Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2018-07-18
12:51:48 sean-k-mooney ill create a script that works and propose it to both os-vif and os-traits
12:52:26 openstackgerrit sahid proposed openstack/nova master: hardware: fix memory check usage for small/large pages https://review.openstack.org/532168
12:54:29 jaypipes sean-k-mooney: I shall let you and stephenfin fight that out in a death match.
12:55:42 sean-k-mooney hehe well what stephenfin has will work. until the point that os-traits gets added to upperconstraints
12:56:17 sean-k-mooney its exactly how nova works but nova is not a lib
12:59:08 openstackgerrit Matt Riedemann proposed openstack/nova stable/queens: Handle HostMappingNotFound when deleting a compute service https://review.openstack.org/583566
13:00:07 do3meli someone able to review this one here: https://review.openstack.org/#/c/578040/ ? thanks
13:01:00 openstackgerrit Matt Riedemann proposed openstack/nova stable/pike: Handle HostMappingNotFound when deleting a compute service https://review.openstack.org/583567
13:27:55 JulienCornuwel Hi, since the deprecation of nova hooks, I wonder what would be the correct way to handle my use case: I'm doing pci_passthrough of NVME drives to instances, and I want to wipe them clean when the customer destroys his instance. I can't use qemu hooks since they are run at stop/start and I don't want to wipe the drive on a simple reboot. Any pointers?
13:34:26 gibi JulienCornuwel: you can wait for instance.delete.end notifications from nova https://docs.openstack.org/nova/latest/reference/notifications.html
13:35:26 pooja_jadhav dansmith: Hello
13:37:01 JulienCornuwel gibi: Is there a way to prevent nova from giving the drive to another customer until I'm done with the wiping? (takes about 30-40 seconds) I've done some tests, and if I run a nova boot during the wipe, nova just takes the device wether I'm done cleaning it or not…
13:38:28 efried do3meli: Reviewed
13:38:28 do3meli thx :efried. will work through it now
13:38:41 efried JulienCornuwel: Are you looking to write code to do this, or are you trying to do it under the existing codebase?
13:38:50 gibi JulienCornuwel: that would require a way to reserve the given PCI device while you are wipeing it
13:39:00 stephenfin sean-k-mooney: I saw that tox install script in a couple of oslo repos but not all of them. I thought mordred or someone had worked around this now, tbh
13:39:32 efried pooja_jadhav: I think Dan is on vacation this week.
13:40:19 pooja_jadhav efried: Ohh.. thanks for update. I want to discuss about the patch https://review.openstack.org/#/c/580720/
13:40:33 mordred stephenfin: we still have tos intall scripts laying around?
13:40:54 stephenfin mordred: aye, lemme root some out
13:40:58 sean-k-mooney stephenfin: maybe but when i made the same change to os-vif it conmplained
13:41:05 openstackgerrit Matt Riedemann proposed openstack/python-novaclient master: Microversion 2.64 - Use new format policy in server group https://review.openstack.org/578261
13:41:08 mgariepy hello, does "nova-manage cell_v2 map_instances" needs to be run on every upgrade ? It ran on my upgrade from newton to ocata and on the ocata to pike taking about 40 minutes each time.
13:41:33 efried pooja_jadhav: Maybe someone here can help. What do you need?
13:41:54 JulienCornuwel efried: We could write some code. But I was wondering if there was already a way that I was missing.
13:42:09 stephenfin mordred: I see oslo.log, oslo.db, and oslo.cache locally
13:42:20 JulienCornuwel gibi: exactly, is there a way to do that.
13:42:28 stephenfin though there could be more that I don't have local copies of
13:42:48 sean-k-mooney stephenfin: as far as i am aware all libs need to have the script
13:43:00 efried JulienCornuwel: I can't think of anything. If you're writing code, you could do as gibi suggests and, in response to the notification, remove the device from inventory reporting (get_available_resource) until you're done with it.
13:43:04 mordred stephenfin: wow. just when I thought we'd gotten rid of all of those :)
13:43:14 gibi JulienCornuwel: I don't think we have current a way to reserver a PCI dev
13:44:13 pooja_jadhav efried: Based on above patch https://review.openstack.org/#/c/580720/ submitted by Dan, When we boot an instance from volume then it should not consider local disk_gb. But I have applied patch and tried to tested the same but i can that local disk gb is getting used by seeing nova hypervisor-stats.
13:44:34 sean-k-mooney stephenfin: mordred this is what i get with the same change to os-vif
13:44:36 sean-k-mooney http://paste.openstack.org/show/726197/
13:44:59 efried pooja_jadhav: libvirt, I assume?
13:45:01 pooja_jadhav see*
13:45:11 stephenfin sean-k-mooney: Stop overriding install_command
13:45:12 pooja_jadhav efried: yes
13:45:59 stephenfin sean-k-mooney: It should simply read 'install_command = pip install {opts} {packages}' (or possibly it can be unset - I'm not sure). mordred can probably explain why that breaks
13:46:08 sean-k-mooney stephenfin: this is what i have install_command = pip install -U -c {env:UPPER_CONSTRAINTS_FILE:https://git.openstack.org/cgit/openstack/requirements/plain/upper-constraints.txt} {opts} {packages}
13:46:19 stephenfin sean-k-mooney: Yup, see above
13:46:30 JulienCornuwel efried: gibi: Thanks for the confirmation. We'll start working on a spec and thinking of a way to do that.
13:46:45 efried pooja_jadhav: So the instance is actually booting from volume - i.e. truly not using the local disk - but nova hypervisor-stats is showing that capacity as having been applied from the local disk?
