| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-06-20 | |||
| 22:09:49 | mriedem | remember me freaking out about this earlier in the week? | |
| 22:10:48 | mriedem | on normal server create we check here https://github.com/openstack/nova/blob/4e6a7c18b1c8340b9fdc93ee02e56b5a29a99648/nova/compute/manager.py#L1930 | |
| 22:11:05 | mriedem | if that fails (it's a vmware host), we abort the build | |
| 22:11:05 | mriedem | fin | |
| 22:12:01 | mriedem | i was thinking we could allow attaching a tagged bdm to a shelved offloaded instance, and during unshelve we basically have the same flag check, and if the compute doesn't support it then we fail the unshelve | |
| 22:17:22 | dansmith | mriedem: well, the options are: | |
| 22:17:23 | dansmith | 1. refuse to attach with a tag to offloaded instances | |
| 22:17:26 | dansmith | 2. allow it and fail on unshelve | |
| 22:17:47 | Guest96259 | ...I guess that's it, right? | |
| 22:18:06 | mriedem | Guest96259: basically | |
| 22:19:16 | Guest96259 | I dunno, #1 seems not that terrible to me, given how much stuff doesn't really work with shelve | |
| 22:19:53 | mriedem | Guest96259: failing in unshelve seems like a can of worms given we don't do anything with the instance if spawn fails | |
| 22:20:34 | Guest96259 | yeah | |
| 22:20:39 | mriedem | we'll revert the task_state, | |
| 22:20:39 | mriedem | record a fault | |
| 22:20:41 | mriedem | but we don't clean anything up, i.e. https://bugs.launchpad.net/nova/+bug/1627694 | |
| 22:20:49 | mriedem | i guess you don't want the instance in ERROR state b/c then you can't attempt to unshelve it again | |
| 22:21:50 | openstack | Launchpad bug 1627694 in OpenStack Compute (nova) "unshelving an instance doesn't rollback volumes connections on failure" [Low,In progress] - Assigned to Shoham Peller (shoham-peller) | |
| 22:24:48 | mriedem | mordred: is your new api user ML a place i can ask if people are actually using some of the compute APIs? like shelve? | |
| 22:37:14 | openstackgerrit | Merged openstack/nova master: libvirt: Extract method _guest_add_video_device https://review.openstack.org/470783 | |
| 22:37:48 | openstackgerrit | Merged openstack/nova master: libvirt: Extract method _guest_add_pci_devices https://review.openstack.org/470784 | |
| 22:41:27 | openstackgerrit | Merged openstack/nova master: libvirt: Extract method _guest_add_watchdog_action https://review.openstack.org/470785 | |
| 22:42:01 | openstackgerrit | Merged openstack/nova master: libvirt: Extract method _guest_add_memory_balloon https://review.openstack.org/470786 | |
| 22:44:29 | openstackgerrit | Merged openstack/nova master: libvirt: Extract method _guest_add_spice_channel https://review.openstack.org/470787 | |
| 22:49:56 | mordred | mriedem: sure! | |
| 22:54:14 | openstackgerrit | Merged openstack/nova master: remove null_safe_int from module scope https://review.openstack.org/455778 | |
| 22:56:37 | Guest96259 | oh double rage | |
| 23:09:56 | openstackgerrit | Ildiko Vancsa proposed openstack/nova master: WIP: Add Cinder v3 detach to swap_volume https://review.openstack.org/456971 | |
| 23:27:39 | openstackgerrit | Merged openstack/nova master: Use plain routes list for os-remote-consoles instead of stevedore https://review.openstack.org/474930 | |
| 23:28:19 | gmann | alex_xu, mriedem1 its ready now - https://review.openstack.org/#/c/467999/ | |
| 23:31:18 | mriedem1 | ok | |
| 23:36:31 | mriedem | alex_xu: i'm not sure if i'll be in the next api meeting but i added something to open discussion in the agenda https://wiki.openstack.org/wiki/Meetings/NovaAPI#Agenda | |
| 23:40:34 | openstackgerrit | melanie witt proposed openstack/nova master: Add get_count_by_vm_state() to InstanceList object https://review.openstack.org/446244 | |
| 23:40:35 | openstackgerrit | melanie witt proposed openstack/nova master: Remove 'reserved' count from used limits https://review.openstack.org/446242 | |
