| opendevreview | chandan kumar proposed openstack/cyborg master: Add NVMeDriver for NVMe controller discovery https://review.opendev.org/c/openstack/cyborg/+/999937 | 04:56 |
|---|---|---|
| opendevreview | chandan kumar proposed openstack/cyborg master: agent: separate driver loading from driver initialisation https://review.opendev.org/c/openstack/cyborg/+/999938 | 04:56 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Route ARQ unbind through conductor-to-agent cleanup RPCs https://review.opendev.org/c/openstack/cyborg/+/999939 | 04:56 |
| opendevreview | chandan kumar proposed openstack/cyborg master: nvme: implement NVMeDriver.cleanup() https://review.opendev.org/c/openstack/cyborg/+/999940 | 04:56 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Add device_state bind guard and fix fake driver topology https://review.opendev.org/c/openstack/cyborg/+/999941 | 04:56 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Expose device_state in device API and register nvme_driver entrypoint https://review.opendev.org/c/openstack/cyborg/+/999942 | 04:56 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Deprecate SSD drivers in favour of NVMe driver https://review.opendev.org/c/openstack/cyborg/+/999943 | 04:57 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Add NVMe driver admin guide and API reference for microversion 2.4 https://review.opendev.org/c/openstack/cyborg/+/999944 | 04:57 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Add per-PF vf_personality sysfs interface https://review.opendev.org/c/openstack/cyborg/+/998603 | 04:57 |
| opendevreview | chandan kumar proposed openstack/cyborg master: pci-sim: introduce personality ops-table dispatch for VF device types https://review.opendev.org/c/openstack/cyborg/+/999125 | 04:57 |
| opendevreview | chandan kumar proposed openstack/cyborg master: pci-sim: add NVMe host kernel module https://review.opendev.org/c/openstack/cyborg/+/999928 | 04:58 |
| opendevreview | chandan kumar proposed openstack/cyborg master: pci-sim: add NVMe VFIO guest probe and host handoff https://review.opendev.org/c/openstack/cyborg/+/999929 | 04:58 |
| opendevreview | chandan kumar proposed openstack/cyborg master: pci-sim: add software MSI-X domain to fake host bridge https://review.opendev.org/c/openstack/cyborg/+/1003255 | 04:58 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Add NVMe driver devstack plugin support https://review.opendev.org/c/openstack/cyborg/+/999945 | 04:58 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Add NVMeDriver for NVMe controller discovery https://review.opendev.org/c/openstack/cyborg/+/999937 | 06:01 |
| opendevreview | chandan kumar proposed openstack/cyborg master: agent: separate driver loading from driver initialisation https://review.opendev.org/c/openstack/cyborg/+/999938 | 06:02 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Route ARQ unbind through conductor-to-agent cleanup RPCs https://review.opendev.org/c/openstack/cyborg/+/999939 | 06:03 |
| opendevreview | chandan kumar proposed openstack/cyborg master: nvme: implement NVMeDriver.cleanup() https://review.opendev.org/c/openstack/cyborg/+/999940 | 06:04 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Add device_state bind guard and fix fake driver topology https://review.opendev.org/c/openstack/cyborg/+/999941 | 06:05 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Expose device_state in device API and register nvme_driver entrypoint https://review.opendev.org/c/openstack/cyborg/+/999942 | 06:06 |
| opendevreview | chandan kumar proposed openstack/cyborg master: agent: separate driver loading from driver initialisation https://review.opendev.org/c/openstack/cyborg/+/999938 | 06:07 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Route ARQ unbind through conductor-to-agent cleanup RPCs https://review.opendev.org/c/openstack/cyborg/+/999939 | 06:07 |
| opendevreview | chandan kumar proposed openstack/cyborg master: nvme: implement NVMeDriver.cleanup() https://review.opendev.org/c/openstack/cyborg/+/999940 | 06:07 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Add device_state bind guard and fix fake driver topology https://review.opendev.org/c/openstack/cyborg/+/999941 | 06:07 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Expose device_state in device API and register nvme_driver entrypoint https://review.opendev.org/c/openstack/cyborg/+/999942 | 06:08 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Deprecate SSD drivers in favour of NVMe driver https://review.opendev.org/c/openstack/cyborg/+/999943 | 06:08 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Add NVMe driver admin guide and API reference for microversion 2.4 https://review.opendev.org/c/openstack/cyborg/+/999944 | 06:08 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Add per-PF vf_personality sysfs interface https://review.opendev.org/c/openstack/cyborg/+/998603 | 06:08 |
