| opendevreview | Merged openstack/nova master: mem-enc: adjust requirement checks for mem_encryption guests https://review.opendev.org/c/openstack/nova/+/967974 | 02:19 |
|---|---|---|
| opendevreview | Merged openstack/nova master: mem-enc: address minor issues pointed out in the review https://review.opendev.org/c/openstack/nova/+/983278 | 02:19 |
| opendevreview | Goutham Pacha Ravi proposed openstack/nova-specs master: Add spec for virtiofs live and cold migration https://review.opendev.org/c/openstack/nova-specs/+/983801 | 05:46 |
| opendevreview | Goutham Pacha Ravi proposed openstack/nova-specs master: Add spec for virtiofs live and cold migration https://review.opendev.org/c/openstack/nova-specs/+/983801 | 06:16 |
| tafkamax | Hi, I have a question regarding nova-pythonclient. Will commands like nova reset-state <SERVER_ID> be migrated? | 06:40 |
| tafkamax | So when python-openstackclient will become the only one standing. | 06:40 |
| jkulik | Isn't it already available through `opennstack server set --state active`? | 06:42 |
| tafkamax | oh okay, will try | 06:47 |
| opendevreview | Kamil Sambor proposed openstack/nova master: Test nova CLI commands with native threading https://review.opendev.org/c/openstack/nova/+/984036 | 08:47 |
| opendevreview | Kamil Sambor proposed openstack/nova master: Enable threading mode for proxy services https://review.opendev.org/c/openstack/nova/+/976089 | 12:27 |
| bauzas | as a reminder, the PTG started for now, you can join us by clicking on https://meetpad.opendev.org/nova-2026.2-ptg | 13:17 |
| dansmith | Uggla: didn't you just say we were supposed to be in the cinder room? | 14:04 |
| dansmith | there was some confusion about where to be (here in the cinder room) | 14:04 |
| Uggla | no liberty Nova room (sorry if it was not clear) | 14:05 |
| melwitt | gibi: thanks for checking on the nova-vtpm patch. I saw it failed again :( so I will investigate | 15:15 |
| gibi | Uggla: your nova bug stats will not like my effort of filing bugs for eventlet tech debt https://bugs.launchpad.net/nova/+bugs?field.tag=eventlet-removal | 15:36 |
| gibi | melwitt: ack, thanks for trying to stabilize it | 15:37 |
| stephenfin | tafkamax: https://docs.openstack.org/python-openstackclient/latest/cli/decoder.html | 15:42 |
| opendevreview | Merged openstack/os-vif master: Bump min version of pyroute2 https://review.opendev.org/c/openstack/os-vif/+/984934 | 15:56 |
| opendevreview | Stephen Finucane proposed openstack/nova master: objects: Prepare for oslo.versionedobjects 3.10.0 https://review.opendev.org/c/openstack/nova/+/985506 | 16:10 |
| tkajinam | I'm just wondering how memfd works for allocation. Should we still calculate memory allocation according to the numa topology ? | 16:10 |
| tkajinam | (assuming we still have to align cpu amd mem in numa-aware way, looking at it may be used for ovs-dpdk | 16:10 |
| tkajinam | by the way memfd is heavily used in cc case (actually instances with snp support or tdx support uses it implicitly and it's interesting after learning the mechanism quite recently | 16:11 |
| Uggla | @gibi, I think the bug you opened are valid so can be triaged. Anyway that's not too bad. | 16:22 |
| Uggla | tkajinam, good to know memfd is something interesting for CC case. | 16:23 |
| Uggla | tkajinam, if you have a doc pointer on it I'm interesting. | 16:23 |
| Uggla | tkajinam, probably a good point to convince my mgmt that this feature is interesting. | 16:24 |
| tkajinam | Uggla, let me find one, but it might not match your use case largely because 1) memfd is implicitly used by qemu 2) the main objective of using memfd is not performance but allocating private/encrypted memory for vm | 16:46 |