| 23:40:36 | openstackgerrit | melanie witt proposed openstack/nova master: Remove useless quota_usage_refresh from nova-manage https://review.openstack.org/446243 | |
| 23:40:38 | openstackgerrit | melanie witt proposed openstack/nova master: Count instances to check quota https://review.openstack.org/416521 | |
| 23:40:41 | openstackgerrit | melanie witt proposed openstack/nova master: Make Quotas object favor the API database https://review.openstack.org/410945 | |
| 23:40:45 | openstackgerrit | melanie witt proposed openstack/nova master: Add online migration to move quotas to API database https://review.openstack.org/410946 | |
| 23:40:45 | openstackgerrit | melanie witt proposed openstack/nova master: Add scatter_gather_all_cells() utility https://review.openstack.org/475957 | |
| #openstack-nova - 2017-06-21 | |||
| 00:05:27 | openstackgerrit | Dan Smith proposed openstack/nova master: Fix and optimize external_events for multiple cells https://review.openstack.org/445142 | |
| 00:16:52 | cfriesen_ | mriedem: dansmith: why would we allow attaching volumes to a shelved instance? do we allow attaching volumes to an instance on a compute node that is "down"? | |
| 00:17:05 | openstackgerrit | Takashi NATSUME proposed openstack/nova master: api-ref: Add X-OpenStack-Request-Id description https://review.openstack.org/474847 | |
| 00:17:34 | mriedem | cfriesen_: the rpc call/cast would fail if the compute was down | |
| 00:17:39 | mriedem | and the instance wasn't shelved offloaded | |
| 00:17:51 | mriedem | but that's true of any instance operation for the guest | |
| 00:18:07 | mriedem | so i don't really consider that case here | |
| 00:18:18 | cfriesen_ | mriedem: it seems to me to be an analogous situation...the instance exists but we can't talk to its compute node (since the shelved one doesn't have one) | |
| 00:20:36 | cfriesen_ | I think it would be resonable to say that the only legal state-changing operations on a shelved instance are to delete it or unshelve it | |
| 00:21:10 | cfriesen_ | no IP address attach/detach, no port attach/detach, no volume attach/detach | |
| 00:22:44 | cfriesen_ | mriedem: in other news, looks like nobody cares about "nova show --minimal" so I'll go ahead and redo it to only show the original name | |
| 00:37:08 | mriedem | cfriesen_: well the cat is already out of the bag with being able to attach a volume to a shelved offloaded instance | |
| 00:37:10 | mriedem | in 2.20 | |
| 00:38:02 | mriedem | so one question is, if i'm an api user and i can attach a volume to a shelved offloaded instance in 2.20+, should i have a reasonable expectation to be able to apply a tag when attaching a volume to a shelved offloaded instance in 2.49? | |
| 00:38:29 | mriedem | because if i'm just looking at it from an api pov, how much different does tagging the thing make that operation? | |
| 00:38:36 | mriedem | or are we just being lazy by not supporting it | |
| 00:40:08 | mriedem | i'm definitely not enjoying thinking about adding more complexity to this change or the shelve/unshelve flow | |
| 01:06:14 | openstackgerrit | STEW TY proposed openstack/nova master: Transform rescue/unrescue instance notifications https://review.openstack.org/385644 | |
| 01:10:34 | oomichi | takashin: hi, can you take a look at https://review.openstack.org/#/c/474847 again ? some nits | |
| 01:16:34 | alex_xu | mriedem: got it | |
| 01:17:21 | Kevin_Zheng | ping mriedem | |
| 01:46:35 | openstackgerrit | Ngo Quoc Cuong proposed openstack/nova master: Remove translation of log messages https://review.openstack.org/466637 | |
| 01:52:49 | mriedem | Kevin_Zheng: hey what's up? | |
| 01:53:24 | Kevin_Zheng | Hi, it's about the delete neutron port thing | |
| 01:54:04 | mriedem | ok | |
| 01:54:26 | Kevin_Zheng | it seems instance obj's info_cache didn't got refreshed | |
| 01:54:54 | Kevin_Zheng | https://github.com/openstack/nova/blob/9ed0d611453aee7a27debd773bdbe146510be0ad/nova/network/neutronv2/api.py#L1274 | |
