| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2021-11-15 | |||
| 15:52:21 | EugenMayer | create the router so Bsub1 can reach Bsub2) the gateway of Bsub1 is pushed | |
| 15:52:21 | EugenMayer | sean-k-mooney: i have an issue with the fact that i can only add a router with those 2 interfaces, if a gateway is presnet. I have a network A (sub1/sub2) and network B (sub1/sub2). I want an instance to be in A sub1 with the default gateway of Asub1, while this instnace is also in Bsub1. The problem is, if Bsub1 also has a gateway (which i need to | |
| 15:52:26 | EugenMayer | (default gateway) | |
| 15:53:20 | sean-k-mooney | so in generall you will create a tenant network and a rounter to connect the tenant network to the external network | |
| 15:53:43 | sean-k-mooney | and if you want to have 2 subnets on that tenant network you just add a second interface to the existing router for the second subnet | |
| 15:54:35 | sean-k-mooney | Zer0Byte: ok if it does not appear in the RP list then if you check the compaute agent log on that host it should have an error | |
| 15:55:08 | sean-k-mooney | Zer0Byte: in the logs o fthe update_avaiable_resouces periodic task if not earlier | |
| 15:56:12 | EugenMayer | sean-k-mooney my problem is with the default gateway pushed, if an instance has 2 interfaces which both have a gateway | |
| 15:56:49 | sean-k-mooney | right that is a common problem | |
| 15:56:57 | sean-k-mooney | you can disable the gateway on the subnet | |
| 15:57:17 | sean-k-mooney | which should prevent that | |
| 15:58:53 | sean-k-mooney | EugenMayer: i think you do "openstack subnet unset --gateway <subnet>" | |
| 15:59:00 | EugenMayer | sean-k-mooney yes, that is what i did, then i cannot route Bsub1 to Bsub2 - you cannot add the interface | |
| 15:59:21 | EugenMayer | if a subnet has not gateway, it cannot be added to the router to route to the subnets | |
| 15:59:30 | sean-k-mooney | you can add the interface but you have to specify the ip manually to be what would be the normal gateway ip | |
| 15:59:59 | EugenMayer | so you mean override the gateway via cloud-init? | |
| 16:00:04 | sean-k-mooney | so when adding the router interface you will need to give it the .1 address or whatever adress is correct for the cidr | |
| 16:00:08 | sean-k-mooney | no | |
| 16:00:17 | sean-k-mooney | do openstack subnet unset --gateway <subnet b> | |
| 16:00:18 | Zer0Byte | "title": "Service Unavailable"}. | |
| 16:00:18 | Zer0Byte | RROR nova.scheduler.client.report [req-4d4167be-e58d-42eb-bf49-ffeb0a21d42c - - - - -] [req-6400ff72-11e0-4953-9980-463cfacf7df1] Failed to retrieve resource provider tree from placement API for UUID 1c684ec9-6ab2-4ef7-8451-c989a0b90f45. Got 503: {"message": "The server is currently unavailable. Please try again at a later time.<br /><br />\nThe Keystone service is temporarily unavailable.\n\n", "code": "503 Service Unavailable", | |
| 16:00:23 | Zer0Byte | is exactly what i suspect | |
| 16:00:47 | Zer0Byte | now my question is how i force the node to populate on placement | |
| 16:03:36 | sean-k-mooney | then do "openstack port create --network <my net> --fixed-ip subnet=<subnet>,ip-address=192.168.2.1 router_subnet_b_port" | |
| 16:03:42 | EugenMayer | create a neutron router to bridge between B-sub1 and B-sub2, since a gateway in B-sub1 seems to be required for that | |
| 16:03:42 | EugenMayer | sean-k-mooney sorry i'am confused. Let met try to explain my dilemma more clearly, maybe you can point me to my mistake. Network A(sub1) and Network B(sub1,sub2) exist. Now i have to options. I disable the gw on B-sub1 so if my instance is in A-sub1 and B-sub1, the default gateway of A-sub1 is used. If i disable that gw, i'am no longer able to | |
| 16:04:05 | sean-k-mooney | then do openstack router add port <router> <port uuid for router_subnet_b_port> | |
