| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-10-11 | |||
| 13:49:34 | sahid | jaypipes: i replied to you about the sriov-trusted-vfs (in case you did not notice it) | |
| 14:02:37 | mdbooth | stephenfin: Can't say I'm a fan of that change. Haven't put a -1 only because it's *probably* better than what we have now. Not sure if the added complexity is worth it, though. | |
| 14:04:48 | mriedem | sahid: which hypervisors don't support this flag? https://libvirt.org/html/libvirt-libvirt-domain.html#VIR_MIGRATE_PARAM_BANDWIDTH | |
| 14:08:11 | sahid | mriedem: hum.. let me check that point | |
| 14:09:47 | openstackgerrit | Merged openstack/nova master: Don't fix protocol-less glance api_servers anymore https://review.openstack.org/505317 | |
| 14:12:02 | sahid | mriedem: according to a simple grep in libvirt source it seems that vz and QEMU are supporting it | |
| 14:12:31 | sahid | but but nothing about vmware, hyperv.. | |
| 14:13:22 | mriedem | sahid: well we wouldn't care about vmware or hyperv in the libvirt driver | |
| 14:13:37 | mriedem | but we care about xen, although i'm not sure if libvirt+xen is listed as supporting live migration anyway | |
| 14:13:57 | openstackgerrit | Eric Fried proposed openstack/nova-specs master: Granular Resource Request Syntax https://review.openstack.org/510244 | |
| 14:13:59 | mriedem | according to this it says we do https://docs.openstack.org/nova/latest/user/support-matrix.html#operation_live_migrate | |
| 14:14:31 | openstackgerrit | Merged openstack/nova master: Move user_data max length check to schema https://review.openstack.org/510704 | |
| 14:14:55 | openstackgerrit | Surya Seetharaman proposed openstack/nova master: nova-manage map_instances is not using the cells info from the API database https://review.openstack.org/510844 | |
| 14:18:51 | sahid | mriedem: we always use that parameter (bandwiddh) for live-migration | |
| 14:19:09 | sahid | do you have a suggestion on how the code should be? | |
| 14:19:26 | mriedem | it must work for xen then | |
| 14:19:33 | sahid | I guess yes | |
| 14:19:33 | mriedem | libvirt+xen i mean | |
| 14:19:37 | sahid | but I can't ensure that | |
| 14:19:54 | mriedem | anthonyper: do you know what the current state of libvirt+xen live migration is? working? | |
| 14:19:58 | mriedem | has anyone tested lately? | |
| 14:25:41 | anthonyper | mriedem: I have no idea :( | |
| 14:33:46 | anthonyper | mriedem: I'll ask other people to try to find out if they have tested live migration recently. | |
| 14:34:36 | mriedem | anthonyper: ok thanks | |
| 14:36:09 | sahid | dansmith: if the wait_for_instance_event has been desgined has you are suggesting so it's not working | |
| 14:38:04 | dansmith | sahid: I'm not sure what you're saying | |
| 14:38:27 | jaypipes | sahid: yep, noticed it. sorry, been spending my last days rebasing the nested resource providers series. | |
| 14:38:42 | jaypipes | sahid: I'll review that slow live-migration patch this morning. | |
| 14:38:50 | openstackgerrit | Merged openstack/nova master: Add connection pool size to vSphere settings https://review.openstack.org/414085 | |
| 14:39:12 | openstackgerrit | Surya Seetharaman proposed openstack/nova master: nova-manage map_instances is not using the cells info from the API database https://review.openstack.org/510844 | |
| 14:39:20 | sahid | jaypipes: no worries and thanks if you can have a look | |
| 14:39:34 | openstackgerrit | Merged openstack/nova master: Clarify conf/compute.py help text for ListOpts https://review.openstack.org/468538 | |
| 14:39:53 | sahid | dansmith: related to one of your comment on https://review.openstack.org/#/c/497457/ | |
| 14:40:20 | sahid | the other one are just nits, right? | |
| 14:41:35 | dansmith | sahid: no, breaking n-net migrations is not a nit :) | |
| 14:42:39 | sahid | dansmith: can you clearly put a suggestion? do you me to add a condition in the code? | |
| 14:43:17 | dansmith | sahid: you can handle it like the spawn code does for if neutron isn't enabled | |
| 14:43:31 | mriedem | if (self._conn_supports_start_paused and | |
| 14:43:31 | mriedem | utils.is_neutron() and not | |
| 14:43:31 | mriedem | vifs_already_plugged and power_on and timeout): | |
| 14:43:31 | mriedem | events = self._get_neutron_events(network_info) | |
| 14:43:31 | mriedem | else: | |
| 14:43:32 | mriedem | events = [] | |
| 14:43:32 | mriedem | yeah | |
| 14:43:44 | mriedem | if this and that and the other thing else something different | |
| 14:43:49 | sahid | dansmith: seems reasonable | |
| 14:44:58 | dansmith | sahid: I also want to know which specific action starts the chain of events that will result in neutron sending an event to nova, so that you can properly wrap that thing with the waiter | |
| 14:45:56 | sahid | dansmith: depending of the mech used, i added a comment in the code | |
| 14:47:24 | dansmith | sahid: ah, so two different places depending on OVS or linuxbridge? that's going to be a mess to handle properly | |