13:46:50 sean-k-mooney oh you put the constatit in deps
13:46:54 tssurya mgariepy: well the aim of map_instances is to just map the instances to the cells, so if you have done it once, and you are sure all instances are mapped, I don't see any reason to keep running it, unless you have new ones to populate
13:47:08 stephenfin sean-k-mooney: Ah, I get it. You don't want to install your local package with constraints but you _do_ want to install your test requirements
13:47:14 stephenfin ...with them
13:47:15 mgariepy tssurya, https://github.com/openstack/openstack-ansible-os_nova/blob/stable/pike/tasks/nova_db_setup.yml#L58-L90
13:47:18 stephenfin sean-k-mooney: yuuup
13:47:47 efried JulienCornuwel: Note that things involving the existing PCI passthrough subsystem are going to be difficult to get approval for, because we're trying to get rid of that whole thing.
13:47:54 sean-k-mooney stephenfin: yep
13:47:56 mgariepy so if the cell_v2 mapping was done in the upgrade, (and the create_cell returns 2, it would be right to skip the map_instances
13:48:39 JulienCornuwel efried: In favor of? Cyborg?
13:48:39 efried JulienCornuwel: That said, I think you could do what you want without really hitting that code; and in such a way that it would be transferrable to placement-based device passthrough in the future.
13:48:58 pooja_jadhav efried: After looking at the code changes done by Dan, its not taking disk_gb resource. When allocation candidates returns the response.
13:49:07 efried JulienCornuwel: Well, yes, but more generally placement.
13:49:17 mordred stephenfin, sean-k-mooney the trick is putting -c{env:UPPER_CONSTRAINTS_FILE:https://git.openstack.org/cgit/openstack/requirements/plain/upper-constraints.txt} into the deps list
13:49:19 mordred not the install_command
13:49:40 tssurya mgariepy: yes exactly
13:49:41 mordred like http://git.openstack.org/cgit/openstack/openstacksdk/tree/tox.ini
13:49:41 sean-k-mooney mordred: ya that is sneaky i like it
13:49:50 mgariepy thanks :)
13:50:06 mordred sean-k-mooney: it's the necessary trick for the self-constraint problem ;)
13:50:07 tssurya mgariepy: no problem
13:50:10 efried pooja_jadhav: It sounds like you should downvote the patch (it needs a rebase anyway, so you won't be the only thing holding it up) and explain what you're seeing.
13:50:33 efried pooja_jadhav: What would be even better is if you could come up with a functional test that demonstrates the behavior.
13:50:40 mordred tox installs the deps line separate from the current project - but it uses install_command for both things
13:50:52 sean-k-mooney mordred: it should also work on cygwin which the install script version did not
13:50:57 mordred so by putting constraints in deps, it winds up on the command line for the deps - but not for the install of the current project
13:51:01 mordred sean-k-mooney: yup
13:51:05 pooja_jadhav efried: But at https://github.com/openstack/nova/blob/master/nova/objects/build_request.py#L64 I can see in this db_instance the root_gb=1 value is going.
13:51:29 mordred sean-k-mooney: there are many reasons we don't like those install script wrappers - we forget to communicate all of them :)
13:52:07 pooja_jadhav efried: ok
13:52:28 openstackgerrit Takashi NATSUME proposed openstack/nova master: Transform volume.usage notification https://review.openstack.org/580345
13:53:06 JulienCornuwel efried: thanks, we'll look into that
13:53:12 sean-k-mooney mordred: well os-vif never used the install script because it broke my cygwin workflow at the time.
13:53:31 tssurya jaypipes: you appear to be the only spec core online now; whenever you get the time could you please have a look at https://review.openstack.org/#/c/581243/ ? It a very easy not fix follow-up on a merged spec
13:53:54 tssurya nit*
13:54:36 mordred sean-k-mooney: \o/
13:57:11 openstackgerrit Takashi NATSUME proposed openstack/nova master: Transform volume.usage notification https://review.openstack.org/580345
13:59:23 efried JulienCornuwel: To be a bit more specific: your notification listener would conceivably start the cleanup process, and also register the device in some kind of cache ("reserved devices"). Add logic to your get_available_resource method to look at that cache and skip reporting any devices therein.
13:59:23 efried So when we make the move to placement-based device reporting, the latter logic would go into update_provider_tree instead of get_available_resource, but everything else would stay the same.
13:59:23 efried There may still be a race in this setup, because there may be a delay before get_available_resource/update_provider_tree is run. I suppose you could get around this by making your spawn method add the device to the "reserved" list, and then have your notification listener thingy only be responsible for removing it when done.
13:59:38 openstackgerrit sean mooney proposed openstack/os-vif master: add upper_constraitns support https://review.openstack.org/583582
14:00:12 sean-k-mooney we should backport ^
14:01:48 sean-k-mooney ok going to grab lunch brb
14:02:09 openstackgerrit do3meli proposed openstack/nova master: docs: add nova host-evacuate command to evacuate documentation https://review.openstack.org/578040
14:06:12 efried sean-k-mooney: We should spell it right, though.
14:08:01 stephenfin efried: It's the main reason we keep jaypipes around, tbh
14:08:03 sean-k-mooney dambit hehe ill fix that when i get back
14:08:16 efried sean-k-mooney: I'll push the edit right quick.
14:08:39 openstackgerrit Eric Fried proposed openstack/os-vif master: add upper_constraints support https://review.openstack.org/583582
14:08:43 stephenfin spellchecker extraordinaire
14:09:02 efried sean-k-mooney: +2
14:09:18 openstackgerrit Matt Riedemann proposed openstack/os-vif master: add upper_constraints support https://review.openstack.org/583582
14:09:33 efried mriedem: wtf, yo
14:10:23 stephenfin efried: It's alright - we'd missed anyway
14:10:59 sean-k-mooney mriedem: yes you are right ill fix those

Earlier   Later