| opendevreview | Bence Romsics proposed openstack/nova master: Functional reproducer for #2051685 https://review.opendev.org/c/openstack/nova/+/977331 | 10:22 |
|---|---|---|
| opendevreview | Bence Romsics proposed openstack/nova master: Suppress error in cleanup of pre live migration https://review.opendev.org/c/openstack/nova/+/982085 | 10:22 |
| carloss | Uggla: o/ hey there - considering the efforts on cold migration for VirtioFS and the future feature plans, I was wondering if we could have another cross team session in the upcoming PTG? | 10:54 |
| opendevreview | ribaudr proposed openstack/nova master: Avoid a CellDatabases fallback lookup deadlock https://review.opendev.org/c/openstack/nova/+/1006935 | 12:00 |
| gibi | cores can I get a second set of eyes on https://review.opendev.org/q/topic:%22bug/2166786%22 this is a somewhat high prio bug fix for me. Sylvain already +2d it. | 12:56 |
| dansmith | gibi: I had this open yesterday and got distracted | 14:07 |
| dansmith | the two regression tests are for regular PCI and PCI-in-placement but otherwise the same right? | 14:13 |
| gibi | right | 14:14 |
| dansmith | oh yeah? I use gpt 5.3 codex for some personal stuff and it seems okay, maybe a little less than claude and opus 4.6.. but haven't tried anything remotely close to redesigning live migration of course | 14:20 |
| dansmith | oops :P | 14:20 |
| gibi | now the truth is out :P | 14:22 |
| dansmith | shame! | 14:22 |
| * gibi thinking about what would be an appropriate penance | 14:23 | |
| dansmith | I'm already reviewing your patch I can't imagine what else you'd want more | 14:23 |
| gibi | good point. I take it! | 14:24 |
| dansmith | :P | 14:24 |
| bauzas | I'm now myself using Cursor Grok 4.6, oops :) | 14:24 |
| dansmith | far more shameful, so I feel better ;P | 14:25 |
| bauzas | .... and I have a Tesla, wooops | 14:28 |
| gibi | does Grok bundled with Tesla these days? | 14:30 |
| dansmith | gibi: left a question about cross-cell cold migration on the last patch | 14:30 |
| dansmith | I really hate having to add another RPC call to fix this, even if that RPC call already exists | 14:30 |
| gibi | dansmith: yeah I somehow forsee that. But asked bauzas first to get some reassurance :) I will check your question soon | 14:33 |
| gibi | I did not considered cross cell stuff so I'm glad you reviewed it | 14:34 |
| bauzas | ah indeed I also didn't thought about cross-cell | 14:35 |
| dansmith | I didn't go look in detail | 14:35 |
| dansmith | but the use of an RPC to solve this aroused my curiosity given my current mindset :) | 14:36 |
| gibi | OK I will track down the cross cell implications | 14:38 |
| gibi | I replied to you in the second patch doing the periodic cleanup | 14:39 |
| gibi | if we can land that that alone will help managing the situation | 14:39 |
| gibi | regardless if the RPC based fix can or cannot be landed | 14:39 |
| dansmith | okay yeah, I was thinking of asking what is the impact of just doing this with periodic.. just a delayed freeing of some PCI resources after an infra-level migration failure.. not too bad | 14:44 |
| gibi | yes exactly | 14:46 |
| gibi | it just clears it up periodically | 14:46 |
| gibi | and if that period is long then it might be visible from the outside that we leak | 14:46 |
| gibi | there is a tradeoff on that periodic frequency | 14:46 |
| gibi | but I accept if we only go with the periodic right now | 14:47 |
| gibi | or say we only land the RPC stuf on master but not backporting it (pending cross cell clearings) | 14:48 |
| bauzas | I'm looking at dansmith's questioin | 14:56 |
| bauzas | because when thinking at the support enveloppe for upgrades, I thought we were fine given the very old API call | 14:56 |
| bauzas | so I guess this is a matter of non-supported routine for calling another cell ? | 14:56 |
| gibi | we cannot call other cells there is no MQ for that | 14:59 |
| gibi | so I have to check if in cross cell migration we can hit the same issue or not | 14:59 |
| gibi | and then if we can then say something about the fix | 14:59 |
| bauzas | ok, that's the potential concern I was thinking | 14:59 |
| gibi | and in the meantime see if that the added call is a problem or not for cross cell migration, probably not but I have to look | 14:59 |
| dansmith | the added call would of course be a problem for cross-cell migration.. the question I think is whether or not you can cross-cell cold migrate with PCI devices at all | 15:07 |
| dansmith | and if you can, if you hit the same problem | 15:07 |
| dansmith | if so, perhaps conductor can do the extra call I guess | 15:07 |
| bauzas | gibi: I'm burned, toasted, dead even. Permission to skip eventlet removal meeting | 15:20 |
| gibi | sure | 15:24 |
| Uggla | Sorry late notice | 15:28 |
| Uggla | Upstream meeting starts in ~2mn (https://meet.google.com/zjr-rxus-hzj?authuser=1&hs=122) | 15:28 |
| gmaan | gibi: dansmith: what you say on this? should we merge and backport this releasenotes updates for leaked task or it si ok as it is? https://review.opendev.org/c/openstack/nova/+/1006095 | 16:36 |
| opendevreview | Merged openstack/nova master: [cover]stabilize test_migrate_from_disabled_host https://review.opendev.org/c/openstack/nova/+/998925 | 18:24 |
| opendevreview | Ghanshyam Maan proposed openstack/nova master: Reduce the unnecessary executors shutdown logs in test jobs https://review.opendev.org/c/openstack/nova/+/1004160 | 21:25 |
Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!