| 14:47:30 | dansmith | but what you have isn't right for either | |
| 14:47:35 | dansmith | although it's much much wronger for OVS | |
| 14:50:01 | sahid | dansmith: i guess you are going to elaborate? | |
| 14:50:43 | dansmith | sahid: I did in my comments | |
| 14:50:48 | dansmith | nova events are not queued | |
| 14:50:58 | dansmith | if you don't have a waiter running before an event arrives, it is dropped | |
| 14:51:20 | dansmith | if you don't want to race, you need to have a waiter set up before you do the thing that triggers the event from the other service | |
| 14:51:46 | dansmith | delaying live migration for five minutes, timeout, log error and then continue is not a good thing to do just because you raced with neutron | |
| 14:54:47 | mdbooth | mriedem: To spec, or not to spec: https://blueprints.launchpad.net/nova/+spec/local-disk-serial-numbers ? | |
| 14:55:19 | mdbooth | dansmith: Same question ^^^ | |
| 14:55:26 | sahid | dansmith: i see so we will have to add this specificly for ovs during pre-live-migration when we are creating the bridge on dest | |
| 14:55:43 | sahid | and when we start calling libvirt.migrateToURI for LB | |
| 14:56:15 | dansmith | sahid: neither option seems very ideal to me, but yes, that's the issue with what you have here | |
| 14:57:00 | dansmith | mdbooth: probably a spec so we can argue about your id generation method | |
| 14:57:21 | mdbooth | dansmith: Yeay! I do love a good opportunity to argue about stuff. | |
| 14:57:26 | mdbooth | Ok | |
| 14:58:25 | mriedem | gmann_afk: commented in that change and the one below it | |
| 15:07:47 | sahid | dansmith: since we are now using the events, do you see a blocker at moving plug_vifs() from pre-live-migration to migration step? | |
| 15:07:50 | mriedem | sahid: i looked at one of the live migration job runs on your patch and it's never waiting for any events | |
| 15:08:07 | openstackgerrit | Eric Fried proposed openstack/nova master: Nix bug msg from ConfGroupForServiceTypeNotFound https://review.openstack.org/511261 | |
| 15:08:17 | efried | mriedem ^ cleanup as requested. | |
| 15:08:45 | sahid | mriedem: what is the mech? for linux bridge we need that patch on neutron side | |
| 15:08:48 | sahid | https://review.openstack.org/#/c/506182/ | |
| 15:09:27 | mriedem | sahid: the depends-on will pull the change in via zuul | |
| 15:09:41 | mriedem | sahid: it's literally building an empty list of events | |
| 15:09:47 | mriedem | meaning there are no vifs in the instance.info_cache | |
| 15:09:53 | mriedem | this would be ovs | |
| 15:10:08 | mriedem | http://logs.openstack.org/57/497457/11/check/legacy-tempest-dsvm-multinode-live-migration/9cd842e/logs/screen-n-cpu.txt.gz#_Oct_02_11_04_06_172675 | |
| 15:10:20 | sahid | mriedem: hm... if there is no vifs so yes i guess it's normal | |
| 15:11:36 | mriedem | hmm wtf http://logs.openstack.org/57/497457/11/check/legacy-tempest-dsvm-multinode-live-migration/9cd842e/logs/screen-n-cpu.txt.gz#_Oct_02_11_04_01_315354 | |
| 15:11:59 | mriedem | cfriesen: ^ this probably explains why i never saw any improvement in the network-changed event queueing patch for live migration | |
| 15:12:00 | sahid | hum... | |
| 15:12:29 | mriedem | cfriesen: this one https://review.openstack.org/#/c/465787/ | |
| 15:17:02 | efried | We needing to pay attention to bogus Zuul merge conflicts at this point? | |
| 15:17:09 | mriedem | no | |
| 15:17:16 | mriedem | well, idk, but i'm not | |
| 15:17:25 | dansmith | today we do | |
| 15:17:28 | efried | Thought they cut over this morning. | |
| 15:17:31 | dansmith | once the cutover starts | |
| 15:35:54 | jianghuaw_ | jaypipes, per the previous, we should populate the inventory for vgpu in the function of get_inventory. So we should re-open this abandoned patch: https://review.openstack.org/#/c/339204/. Is it correct? | |
| 15:38:38 | openstackgerrit | Merged openstack/nova master: Updated from global requirements https://review.openstack.org/511021 | |
| 15:40:39 | bauzas | jianghuaw: I tend to say yes | |
| 15:40:42 | jaypipes | jianghuaw_: yes | |
| 15:40:50 | bauzas | jianghuaw: FWIW, I'm working on the libvirt side | |
| 15:40:51 | jaypipes | jianghuaw_: feel free to take it. | |
| 15:41:18 | jaypipes | jianghuaw_: I restored it just now. feel free to git review -d 339204 and take over ownership | |
| 15:41:36 | jaypipes | jianghuaw_: or cherry-pick it as you want\ | |
| 15:41:39 | jianghuaw_ | bauzas, jaypipes: great. Thanks. | |
| 15:42:47 | openstackgerrit | Eric Fried proposed openstack/nova master: Get auth from context for glance endpoint https://review.openstack.org/490057 | |
| 15:44:18 | openstackgerrit | Eric Fried proposed openstack/nova master: Use ksa adapter for keystone conf & requests https://review.openstack.org/507693 | |
| 15:44:33 | openstackgerrit | Eric Fried proposed openstack/nova master: Use ksa adapter for cinder client https://review.openstack.org/508345 | |