| opendevreview | chandan kumar proposed openstack/cyborg master: pci-sim: introduce personality ops-table dispatch for VF device types https://review.opendev.org/c/openstack/cyborg/+/999125 | 06:09 |
| opendevreview | chandan kumar proposed openstack/cyborg master: pci-sim: add NVMe host kernel module https://review.opendev.org/c/openstack/cyborg/+/999928 | 06:09 |
| opendevreview | chandan kumar proposed openstack/cyborg master: pci-sim: add NVMe VFIO guest probe and host handoff https://review.opendev.org/c/openstack/cyborg/+/999929 | 06:09 |
| opendevreview | chandan kumar proposed openstack/cyborg master: pci-sim: add software MSI-X domain to fake host bridge https://review.opendev.org/c/openstack/cyborg/+/1003255 | 06:09 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Add NVMe driver devstack plugin support https://review.opendev.org/c/openstack/cyborg/+/999945 | 06:10 |
| chandankumar | Hello sean-k-mooney, I think we can push nvme generic driver work to next release. | 11:51 |
| chandankumar | I am not sure we can get it ready today for merge. | 11:51 |
| sean-k-mooney | ack, that would reduce the review pressure | 11:54 |
| sean-k-mooney | in that case we can shift focus to createing the release note prelude and marketing highlihgt | 11:55 |
| sean-k-mooney | and then cutting rc1 | 11:55 |
| chandankumar | Recent nvme changes are working fine on both hardware and pci-sim implementation | 11:56 |
| sean-k-mooney | we can then continue working on this once the specc is repoposed and aprpoved which will be easy enough | 11:56 |
| chandankumar | let me get a review up for prelude by monday | 11:57 |
| sean-k-mooney | ack if you can do that it would be cool i was going to do it if i didnt see one by the end of the day | 11:58 |
| sean-k-mooney | i do have your nvme series open now by the way but i only stared revieweing it 20 mins ago | 11:58 |
| sean-k-mooney | i didnt get to it yesterday evening after you revised it | 11:59 |
| chandankumar | let'see how much we can get it | 11:59 |
| chandankumar | I will get the prelude cr ready then | 11:59 |
| sean-k-mooney | chandankumar: direct db access is not allowed form the comptue agent and your later patches are doing that | 14:15 |
| sean-k-mooney | https://review.opendev.org/c/openstack/cyborg/+/1002254/1 is needed | 14:15 |
| sean-k-mooney | i think we will just stop there for this cycle | 14:15 |
| sean-k-mooney | im going to drop for a while | 14:15 |
| chandankumar | ah, in that case, I will put these two patches in my series | 14:21 |
| chandankumar | sounds good | 14:21 |
| sean-k-mooney | i think we will want to merge the first 3 patches in this sersie https://review.opendev.org/c/openstack/cyborg/+/1002255/1 | 14:41 |
| sean-k-mooney | and then you can add the extra remotabels methods | 14:41 |
| sean-k-mooney | but those are not entrily ready yet | 14:42 |
| sean-k-mooney | so we may want to pasue for a few days to plan how best to proceed and make sure we are not missign anything | 14:42 |
| opendevreview | chandan kumar proposed openstack/cyborg master: Add NVMeDriver for NVMe controller discovery https://review.opendev.org/c/openstack/cyborg/+/999937 | 14:43 |
| opendevreview | chandan kumar proposed openstack/cyborg master: agent: separate driver loading from driver initialisation https://review.opendev.org/c/openstack/cyborg/+/999938 | 14:43 |
| chandankumar | for right now. I am putting 3 patches on top of 999938 | 14:45 |
| sean-k-mooney | dont move my patches | 14:46 |
| sean-k-mooney | you can put your series on top of them but dont ther first patch is a bugfix that shoudl merge before your feature changes | 14:46 |
| sean-k-mooney | or at least that shoudl not depend on anything in your series | 14:47 |
| sean-k-mooney | it has helper that your sersie coudl reuse by the way | 14:47 |
| sean-k-mooney | https://review.opendev.org/c/openstack/cyborg/+/1002253/1/cyborg/conductor/manager.py#147 | 14:47 |
| sean-k-mooney | to determin if a device is in use | 14:47 |
| sean-k-mooney | but that part of the logic im not sure is 100% correct yet | 14:49 |
| chandankumar | let me also take a pause | 14:49 |
