Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2023-06-09
13:17:50 bauzas Uggla_: sean-k-mooney: I need to review the scaphandre patches tbh, but I don't have any problem to use a specific local share type
13:17:59 bauzas so we could persist it
13:23:07 Uggla_ sean-k-mooney, bauzas to be honest I thought that was a not so bad approach as 90% of the code is the same except that the mountpoint on the compute does not come from manila.
13:23:38 Uggla_ but let's wait sean-k-mooney to be back.
13:41:20 dansmith I haven't reviewed them, but if the idea is to create share objects for the local shares as if they were manila shares, I agree that seems wrong
13:42:34 sean-k-mooney that is the current approch
13:43:36 sean-k-mooney if we were to have a local share type and model it tha tway i would expect use to provde a way for end users to attach/detach them and have nova create a folder dynmicaly to hold the content and copy the content form one host to anohter on cold migration
13:43:52 sean-k-mooney the share attchment api as propsoed is not desginse to hanel that
13:44:40 sean-k-mooney for scaphandre there is nor requiremtn to copy data on cold migration it doesnt even make sense in that specific case
13:44:50 dansmith presumably it means the user sees that local share in the shares api, and (hopefully) is barred from deleting it
13:44:51 sean-k-mooney the share api has no way to speicf a capasity
13:45:34 sean-k-mooney dansmith: ya so we woudl not want it to be listed or deletable as that provides a second way to configure scaphandre mounts
13:45:43 dansmith right
13:46:09 sean-k-mooney the only thingn the manilla shares feature and scaphandre have in common is they both use virtio fs
13:46:15 dansmith right
13:46:38 dansmith so I would treat these local shares as "do whatever the conf says" that we handle when we're building the xml on reboot or whatever
13:46:45 sean-k-mooney the flavor extra spec/image porpety is the only metadta reiqured ot know if the feature is used
13:46:48 dansmith and not represent them in the database, api, etc
13:46:55 dansmith more like configdrive than anything else
13:47:02 sean-k-mooney yes
13:47:08 sean-k-mooney configdrive is a good example
13:49:18 sean-k-mooney by the way i can see usecasue for being able to have local storage on a host exposed to the guest as virtio-fs passthough. i just dont htinke the share attachemt feature is what we should use to build that
13:49:54 dansmith you mean more use cases than just scaphandre right? I definitely can
13:50:02 sean-k-mooney that is its own seperate feature
13:50:06 dansmith but not user-controllable and not visible like an attachment
13:50:45 sean-k-mooney well i coudl see it being user contoleable but there is a quota aspect that woudl have to be adress
13:51:08 dansmith when would we ever want to allow the user to request "show me /etc on the host" ?
13:51:24 sean-k-mooney that not what i mean
13:51:56 sean-k-mooney i mean we coudl allow the user to have addtional storage form a folder we create on the host like ephemeal storage
13:52:19 dansmith oh sure, right
13:52:30 dansmith not sure we should expose that as a share in the same way as manila though
13:52:32 sean-k-mooney i.e. a way to add addtional sotrage without having to do it via a resize
13:52:48 sean-k-mooney ya its a seperate thing
13:52:52 dansmith agree
13:53:11 dansmith I was just typing out "we might want to expose ephemeral disks as filesystems instead" before you said it
13:53:25 dansmith so yeah, I can definitely see that, but I think that's a sort of third use case for this
13:53:46 sean-k-mooney its more "As a user i would like to be ablt to attch addtion local storage to an instance without needign to resize"
13:53:50 dansmith first being manila, second being "the admin wants this configdrive-like thing inserted into me", and third epehemeral fs space
13:54:00 dansmith well, resize or mkfs but yeah
13:54:39 sean-k-mooney honestly the motivation for ti is pretty low until you can do it while the vm is running
13:55:53 dansmith I can also imagine people wanting two instances with hard affinity to share a fast local disk as a filesystem
13:56:01 dansmith i.e. db-and-http workers and things
13:56:11 sean-k-mooney anyway Uggla_ i have concerns with the current apptoch. i have not done a full review of the patchs but i thing directionally we shoudl try and only share the virtiofs xml generation btween these two features
13:56:36 dansmith but the local share thing should really be a tack-on at the end of this right?
13:56:59 dansmith the set is extremely long right now so just breaking that stuff off the end for now would make sense to me and focus on getting the manila bit working as the primary goal
13:57:02 sean-k-mooney well i atcully dont think we shoudl do the local share as part of this work this cyle
13:58:07 sean-k-mooney i was ortgnallly hoping that the virtiofs xml change woudl be first then scaphandre abd mainila could progess in parallel
13:58:45 sean-k-mooney but i entrily expect us to be able to merge scaphandre support without any of the manilla supprot being merge outside the virtiofs part in the libvirt driver
13:59:20 dansmith yes I would think so too, but you're also saying do that after manila (now) right?
13:59:32 sean-k-mooney not really
13:59:43 dansmith [06:57:02] <sean-k-mooney> well i atcully dont think we shoudl do the local share as part of this work this cyle
13:59:51 dansmith what does that mean then?
14:00:06 sean-k-mooney the intoduction of a local share type
14:00:28 sean-k-mooney if we were to have that i would prefer to defer that to the other uscase of epmperal fs or simialr
14:00:30 dansmith the scaphandre case is just a special configuration of a local share no?
