| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-11-20 | |||
| 19:28:32 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Extract SortContext/SortWrapper of instance_list into utils https://review.openstack.org/518871 | |
| 19:28:32 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Add cross cell sort support for get_migrations https://review.openstack.org/517273 | |
| 19:28:33 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Add pagination and Changes-since filter support for os-migrations. https://review.openstack.org/330406 | |
| 19:36:38 | cfriesen | does the current master branch scheduler handle working as a hybrid scheduler for baremetal as well as VM compute nodes? I'm mostly wondering about things like the NUMATopologyFilter. | |
| 19:37:48 | mriedem | don't think so | |
| 19:37:53 | mriedem | b/c there are different host managers | |
| 19:38:03 | mriedem | HostManager vs IronicHostManager | |
| 19:38:27 | mriedem | although, | |
| 19:38:40 | mriedem | with resource classes in pike, i'm not sure how much you need the IronicHostManager anymore | |
| 19:38:41 | cfriesen | mriedem: yeah...IronicHostManager mostly calls the superclass functions | |
| 19:39:12 | mriedem | and use_baremetal_filters and baremetal_enabled_filters are deprecated | |
| 19:39:38 | mriedem | so, IronicHostManager should probably be deprecated | |
| 19:39:40 | cfriesen | mriedem: we're making it work on Pike and need to modify some of the filters. was just wondering if this was something that I should propose for Rocky as a spec | |
| 19:39:47 | mriedem | since the related config options for the filters are also deprecated | |
| 19:39:52 | cfriesen | or if it's just bugfix material | |
| 19:40:03 | mriedem | spec for what? | |
| 19:40:13 | mriedem | same scheduler serving vm and bm? | |
| 19:40:19 | cfriesen | yes | |
| 19:40:22 | mriedem | i think you get that with custom resource classes | |
| 19:40:34 | cfriesen | but they don't handle the fancy filters, right? | |
| 19:40:47 | cfriesen | so if you enable numatopologyfilter it breaks baremetal | |
| 19:40:49 | mriedem | i don't know how the fancy filters handle ironic nodes | |
| 19:40:57 | dansmith | I didn't think you needed ironic host manager even before placement | |
| 19:41:14 | mriedem | i bet johnthetubaguy would love to talk about this | |
| 19:41:18 | dansmith | like, originally you did but I thought aside from a couple details you were good now | |
| 19:41:31 | mriedem | well, difference is in the filters that are used | |
| 19:41:39 | mriedem | IronicHostManager used the Exact* filters | |
| 19:41:46 | mriedem | and all of those are deprecated since pike | |
| 19:42:19 | dansmith | right, but if you didn't need those filters, then you could use the same host manager as virt | |
| 19:42:29 | cfriesen | dansmith: the docs still say to enable it. there are some differences in how states are reported | |
| 19:42:57 | cfriesen | it looks like IronicHostManager is supposed to work with VM nodes | |
| 19:42:59 | mriedem | so this | |
| 19:42:59 | mriedem | def _locked_consume_from_request(self, spec_obj): | |
| 19:43:00 | mriedem | """Consume nodes entire resources regardless of instance request.""" | |
| 19:43:00 | mriedem | self.free_ram_mb = 0 | |
| 19:43:00 | mriedem | self.free_disk_mb = 0 | |
| 19:43:00 | mriedem | self.vcpus_used = self.vcpus_total | |
| 19:43:02 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Merge flavor extensions controller code https://review.openstack.org/516104 | |
| 19:43:03 | cfriesen | but some of the filters don't | |
| 19:43:06 | dansmith | I don't know all the details, but before placement came and kindof distracted us I thought we were about good to remove that | |
| 19:43:36 | mriedem | the ironic driver will consume all memory_mb/disk_gb/vcpu inventory on a node in placement | |
| 19:43:40 | dansmith | the docs are mostly aimed at just ironic deployments though where there's no reason not to enable the dedicated thing right? | |
| 19:43:47 | mriedem | so that IronicHostState is irrelevant | |
| 19:44:09 | mriedem | *IronicNodeState | |
| 19:44:15 | cfriesen | yeah...there's not a lot of info on how to set up hybrid. I did talk with Julia at the summit about some of this. | |
| 19:45:33 | mriedem | sdague: easy api extension merge-a-roo https://review.openstack.org/#/c/516104/ | |
| 19:45:45 | cfriesen | currently we modified NUMATopologyFilter.host_passes() to just return True for ironic nodes...wondering if this is something to push upstream | |
| 19:46:32 | mriedem | can ironic report numa capabilities? from what i gather, lots of people rely on the CapabilitiesFilter for ironic | |
| 19:46:44 | cfriesen | no idea. :) | |
| 19:46:45 | mriedem | *ComputeCapabilitiesFilter | |
| 19:51:46 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Fix some incorrect option references for scheduler filters https://review.openstack.org/521645 | |
| 19:59:18 | openstackgerrit | Andrey Kurilin proposed openstack/python-novaclient master: [ci] Use pseudo-random names for new resources https://review.openstack.org/479777 | |
| 20:01:18 | openstackgerrit | Dan Smith proposed openstack/nova master: Alternate fix for validating image on rebuild https://review.openstack.org/521186 | |
| 20:14:45 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Deprecate the IronicHostManager https://review.openstack.org/521648 | |
| 20:28:33 | mriedem | sean-k-mooney: i'm confused, is this a nova bp for just work that is all done in os-vif? https://blueprints.launchpad.net/nova/+spec/vif-port-profile | |