| sean-k-mooney | the reason im not pushign to get my patches merged is i have not done the deep dive on them to fully snrue there logic is correct | 14:50 |
| chandankumar | in that case, I will keep a patch what is need for my patch instead of stacking | 14:50 |
| sean-k-mooney | they appare to solve the bug but im unfomoftbale iwth the handelign of device vs atchement handel vs deployable | 14:50 |
| chandankumar | s/patch/piece | 14:50 |
| sean-k-mooney | the beit that we need to reconsile is as follows https://github.com/openstack/cyborg/blob/master/cyborg/db/sqlalchemy/models.py#L78-L141 | 14:53 |
| sean-k-mooney | a device is a recored for a phasical device | 14:53 |
| sean-k-mooney | the deployable is a pool of devices modled in placement as a single resouce provider with an nvientory of n acclerators | 14:53 |
| sean-k-mooney | each alclare correspodnign to a attachment handel/device | 14:54 |
| sean-k-mooney | i think i have the names of these two functions reverst | 14:54 |
| sean-k-mooney | https://review.opendev.org/c/openstack/cyborg/+/1002253/1/cyborg/conductor/manager.py#147 | 14:55 |
| sean-k-mooney | so what im callign _deployable_in_use is really attachment handel in use or device in use | 14:55 |
| sean-k-mooney | and _device_in_use is deployable in use | 14:55 |
| sean-k-mooney | a deployable is in use if any of its devices/attachmetn handels are in use | 14:56 |
| sean-k-mooney | a device is only in uses if its bound or its correspodnign attachment handel is in use | 14:56 |
| sean-k-mooney | in general the reserved value in placment for a given inventory shoudl be equal to the numebr of insue attacheent handels/devices | 14:57 |
| sean-k-mooney | def _deployable_in_use(driver_dep_obj): | 14:59 |
| sean-k-mooney | # DriverAttachHandle.list() combines the persisted usage marker with | 14:59 |
| sean-k-mooney | # any remaining ExtARQ ownership reference. | 14:59 |
| sean-k-mooney | return any( | 15:00 |
| sean-k-mooney | attach_handle.in_use | 15:00 |
| sean-k-mooney | for attach_handle in driver_dep_obj.attach_handle_list | 15:00 |
| sean-k-mooney | ) | 15:00 |
| sean-k-mooney | that is actully correct | 15:00 |
| sean-k-mooney | its | 15:00 |
| sean-k-mooney | @classmethod | 15:00 |
| sean-k-mooney | def _device_in_use(cls, driver_dev_obj): | 15:00 |
| sean-k-mooney | return any( | 15:00 |
| sean-k-mooney | cls._deployable_in_use(deployable) | 15:00 |
| sean-k-mooney | for deployable in driver_dev_obj.deployable_list | 15:00 |
| sean-k-mooney | ) | 15:00 |
| sean-k-mooney | that is not | 15:00 |
| sean-k-mooney | the correct way to check fi a device is in use is to fined the assocated atchement handel and check if that is in use | 15:00 |
| sean-k-mooney | thats what i need to fix and then i need to check the sourinding logic | 15:01 |
| sean-k-mooney | so i think the depensces are going to be as follows, 1 i will fix https://bugs.launchpad.net/openstack-cyborg/+bug/2017513, then we will intoduce the ovo interaction api and move the arq bidning to the comptue agent adn finally we will resume the nvme driver implmeention once those 3 things are complete | 15:07 |
| sean-k-mooney | we may be able to do it before moving bidn to the agent but we need to thikn about it. | 15:08 |
| chandankumar | ok | 15:10 |
| sean-k-mooney | the fact that the ovo indriection api was not in place was something i was not aware of during the spec review | 15:10 |
| sean-k-mooney | without that he only other option we have woudl be a dedicated rpc form the comptue to the conductor | 15:11 |
| sean-k-mooney | to marke the device as cleaned | 15:11 |
| sean-k-mooney | or relying on a perodic update | 15:11 |
| sean-k-mooney | basiclly calling report_data https://github.com/openstack/cyborg/blob/master/cyborg/conductor/rpcapi.py#L50-L60 | 15:12 |
| sean-k-mooney | that is an option today | 15:13 |
| sean-k-mooney | after the deivce is cleaned | 15:13 |
| chandankumar | for now, I will copy _deployable_in_use and run ut on the conductor against device attach handle. | 15:13 |
| chandankumar | adn then count the inuse attachment handles first, then set reserved to the inuse attach handle and totel to number of accelerator. | 15:13 |
| sean-k-mooney | ack | 15:14 |
| sean-k-mooney | i want to meove _deployable_in_use to the deployable ovo eventually | 15:14 |
| sean-k-mooney | as a property so you can just do deployable.in_use | 15:15 |
| sean-k-mooney | anyway i have to go now o/ | 15:15 |
| chandankumar | sure, I will update the patch based on above discussion | 15:16 |
| chandankumar | thank you! | 15:16 |
Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!