14:00:54 sean-k-mooney im conlifcted on how to respond to that
14:01:20 sean-k-mooney i want to say no because i dont think scaphandre's use of virtio-fs shoudl me modeled as a share in the db
14:01:23 dansmith by "local share" I'm meaning the "operator has configured nova to always insert this share into all guests on this host"
14:01:28 sean-k-mooney so its a special case fo virtio-fs
14:01:36 dansmith maybe we need to define some terms
14:02:14 sean-k-mooney i would cinsider it a special case foinsterting this "filesystem path" into all guest that request it
14:02:44 dansmith okay I'm a bit confused then
14:02:44 sean-k-mooney i would reall like to keep "share" to refer only to manialla shares
14:02:52 dansmith I agree with that
14:03:27 sean-k-mooney if we want to allow arbairy addtional filesystem to be provided to a guest via an api that woudl be a new filesystem attach api
14:03:48 sean-k-mooney for scaphandre as you said i think config drive is a better mental model
14:04:15 sean-k-mooney we have an image property/extra spec that act as a request for it and a trait on host that supprot it
14:04:38 dansmith how about a 15 minute meet on the topic real quick?
14:04:42 sean-k-mooney there is no requirement for modileign it in the db or as object or in the api beyond the image property/extra spec
14:04:45 dansmith Uggla_: ?
14:06:07 sean-k-mooney sure i can do that
14:06:27 dansmith meet.google.com/duc-bfai-mrv
14:07:47 sean-k-mooney anyone who is interested is welcome to join by the way ^
14:07:54 dansmith yup
14:14:21 Uggla_ dansmith, sean-k-mooney sorry was out to get Rose, reading
14:18:52 bauzas dansmith: sean-k-mooney: Uggla_ : doh, I need to go to get my daughter in 5 mins
#openstack-nova - 2023-06-11
16:19:50 opendevreview Amit Uniyal proposed openstack/nova master: WIP: Refactor CinderFixture https://review.opendev.org/c/openstack/nova/+/885756
16:19:50 opendevreview Amit Uniyal proposed openstack/nova master: Reproducer for dangling bdms https://review.opendev.org/c/openstack/nova/+/881457
16:19:51 opendevreview Amit Uniyal proposed openstack/nova master: Delete dangling bdms https://review.opendev.org/c/openstack/nova/+/882284
#openstack-nova - 2023-06-12
06:47:53 opendevreview Amit Uniyal proposed openstack/nova master: Reproducer for dangling bdms https://review.opendev.org/c/openstack/nova/+/881457
06:47:54 opendevreview Amit Uniyal proposed openstack/nova master: Delete dangling bdms https://review.opendev.org/c/openstack/nova/+/882284
07:38:16 opendevreview Amit Uniyal proposed openstack/nova master: Reproducer for dangling bdms https://review.opendev.org/c/openstack/nova/+/881457
07:38:17 opendevreview Amit Uniyal proposed openstack/nova master: Delete dangling bdms https://review.opendev.org/c/openstack/nova/+/882284
09:10:01 opendevreview Sofia Enriquez proposed openstack/nova-specs master: Implement support for LUKS inside qcow2 volumes when using Libvirt https://review.opendev.org/c/openstack/nova-specs/+/883516
09:34:46 opendevreview Sofia Enriquez proposed openstack/nova-specs master: Implement support for LUKS inside qcow2 volumes when using Libvirt https://review.opendev.org/c/openstack/nova-specs/+/883516
11:12:49 opendevreview Amit Uniyal proposed openstack/nova master: Reproducer for dangling bdms https://review.opendev.org/c/openstack/nova/+/881457
11:12:49 opendevreview Amit Uniyal proposed openstack/nova master: Delete dangling bdms https://review.opendev.org/c/openstack/nova/+/882284
16:42:25 opendevreview Merged openstack/nova-specs master: Add compute-object-ids spec for 2023.2 https://review.opendev.org/c/openstack/nova-specs/+/877291
#openstack-nova - 2023-06-13
08:14:13 opendevreview Fabian Wiesel proposed openstack/nova master: Add more password generation options https://review.opendev.org/c/openstack/nova/+/865669
15:33:58 manuvakery1 Hi, Few of my VM live migration never completes, it always get stuck at the end of memory copy , going from 1% to 0 and then to 1%. I tried to set the following params
15:33:58 manuvakery1 live_migration_downtime: 500
15:33:58 manuvakery1 live_migration_downtime_steps: 10
15:33:58 manuvakery1 live_migration_downtime_delay: 50, Still its not helping
15:34:32 manuvakery1 I see its sets downtime only once to 50ms
15:35:04 manuvakery1 Any help on this
15:36:38 manuvakery1 2023-06-13 15:01:20.336 3223605 INFO nova.virt.libvirt.migration [req-42a1d50d-1411-416e-8084-cf929cf892f9 - - - - -] [instance: 09658040-2af3-4fbf-884f-601a4d1f2ce2] Increasing downtime to 50 ms after 0 sec elapsed time
15:37:02 frickler manuvakery1: you may wnat to look into auto-completion or post-copy https://docs.openstack.org/nova/yoga/admin/configuring-migrations.html
15:38:18 manuvakery1 frickler: Sure i will .. but based on the configs I set it should increase the downtime to 100ms and then 150ms until it reaches 500ms
18:47:42 stephenfin sean-k-mooney[m]: melwitt: Would you folks be able to look at https://review.opendev.org/c/openstack/placement/+/880623/
18:48:27 stephenfin (Context is we discussed bumping SQLAlchemy to 2.x at a TC forum session this morning and I realised this was proposed but never merged /o\)

Earlier   Later