| opendevreview | Hiroki Narukawa proposed openstack/nova master: libvirt: retry libvirt connection on live_migration_monitor https://review.opendev.org/c/openstack/nova/+/867077 | 06:07 |
|---|---|---|
| opendevreview | Hiroki Narukawa proposed openstack/nova master: Reproducer for bug 1999607 https://review.opendev.org/c/openstack/nova/+/877582 | 06:07 |
| opendevreview | Hiroki Narukawa proposed openstack/nova master: Reproducer for bug 1999607 https://review.opendev.org/c/openstack/nova/+/877582 | 06:39 |
| opendevreview | Hiroki Narukawa proposed openstack/nova master: libvirt: retry libvirt connection on live_migration_monitor https://review.opendev.org/c/openstack/nova/+/867077 | 06:39 |
| opendevreview | Hiroki Narukawa proposed openstack/nova master: Reproducer for bug 1999607 https://review.opendev.org/c/openstack/nova/+/877582 | 07:22 |
| opendevreview | Hiroki Narukawa proposed openstack/nova master: libvirt: retry libvirt connection on live_migration_monitor https://review.opendev.org/c/openstack/nova/+/867077 | 07:22 |
| opendevreview | Amit Uniyal proposed openstack/nova master: Allow swap resize from non-zero to zero https://review.opendev.org/c/openstack/nova/+/857339 | 07:58 |
| *** gibi_pto is now known as gibi | 08:04 | |
| opendevreview | Amit Uniyal proposed openstack/nova master: Allow swap resize from non-zero to zero https://review.opendev.org/c/openstack/nova/+/857339 | 08:04 |
| opendevreview | Alexey Stupnikov proposed openstack/nova master: Preserve cached base images for failed resize ops https://review.opendev.org/c/openstack/nova/+/877410 | 08:23 |
| *** thelounge554 is now known as thelounge55 | 08:37 | |
| *** elodilles_pto is now known as elodilles | 09:26 | |
| opendevreview | Danylo Vodopianov proposed openstack/nova-specs master: VirtIO PackedRing Configuration support https://review.opendev.org/c/openstack/nova-specs/+/868377 | 09:30 |
| *** EugenMayer46 is now known as EugenMayer4 | 10:02 | |
| songwenping | sean-k-mooney: hi, i have two same gpus on one host and want to passthrough two vms, my passthrough_whitelist is set {'vendor_id':'10de','product_id':'1e04'} and alias is set {"vendor_id":"10de","product_id" : "1e04","name": "gpu"}, but i can only create one vm, the other failed with no valid host filetered by pcipassthroughfilter. so how can i make the configuration? | 10:20 |
| bauzas | songwenping: which version and what says the filter ? | 10:31 |
| songwenping | Rocky and PciPassthroughFilter: (start: 1, end: 0) | 10:33 |
| sean-k-mooney | that should work | 10:34 |
| bauzas | you would need DEBUG levels | 10:34 |
| sean-k-mooney | songwenping: can you check the pci_devices table in the nova cell db | 10:35 |
| sean-k-mooney | and make sur eboth are there | 10:35 |
| sean-k-mooney | songwenping: you alos need to make sure xorg (if its installed) is not using one of the gpus | 10:35 |
| sean-k-mooney | i.e. make sure they are both bound to vfio_pci | 10:35 |
| songwenping | pci_devices table has two gpu record. | 10:35 |
| sean-k-mooney | ok | 10:36 |
| sean-k-mooney | and one is aviable | 10:36 |
| sean-k-mooney | and the other asigned to the first vm | 10:36 |
| bauzas | sean-k-mooney: that's a scheduler filter issue | 10:36 |
| bauzas | so not related to the driver | 10:36 |
| sean-k-mooney | ya it is but i dont think its a bug | 10:36 |
| bauzas | yup | 10:37 |
| sean-k-mooney | bauzas: right i jsut mentioned the driver for after this is fixed | 10:37 |
| bauzas | anyway, I need to get off the keyboard | 10:37 |
| songwenping | i am opening the debug level to get more info | 10:37 |
| sean-k-mooney | songwenping: are you useing numa by the way | 10:37 |
| sean-k-mooney | i.e. cpu pinning or hugepages | 10:37 |
| songwenping | no using numa | 10:37 |
| sean-k-mooney | ok | 10:37 |
| sean-k-mooney | if you can pastebin the entires for the pci_devices and debug secduler logs we can take a look | 10:38 |
| sean-k-mooney | i just want to look at the 2 for the gpu | 10:39 |
| sean-k-mooney | its odd that one worked and the other is failing | 10:39 |
| sean-k-mooney | i would have expect both to fail or both to work | 10:39 |
| sean-k-mooney | im wonderign what device_type is set to i assum type-PCI but it could be type-PF depending on the type of gpu | 10:40 |
| songwenping | sean-k-mooney,bauzas: that's becuase the disk is not enough, i use the local disk. thanks a lot. | 10:55 |
| sean-k-mooney | huh the pci filter shoudl not be | 10:56 |
| sean-k-mooney | oh the disk filter elimiated the host that had the free pci device | 10:56 |
| sean-k-mooney | and then the pci filter elimiated the remaining host because there was not a free gpu there | 10:56 |
| sean-k-mooney | ok makes sense | 10:56 |
| songwenping | yes, the filter error is mislead. | 10:57 |
| sean-k-mooney | ya you bacially need to look at the set of all filters when debugging it | 10:57 |
| sean-k-mooney | not just the last one that elimiated the final hosts | 10:57 |
| songwenping | yeah the debug info is very useful. | 10:58 |
| sean-k-mooney | some operators run with dbug on all the time for what its worth | 10:58 |
| sean-k-mooney | you might now want it enabeld on all the services | 10:59 |
| sean-k-mooney | but i like having it on for the scheduler at least | 10:59 |
| bauzas | ok gtk | 10:59 |
| sean-k-mooney | there is an argument to be made for moving those specific logs to info level | 10:59 |