| * tkajinam is looking for qemu commit, in addition to https://github.com/AMDESE/AMDSEV/issues/236 | 16:46 | |
| Uggla | tkajinam maybe part of an answer to your question --> https://lwn.net/Articles/990663/ | 16:50 |
| tkajinam | ok so you still need to maintain memory:numa mapping (seeing that host-nodes=0) | 16:51 |
| tkajinam | Uggla, this might be too-much information but captures the qemu's internal logic to leverage memfd for private memory area https://review.opendev.org/c/openstack/nova-specs/+/983376/comment/658b8a62_20071ca8/ | 16:52 |
| sean-k-mooney | tkajinam: memfd can have the shared flag set or not i bleive | 16:54 |
| tkajinam | yeah | 16:54 |
| sean-k-mooney | i woudl prefer to always have memfd shared=true | 16:54 |
| sean-k-mooney | but keep "annoumus" spelled correctly for those that want to opt out | 16:55 |
| sean-k-mooney | i.e. evnetually memfd i think shoudl become the default | 16:55 |
| sean-k-mooney | but ya in newer qemu memfd does not need numa | 16:55 |
| sean-k-mooney | but it did when it was first done | 16:55 |
| sean-k-mooney | the fact that memfd does not require numa affinity is one of the reaosn its intersting | 16:57 |
| tkajinam | but not requiring numa is valid only when you don't need numa aware things, right ? | 17:01 |
| tkajinam | I'm a bit confused because the spec mentions OVS-DPDK which likely require numa-aware allocation of cpu/mem | 17:01 |
| tkajinam | while it might not be needed for virtio-fs thing | 17:01 |
| sean-k-mooney | correct | 17:02 |
| sean-k-mooney | memfd and numa affinity are ortoganal | 17:02 |
| sean-k-mooney | if you want both you shoudl request both | 17:02 |
| tkajinam | ok. that makes sense | 17:02 |
| sean-k-mooney | today to use numa in nova you have to set hw:mem_page_size to any vallid value or use file backed memory | 17:03 |
| sean-k-mooney | so memfd would not change that | 17:04 |
| tkajinam | ok | 17:05 |
| Uggla | dansmith, gouthamr, bauzas What I meant is during the PTG topic: | 17:14 |
| Uggla | 1- We blocked actions on instances with an attached share to reduce the scope. | 17:14 |
| Uggla | 2- Also mainly because older virtiofsd (<1.11) doesn’t support migration — per libvirt docs: “Older versions of virtiofsd (prior to 1.11) do not support migration…”, so ops like live-migration, save/restore, or memory snapshots may not work with a virtiofs share attached. (https://libvirt.org/kbase/virtiofs.html) | 17:14 |
| Uggla | 3- So the blocked operation due to libvirt/qemu is not only live-migation but also suspend and shelve (I was unsure about the impacted ones during the call) | 17:26 |
| dansmith | suspend I understand, but not shelve.. how does that have any requirement on libvirt/qemu? | 17:29 |
| sean-k-mooney | i think we dont supprot snapshot and that the issue with shelve but we do allow nova to stop the vm to do snapshot | 17:33 |
| sean-k-mooney | and for shelve specificly | 17:34 |
| sean-k-mooney | we only do the snapshot in the shelve offload pahse | 17:34 |
| sean-k-mooney | so the vm is stopped so ya i dont knwo why we could not od that today | 17:34 |
| dansmith | I thought shelve did the snapshot not offload, but either way, the instance should be stopped already so yeah I can't think of why that would be a problem | 17:35 |
| sean-k-mooney | offload might just bee the delete but i think the code path also diverges based on BFV vs image_type=rbd vs images_type=qcow | 17:43 |
| opendevreview | chanyeol yoon proposed openstack/nova master: Fix infinite WARNING loop in _reclaim_queued_deletes https://review.opendev.org/c/openstack/nova/+/985068 | 23:35 |
Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!