| 01:55:28 | Kevin_Zheng | after compute_utils.refresh_info_cache_for_instance(context, instance) instance.info_cache still have only one record | |
| 01:56:22 | Kevin_Zheng | and if I do instance.info_cache_refresh() again after nw_info = self._build_network_info_model(context, instance, networks,port_ids, admin_client, preexisting_port_ids) | |
| 01:56:42 | Kevin_Zheng | it contains the new record, and everything will be correct | |
| 01:57:17 | mriedem | so it starts from here correct? https://github.com/openstack/nova/blob/9ed0d611453aee7a27debd773bdbe146510be0ad/nova/network/neutronv2/api.py#L881 | |
| 01:57:34 | mriedem | ordered_ports and preexisting_port_ids are the same there i think | |
| 01:57:46 | mriedem | when attaching a preexisting port to an instance | |
| 01:58:09 | mriedem | so that's calling this https://github.com/openstack/nova/blob/9ed0d611453aee7a27debd773bdbe146510be0ad/nova/network/base_api.py#L246 | |
| 01:58:26 | mriedem | which calls back into the neutronv2 api code https://github.com/openstack/nova/blob/9ed0d611453aee7a27debd773bdbe146510be0ad/nova/network/base_api.py#L249 | |
| 01:58:37 | mriedem | https://github.com/openstack/nova/blob/9ed0d611453aee7a27debd773bdbe146510be0ad/nova/network/neutronv2/api.py#L1274 | |
| 01:58:57 | mriedem | at that point, compute_utils.refresh_info_cache_for_instance(context, instance) won't have the newly attached port in it | |
| 01:59:04 | Kevin_Zheng | yeah | |
| 01:59:11 | Kevin_Zheng | Thats what I tested | |
| 01:59:12 | mriedem | because _build_network_info_model is going to process it | |
| 01:59:20 | Kevin_Zheng | yeah | |
| 01:59:32 | mriedem | so we call _build_network_info_model with the port_ids and preexisting_port_ids list, which are going to have the same single entry | |
| 01:59:52 | mriedem | and network_info returned from that will have a single vif in it | |
| 02:00:18 | mriedem | which is asserted here https://github.com/openstack/nova/blob/9ed0d611453aee7a27debd773bdbe146510be0ad/nova/compute/manager.py#L5080 | |
| 02:00:37 | mriedem | ok, so that explains it | |
| 02:00:53 | mriedem | that also explains a bug that vladikr is trying to fix | |
| 02:00:59 | mriedem | https://review.openstack.org/#/c/468758/ | |
| 02:02:26 | Kevin_Zheng | so, should we do instance.info_cache.refresh() again after _build_network_info_model ? | |
| 02:02:54 | openstackgerrit | Ngo Quoc Cuong proposed openstack/nova master: Remove translation of log messages https://review.openstack.org/466957 | |
| 02:02:55 | openstackgerrit | Li Yingjun proposed openstack/nova master: Fix duplicate Neutron ports on instance reschedule https://review.openstack.org/467509 | |
| 02:06:01 | mriedem | Kevin_Zheng: that won't do it i don't think | |
| 02:06:11 | mriedem | the info cache is updated on the instance here https://github.com/openstack/nova/blob/9ed0d611453aee7a27debd773bdbe146510be0ad/nova/network/base_api.py#L255 | |
| 02:06:48 | Kevin_Zheng | it seems it only saved info_cache to DB | |
| 02:07:05 | mriedem | yes this right? https://github.com/openstack/nova/blob/9ed0d611453aee7a27debd773bdbe146510be0ad/nova/network/base_api.py#L50 | |
| 02:07:14 | Kevin_Zheng | the instance obj in memory still didn't contain new record | |
| 02:07:37 | mriedem | hmm, why don't we just set instance.info_cache = ic there? | |
| 02:08:03 | Kevin_Zheng | yeah, that will also work | |
| 02:08:34 | artom | mriedem, re "care about supporting attaching tagged volumes to a shelved offloaded instance" - I think it's less bad in the create BuildAbort case because the feedback the user gets is more immediate | |
| 02:09:27 | artom | Whereas with shelved offloaded there could be a delay between attaching a tagged volume and unshelving the instance | |
| 02:09:31 | artom | I dunno | |
| 02:09:47 | mriedem | Kevin_Zheng: can you test that out and see if it fixes the bug? because that would be much easier and convenient since we'd have the object we need right there | |