| 21:03:03 | efried | jaypipes What is the proper way to talk about a RP's association with an aggregate? Can I say the RP is *in* the aggregate? | |
| 21:03:12 | jaypipes | efried: yes | |
| 21:03:14 | efried | k | |
| 21:03:23 | efried | jaypipes So here's what I'm thinking. | |
| 21:04:24 | efried | jaypipes Today the report client keeps aggregate associations outside of ProviderTree. I believe we need to move what is currently SchedulerReportClient._provider_aggregate_map into ProviderTree. Specifically, _Provider has a set() of aggregate UUIDs. | |
| 21:05:23 | efried | jaypipes Because the virt driver will ultimately be in charge of assigning aggregate associations, so we'll want the same suite of methods on ProviderTree to ask if a given RP is "in" an aggregate, add (or replace) aggregate associations; etc. | |
| 21:06:17 | efried | jaypipes Same thing for traits, which is where that change set of a couple hours ago was coming from - to be followed up by SRC methods to populate/query it. | |
| 21:11:14 | jaypipes | efried: not sure the virt driver will be responsible for associated an aggregate to a provider -- unless the virt driver is also creating those aggregates. Does PowerVM do that? | |
| 21:11:40 | jaypipes | efried: for traits, yeah, totes agree. | |
| 21:12:18 | efried | jaypipes In the only real world case we have right now - shared storage pool - yes. I can't think how else the aggregate would get created. | |
| 21:12:45 | jaypipes | efried: it was originally intended to be created by an admin or an external agent (for example, Cinder...) | |
| 21:14:03 | efried | jaypipes Yeah, I suppose I can see that. In our case, we use SSP for ephemeral, so it would have to be more like the admin. It would just be another step they have to do when they're setting up the host. | |
| 21:14:22 | jaypipes | efried: understood. | |
| 21:14:26 | efried | jaypipes BTW, is there any rule against me using the same UUID for an aggregate and a RP that's in it? | |
| 21:14:42 | efried | I mean, I'm sure you'll say that's not a good idea, but is anything gonna stop me? | |
| 21:15:00 | jaypipes | efried: not a good idea, but nothing will prevent you from doing that. | |
| 21:15:14 | efried | k | |
| 21:18:18 | efried | jaypipes Okay, so one more thing: If we're moving aggregates (and adding traits) to ProviderTree, we could keep track of the last-refreshed time in either place. I'm assuming we'd like to keep doing that part at the SRC rather than in ProviderTree? | |
| 21:18:55 | efried | since he's the one doing the refreshing. | |
| 21:21:43 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Remove the deprecated TrustedFilter and related options https://review.openstack.org/521659 | |
| 21:26:07 | jaypipes | efried: yes | |
| 21:26:15 | efried | ight | |
| 21:30:15 | openstackgerrit | Dan Smith proposed openstack/nova master: Add regression test for rebuild with new image doubling allocations https://review.openstack.org/521153 | |
| 21:30:16 | openstackgerrit | Dan Smith proposed openstack/nova master: Fix doubling allocations on rebuild https://review.openstack.org/521662 | |
| 21:38:22 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Use the RequestSpec when getting scheduler_hints in compute https://review.openstack.org/515461 | |
| 21:38:23 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Pass RequestSpec to ConductorTaskAPI.build_instances https://review.openstack.org/515495 | |
| 21:47:37 | openstackgerrit | Dan Smith proposed openstack/nova master: Fix doubling allocations on rebuild https://review.openstack.org/521662 | |
| 21:57:06 | efried | jaypipes Confirm: We weren't actually using the SRC aggregate map for anything yet? | |
| 22:12:37 | mriedem | efried: what happened here? https://review.openstack.org/#/q/status:abandoned+topic:bp/add-trait-support-in-allocation-candidates | |
| 22:12:44 | mriedem | is that all superseded by jay's other series? | |
| 22:13:15 | efried | mriedem Most of it, yes. | |
| 22:13:20 | efried | Let me look for outliers... | |
| 22:14:20 | efried | mriedem Yeah; alex_xu started off doing the refactor one way; then jaypipes did it a different way and jaypipes "prevailed" :) | |
| 22:14:50 | mriedem | ok, the new series isn't tracked in launchpad as far as i can tell so i don't know what's going on there | |
| 22:17:06 | efried | mriedem TBH, add-trait-support-in-allocation-candidates may be superseded by granular-resource-requests, depending on timing. | |
| 22:17:22 | hangyang | Hi there, I have some questions for time sync in vm and its KVM host. I have some RHEL7 instances in Ocata openstack cluster. The VMs are not running ntpd and after suspend/resume, I notice there is a time drift between the VM and its KVM host and the VM will never catch up with host. But for stop/start, the vm can keep time in sync with its host. Any idea about how to solve the suspend issue? Should I run ntpd in the guest | |
| 22:17:29 | efried | mriedem See my comments in https://review.openstack.org/#/c/479776/ | |
| 22:17:36 | efried | (which is the last open one tracked by that bp) | |
| 22:23:15 | efried | mriedem Since you seem to be looking, the rest of the ksa patches are gonna be based on https://review.openstack.org/#/c/490057/ | |
| 22:23:27 | efried | ...which I believe you once referred to as "scary". | |
| 22:23:51 | mriedem | yes | |
| 22:24:12 | mriedem | i'm going through all of our queens blueprints and tracking latest status in https://etherpad.openstack.org/p/nova-queens-blueprint-status | |
| 22:24:14 | mriedem | if you must know | |
| 22:24:25 | mriedem | only about 35 more to go | |
| 22:24:35 | flwang | mriedem: sorry for chiming in | |