericaly in a future release.
13:51:06 sean-k-mooney and posibly even backport it as its a know bug/limiation depending on what the fix looks like
13:53:05 opendevreview Arnaud Morin proposed openstack/nova master: Allow plugging port when device_id is instance_id https://review.opendev.org/c/openstack/nova/+/854627
13:53:33 sean-k-mooney so hard no on ^
13:53:35 sean-k-mooney #right
13:54:18 amorin hello
13:54:30 sean-k-mooney o/
13:54:40 sean-k-mooney amorin: we do not want to supprot that workflow

Earlier   Later