| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2021-11-23 | |||
| 17:43:35 | sean-k-mooney | assuming likely that this chagne had already merged | |
| 17:44:08 | sean-k-mooney | ganso: there is a docs change in this serise anyway https://review.opendev.org/c/openstack/nova/+/792362/8 | |
| 17:44:18 | sean-k-mooney | which can be used to do any more change that are required | |
| 17:48:40 | opendevreview | Rodrigo Barbieri proposed openstack/nova master: Move 'hw:pmu', 'hw_pmu' parsing to nova.virt.hardware https://review.opendev.org/c/openstack/nova/+/792364 | |
| 17:49:09 | ganso | sean-k-mooney: oh thanks I had forgotten about the other patches in the topic | |
| 20:21:21 | artom | Until Neutron starts telling us authoritatively what kind of events to wait for and when, we're pushing the responsibility on the operator to tell us in config options. | |
| 20:47:21 | sean-k-mooney | artom: maybe | |
| 20:48:02 | sean-k-mooney | we have a simialir issue wiht disable ports and evacuatate apparently | |
| 20:48:21 | sean-k-mooney | im not sure we can really just set thse for every operation | |
| 21:01:09 | opendevreview | Dan Smith proposed openstack/nova master: Revert project-specific APIs for servers https://review.opendev.org/c/openstack/nova/+/816206 | |
| 21:43:14 | artom | sean-k-mooney, ugh | |
| 21:47:12 | sean-k-mooney | the disable prot issue is simpel to fix | |
| 21:47:34 | sean-k-mooney | but im not sure if the downstream evacute issue is related to the revirt migrate issue or not | |
| 21:47:59 | sean-k-mooney | artom: the event handeling is all a bit of a mess since neutron is so inconsitent when sending events | |
| 21:48:31 | sean-k-mooney | artom: latest live migration issue https://bugs.launchpad.net/nova/+bug/1951623 | |
| 21:49:11 | sean-k-mooney | i outlined the fix in comment 3 but technically neutron should be seind plug event even for disable interface | |
| 21:49:36 | sean-k-mooney | because we do plug them the agent is just ment to set the port down | |
| 21:49:36 | artom | sean-k-mooney, yeah, so if Neutron's not fixing themselves, and we're not dumping all that complexity on the operator, then... just stop using external events altogether? | |
| 21:49:55 | artom | And too bad if the instance has no network for the first few seconds? | |
| 21:50:10 | sean-k-mooney | no we need to find time to fix it | |
| 21:51:15 | sean-k-mooney | you can just disable treating vif plug failrue as error today | |
| 21:51:32 | sean-k-mooney | its not the right solution long term but you can if you need too | |
| 21:51:58 | sean-k-mooney | and we have a second option specificly for live migration i belte and now gibi has added a new one | |
| 21:52:20 | opendevreview | Artom Lifshitz proposed openstack/nova master: Add nova-next-hybrid-plug job https://review.opendev.org/c/openstack/nova/+/817303 | |
| 21:52:52 | sean-k-mooney | https://docs.openstack.org/nova/latest/configuration/config.html#DEFAULT.vif_plugging_is_fatal | |
| 21:52:54 | sean-k-mooney | and https://docs.openstack.org/nova/latest/configuration/config.html#compute.live_migration_wait_for_vif_plug | |
| 21:53:07 | artom | True, we have the first config option | |
| #openstack-nova - 2021-11-24 | |||
| 00:52:28 | artom | Well crap, https://review.opendev.org/c/openstack/nova/+/817303 passed now | |
| 05:27:58 | frickler | artom: seems you only define the job now and don't actually run it | |
| 06:18:49 | opendevreview | Lee Yarwood proposed openstack/nova-specs master: libvirt: Allow Manila shares to be directly attached to instances https://review.opendev.org/c/openstack/nova-specs/+/813180 | |
| 08:50:08 | opendevreview | Hemanth N proposed openstack/nova master: clear dhcp_server in network_info when dhcp is disabled https://review.opendev.org/c/openstack/nova/+/819069 | |
| 10:51:56 | DK4 | im still having problems creating instances in a new deployment. instance is stuck in creating/spawning state, any hints or advices? | |
| 11:20:32 | opendevreview | Lee Yarwood proposed openstack/nova-specs master: libvirt: Allow Manila shares to be directly attached to instances https://review.opendev.org/c/openstack/nova-specs/+/813180 | |
| 11:23:21 | lyarwood | DK4: Use `openstack server event list $instance` to find the request-id associated with the creation request and trace that through the system if you have access | |
| 11:23:55 | lyarwood | DK4: you can also use openstack server event show $instance $request-id to dump any details about the create request and associated events if you don't have access to the env | |
| 11:24:31 | lyarwood | DK4: but tbh either way it's likely a messaging issue in your env between the conductor and computes | |
| 12:06:58 | opendevreview | Dr. Jens Harbott proposed openstack/nova master: Add nova-next-hybrid-plug job https://review.opendev.org/c/openstack/nova/+/817303 | |
| 12:12:20 | DK4 | lyarwood: thank your for help. im testing on proxmox7 it worked there before. i guess one of the kernel updates on proxmox host itself broke something instead. | |
| 12:25:14 | DK4 | lyarwood: the pvekernel 5.13 seems to break my openstack deployments. after switching back to 5.11 it starts to work again. thank you for your input! | |
