| *** takashin has joined #openstack-placement | 00:41 | |
| *** tetsuro has joined #openstack-placement | 00:47 | |
| *** yikun has joined #openstack-placement | 00:52 | |
| *** takashin has left #openstack-placement | 03:02 | |
| *** tetsuro has quit IRC | 04:00 | |
| *** e0ne has joined #openstack-placement | 05:08 | |
| *** e0ne has quit IRC | 05:49 | |
| *** tetsuro has joined #openstack-placement | 07:06 | |
| *** helenafm has joined #openstack-placement | 07:31 | |
| openstackgerrit | Zhenyu Zheng proposed openstack/nova-specs master: Detach and attach boot volumes - Stein https://review.openstack.org/600628 | 07:42 |
|---|---|---|
| *** e0ne has joined #openstack-placement | 08:10 | |
| *** e0ne has quit IRC | 08:20 | |
| *** tssurya has joined #openstack-placement | 08:54 | |
| *** e0ne has joined #openstack-placement | 09:04 | |
| *** tetsuro has quit IRC | 09:07 | |
| *** ttsiouts has joined #openstack-placement | 10:02 | |
| *** ttsiouts has quit IRC | 10:24 | |
| *** ttsiouts has joined #openstack-placement | 10:25 | |
| *** e0ne has quit IRC | 10:27 | |
| *** ttsiouts has quit IRC | 10:29 | |
| *** ttsiouts has joined #openstack-placement | 10:30 | |
| *** ttsiouts has quit IRC | 11:02 | |
| *** jaypipes has joined #openstack-placement | 11:17 | |
| *** ttsiouts has joined #openstack-placement | 11:49 | |
| *** e0ne has joined #openstack-placement | 11:53 | |
| *** ttsiouts has quit IRC | 12:55 | |
| *** ttsiouts has joined #openstack-placement | 13:03 | |
| *** mriedem has joined #openstack-placement | 13:03 | |
| *** ttsiouts has quit IRC | 13:19 | |
| openstackgerrit | Eric Fried proposed openstack/placement master: Fix member_of doc in RequestGroup.dict_from_request https://review.openstack.org/605427 | 13:53 |
| *** ttsiouts has joined #openstack-placement | 14:03 | |
| *** ttsiouts has quit IRC | 14:14 | |
| *** ttsiouts has joined #openstack-placement | 14:21 | |
| *** helenafm has quit IRC | 14:47 | |
| openstackgerrit | Artom Lifshitz proposed openstack/nova-specs master: Re-propose numa-aware-live-migration spec https://review.openstack.org/599587 | 15:31 |
| *** tssurya has quit IRC | 15:55 | |
| *** gibi has quit IRC | 16:14 | |
| *** ttsiouts has quit IRC | 16:35 | |
| *** e0ne has quit IRC | 16:43 | |
| stephenfin | jaypipes: Out of curiosity, in a world where PCPUs *and* NUMA nodes are modelled in placement, how do we map PCPUs to a NUMA node in nova-compute? | 17:12 |
| stephenfin | jaypipes: For example, node #0 has 8 CPUs, node #1 has 6 CPUs. We boot an instance with 6 CPUs and placement takes those from node #1. We need to make sure we pin to CPUs on that NUMA node, otherwise what exists in placement doesn't match what's on the host, right? | 17:13 |
| jaypipes | stephenfin: each NUMA provider would have an inventory of PCPU resource. | 17:13 |
| jaypipes | stephenfin: the *assignment* of a particular host CPU to a guest virtual CPU thread isn't done in placement at all. | 17:14 |
| stephenfin | How do we avoid that mismatch above? | 17:14 |
| stephenfin | If something other than placement (nova-compute) does the calculation and pins an instance to the "wrong" NUMA node, do we go back and update the allocations? | 17:15 |
| jaypipes | stephenfin: the scheduler will need to pick an allocation request from the list of allocation candidates. that allocation request will take the 6 from one of the other NUMA node's resource provider. We will need to return the resource provider's name in the provider_summaries section of ht eallocation candidates response in order for the scheduler to know which NUMA node corresponds to a prticular resource provider UUID. | 17:16 |
| jaypipes | stephenfin: we never "go back and update allocations". | 17:16 |
| stephenfin | jaypipes: Ah, so some unique identifier about the node will be encoded in the name? | 17:17 |
| stephenfin | I know we discussed this in Denver but only from the context of cyborg | 17:17 |
| jaypipes | stephenfin: yes. probably will be just {compute_node_hypervisor_hostname}_NUMA{index} or something like that. chat with bauzas :) | 17:19 |
| stephenfin | I figured we wouldn't go back and update but I was missing how we'd know placement's NUMA provider with an inventory of 8 CPUs corresponded to NUMA node 1 and not NUMA node 0 | 17:20 |
| stephenfin | That sounds as good as anything though | 17:20 |
| jaypipes | stephenfin: for PCI devices, the proposed naming convention is {hypervisor_hostname}_{pci_device_address}. Would be great if NUMA nodes had a similar addressing scheme, but they don't... | 17:24 |
| *** gibi has joined #openstack-placement | 18:18 | |
| *** e0ne has joined #openstack-placement | 19:15 | |
| *** e0ne has quit IRC | 19:30 | |
| *** gibi has quit IRC | 20:12 | |
| *** gibi has joined #openstack-placement | 20:45 | |
| openstackgerrit | Chris Friesen proposed openstack/nova-specs master: Add support for emulated virtual TPM https://review.openstack.org/571111 | 22:05 |
| openstackgerrit | Chris Friesen proposed openstack/nova-specs master: Add support for emulated virtual TPM https://review.openstack.org/571111 | 22:08 |
| *** mriedem is now known as mriedem_away | 23:30 | |
| *** openstackgerrit has quit IRC | 23:49 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!