| sean-k-mooney | we didnt in the past because it can be a bit noisy but its a trade off | 11:00 |
| Uggla | sean-k-mooney, gibi, bauzas Hi, is the a way to disable a flag set with self.flags(...) ? | 11:11 |
| sean-k-mooney | Uggla: well you can just set it to what you want but if you mean you want to revert it to the unset state then you likely need to do thave via oslo | 11:15 |
| sean-k-mooney | https://docs.openstack.org/oslo.config/latest/reference/locations.html | 11:16 |
| sean-k-mooney | so flags i think is calling set_override | 11:16 |
| sean-k-mooney | https://docs.openstack.org/oslo.config/latest/reference/api/oslo_config.html#oslo_config.cfg.ConfigOpts.reset | 11:17 |
| sean-k-mooney | there is a reset funciton | 11:17 |
| sean-k-mooney | so you would do CONF.my_section.my_var.reset() | 11:17 |
| Uggla | sean-k-mooney, cool I'm going to have a look. Thanks | 11:17 |
| sean-k-mooney | that will clear any orverdie ro default value | 11:17 |
| sean-k-mooney | reset()ΒΆ | 11:18 |
| sean-k-mooney | Clear the object state and unset overrides and defaults. | 11:18 |
| sean-k-mooney | Uggla: feel free to wrap that up in a reset_flags helper funciton if you want to use it | 11:19 |
| sean-k-mooney | i have never needed to od this in a test before so before you do are you sure you need this | 11:20 |
| sean-k-mooney | usually i woudl sugges makeign sure the parent of your test class does not do thing you dont need | 11:20 |
| sean-k-mooney | instead of trying to reset teh config state | 11:21 |
| Uggla | sean-k-mooney, I'm not sure I try to have a src node configure with a flag and a dest node without. But it seams that flags are 'global' in functional test. | 11:23 |
| sean-k-mooney | oh you cant do that | 11:24 |
| sean-k-mooney | its a limiation of our test infra | 11:24 |
| sean-k-mooney | there is a way to make it work with context manger which i demonstarted a few years ago but we decided not to do that | 11:24 |
| sean-k-mooney | if your doing this in a functional test you will need to mock it differntly | 11:25 |
| Uggla | sean-k-mooney, the idea is to simulate a live migration from src node with a flag to dst node without this flag | 11:25 |
| sean-k-mooney | yes you cant do that today in our func tests without a lot of hacks | 11:26 |
| Uggla | sean-k-mooney, yes I manage to do what I wanted with mock. But I was wondering if something simpler can be used. | 11:29 |
| Uggla | Is the reset proposed above a potential way to do that ? | 11:29 |
| sean-k-mooney | not really | 11:30 |
| sean-k-mooney | when you reset it it will reset it for both computes | 11:30 |
| Uggla | sean-k-mooney, in my case I think it will be ok, I'll try | 11:31 |
| sean-k-mooney | it likely will be racy | 11:32 |
| sean-k-mooney | basicaly this is something we currently try not to do | 11:32 |
| sean-k-mooney | so if you get ti workign we will have to review it very carefully | 11:32 |
| sean-k-mooney | Uggla: https://stackoverflow.com/questions/33504543/passing-import-context-variables-to-decorators/33507308#33507308 | 11:38 |
| sean-k-mooney | thats not quite what you want but i had an example for this in nova somewhere in gerrit with context manager but i cant find it | 11:42 |
| opendevreview | Alexey Stupnikov proposed openstack/nova stable/victoria: Cleanup old resize instances dir before resize https://review.opendev.org/c/openstack/nova/+/864730 | 13:02 |
| opendevreview | Amit Uniyal proposed openstack/nova master: Added context manager for instance lock https://review.opendev.org/c/openstack/nova/+/873648 | 13:33 |
| opendevreview | Amit Uniyal proposed openstack/nova master: Disconnecting volume from the compute host https://review.opendev.org/c/openstack/nova/+/877446 | 13:33 |
| opendevreview | Amit Uniyal proposed openstack/nova master: Disconnecting volume from the compute host https://review.opendev.org/c/openstack/nova/+/877446 | 13:37 |
| opendevreview | Amit Uniyal proposed openstack/nova master: Allow swap resize from non-zero to zero https://review.opendev.org/c/openstack/nova/+/857339 | 17:32 |
| darkhorse | hi team, i noticed that the periodic tasks defined with @periodic_task decorator at scheduler/manager.py performs tasks on all workers instead of running once at defined spacing. is that a bug or a designed behavior? | 18:12 |
| darkhorse | hi team, i noticed that the periodic tasks defined with @periodic_task decorator at scheduler/manager.py performs tasks on all workers instead of running once at defined spacing. is that a bug or a designed behavior? | 18:12 |
| dansmith | darkhorse: which periodic are you looking at? but yes, it's *expected* and the periodic should use some locking to prevent it from running simultaneously if need-be | 18:24 |
| darkhorse | dansmith: thank you for clarification. i have added a custom periodic task that should run every minute. will you point me to an example code that uses locking? | 18:26 |
| dansmith | darkhorse: oslo lockutils :) | 18:27 |
| dansmith | with external=true so it uses a lock file, which all the others will see | 18:27 |
| dansmith | if not external, then it'd just be per-process, which wouldn't help | 18:27 |
| darkhorse | ok thank you! | 18:29 |
| *** artom_ is now known as artom | 21:15 | |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!