| 13:04:20 | artom | frickler, *facepalm* yep, thanks for the fix :( | |
| 13:04:22 | artom | Err, :) | |
| 13:42:32 | frickler | artom: np, I've left the more interesting part of actually fixing the job for you ;) but I like the idea of testing that scenario, because that's how some of my deployments still look like | |
| 13:49:18 | ralonsoh | sean-k-mooney, hi! Can I ask you about https://review.opendev.org/c/openstack/neutron/+/818338/3/neutron/agent/linux/interface.py#369 ? | |
| 13:50:03 | ralonsoh | when using DPDK, the interfaces connected to a namespace are type=TAP? | |
| 13:51:02 | ralonsoh | well, not the interfaces but the ports | |
| 13:57:39 | sean-k-mooney | you mean for dhcp agent and l3 agent router instnaces | |
| 13:57:53 | sean-k-mooney | ralonsoh: with ovs-dpdk we do not supprot vnic_type normal | |
| 13:57:59 | sean-k-mooney | only vhost-user offically | |
| 13:58:21 | ralonsoh | sean-k-mooney, yeah but the Interface type=tap?? | |
| 13:58:26 | ralonsoh | in this patch | |
| 13:58:27 | sean-k-mooney | but the dhcp server and l3 agent can create port which should get create on ovs as interface_type=internal | |
| 13:58:29 | ralonsoh | is that correct? | |
| 13:58:34 | sean-k-mooney | no | |
| 13:58:57 | ralonsoh | ok then, I'll comment on the BZ | |
| 13:59:02 | ralonsoh | sean-k-mooney thanks a lot | |
| 13:59:06 | sean-k-mooney | in ovs its interface_type=internal which will create a tap device but there is not ovs interface of type tap | |
| 13:59:19 | ralonsoh | right | |
| 13:59:20 | sean-k-mooney | and we do not use vif_type=tap | |
| 13:59:42 | sean-k-mooney | ralonsoh: do you have the bz link | |
| 13:59:50 | ralonsoh | sean-k-mooney, no, this is only U/S | |
| 13:59:57 | ralonsoh | https://launchpad.net/bugs/1951493 | |
| 14:00:07 | sean-k-mooney | ob you said BZ so i got confused | |
| 14:00:14 | ralonsoh | sorry, my bad | |
| 14:00:40 | sean-k-mooney | no worries | |
| 14:01:12 | sean-k-mooney | so ovs is not namespace aware | |
| 14:01:42 | ralonsoh | well, once we create the port, we move it to the namespace | |
| 14:01:46 | sean-k-mooney | if if ovs-vswitchd is restarted then i can see this happening | |
| 14:01:54 | ralonsoh | but that could not work with dpdk | |
| 14:02:00 | sean-k-mooney | it will | |
| 14:02:15 | ralonsoh | I mean that interface doesn't exist in the kernel namespace | |
| 14:02:22 | ralonsoh | if this port is an OVS DPDK port | |
| 14:02:30 | sean-k-mooney | but the issue is for some reason when ovs-vswitchd is restart the tap is not remvoed | |
| 14:02:45 | sean-k-mooney | in this case it wil | |
| 14:03:04 | sean-k-mooney | the agents are usign interface tyep internal | |
| 14:03:08 | sean-k-mooney | and that can then be moved | |
| 14:03:11 | ralonsoh | yes | |
| 14:03:16 | sean-k-mooney | when ovs-dpdk stops it shoudl delete the tap | |
| 14:03:49 | sean-k-mooney | the l3/dhcp shoudl then pool for it to be recated and move it again once ovs starts again | |
| 14:04:06 | sean-k-mooney | that or ovs-dpdk need to check alls network namespace and reconenct | |
| 14:04:40 | ralonsoh | well, this "_ovs_add_port" method should check that | |
| 14:05:06 | ralonsoh | so we should check for the namespace and use the existing port, right? | |
| 14:05:18 | ralonsoh | existing tap port created inside the namespace | |
| 14:05:22 | sean-k-mooney | so this https://review.opendev.org/c/openstack/neutron/+/818338/3/neutron/agent/linux/interface.py seams somewhat valid but attrs.insert(0, ('type', 'tap')) is not | |
| 14:06:52 | sean-k-mooney | left comments on https://review.opendev.org/c/openstack/neutron/+/818338/3/neutron/agent/linux/interface.py | |
| 14:07:00 | ralonsoh | thanks! | |
| 14:07:46 | ralonsoh | sean-k-mooney, but L419 should be conditional | |
| 14:07:52 | ralonsoh | depending on the datapath type | |
| 14:08:01 | ralonsoh | we can't execute this if netdev | |
| 14:08:19 | ralonsoh | but setting the netns option in the namespace | |
| 14:08:31 | ralonsoh | *in the interface options | |
| 14:09:50 | sean-k-mooney | this https://review.opendev.org/c/openstack/neutron/+/818338/3/neutron/agent/linux/interface.py#149 ? | |
| 14:10:11 | sean-k-mooney | deleting the ip address? | |
| 14:10:21 | sean-k-mooney | why woudl that fail | |
| 14:10:42 | sean-k-mooney | wehn we are suign prot with type internal even with ovs-dpdk these are kerenle interfaces | |
| 14:11:03 | sean-k-mooney | so we can treat them like any other kernel interface | |
| 14:11:36 | sean-k-mooney | oh i miss read that | |
| 14:11:47 | sean-k-mooney | you ment https://review.opendev.org/c/openstack/neutron/+/818338/3/neutron/agent/linux/interface.py#419 | |
| 14:12:06 | sean-k-mooney | i think that shoudl also work for ovs-dpdk | |
| 14:12:51 | ralonsoh | OK, I'll comment that in the review. Actually this should be present in the namespace | |
| 14:12:52 | sean-k-mooney | addint the device to a network namespace shoudl work fine however it likely shoudl haveppn after the network namespaces is set on the ovs db and that shoudl eb done during prot add | |