| 16:05:10 | sean-k-mooney | EugenMayer: if the gateway is disable i think you have to manually create the port rather then addign the subnet to the router a is descibed above ^ | |
| 16:05:24 | EugenMayer | interesting | |
| 16:05:38 | sean-k-mooney | Zer0Byte: i think you jsut need to restart the compute agent | |
| 16:05:57 | sean-k-mooney | it should try again the next time the perodic runs if your keystone issues are resovled | |
| 16:06:07 | EugenMayer | so i should stick to B sub1/sub2 with gw disabled on sub1 and try to workarround the issue to bridge Bsub1 to Bsub2 | |
| 16:08:12 | sean-k-mooney | so your real issue is that dhcp is going to fight with the default router right | |
| 16:08:25 | sean-k-mooney | normally i fix this with cloud-init manually and set the metic | |
| 16:08:38 | sean-k-mooney | what sound like would be useful would be some way to configure the metric per subnet | |
| 16:09:38 | sean-k-mooney | if you never needed to have two port on the same vm (from both subnets) this woudl not normlay be an issue | |
| 16:11:00 | EugenMayer | well i need that in my case. It is an kubernetes control-plane while A is its 'intranet' for others to access apps internally and also get routed to a floating IP. B is the cluster internal network | |
| 16:12:06 | EugenMayer | sean-k-mooney i'am not sure what your advice is. Try to go with disabling the gw and try adding the port manually or try to use cloud-init to somehow change the metric for DHCP | |
| 16:14:08 | Zer0Byte | @sean-k-mooney restarning the compute agent don;t work also no error on logs :( | |
| 16:15:23 | Zer0Byte | for understand the flow nova-agent (the service isntaled on the compute node) will contact placement directly right | |
| 16:39:09 | hyang[m] | Hi Nova core reviewers, can someone help to check the patch https://review.opendev.org/c/openstack/nova/+/811521? It can resolve a bug that affects both Nova and Neutron. | |
| 16:55:10 | sean-k-mooney | hyang[m]: that is partly a feature and partly a bug | |
| 16:55:30 | sean-k-mooney | i say its partly a feature as support for neutron rbac in nova is a new feature | |
| 16:56:05 | sean-k-mooney | its partly a bug in that that should have been part of the development of security group shareing in neutorn | |
| 16:56:14 | sean-k-mooney | there shoudl have been a sibligh spec to cover the nova changes | |
| 16:56:51 | sean-k-mooney | actully is this a new neutron featfure this cycle | |
| 16:57:07 | sean-k-mooney | if so then you shoudl convert that patch to a blueprint | |
| 16:59:24 | Zer0Byte | is very weird i ssue | |
| 16:59:28 | Zer0Byte | i put the agent on debug | |
| 16:59:38 | Zer0Byte | Final resource view: name=tc-m2mem-compute2-1210 is registering on placement | |
| 16:59:44 | Zer0Byte | but don;t show as a resource | |
| 16:59:50 | sean-k-mooney | hyang[m]: actully this should have a spec or specless bluerpint its a new feature not a bug | |
| 17:00:07 | sean-k-mooney | hyang[m]:this si not something that should be backported | |
| 17:05:35 | bauzas | folks, gentle reminder that tomorrow is spec review day | |
| 17:05:38 | bauzas | sharpen your pens | |
| 17:06:18 | gibi | bauzas: good reminder | |
| 17:06:20 | gibi | thanks | |
| 17:07:40 | bauzas | gibi: you can thank sean-k-mooney for ringing me a bell in my empty closet that's called "a brain" | |
| 17:07:51 | gibi | sean-k-mooney: thanks | |
| 17:07:52 | gibi | :D | |
| 17:08:13 | sean-k-mooney | as i said downstream the only reason i rememebred is i tought i missed it last week so i checked :) | |
| 17:09:39 | gibi | bauzas: I think you can abandon these https://review.opendev.org/q/project:openstack/nova-specs+status:open+label:Code-Review-2 as all of them are heavily outdated | |
| 17:10:12 | gibi | just to have a clean spec list for tomorrow | |
| 17:11:46 | Zer0Byte | any another idea @sean-k-mooney ? | |
| 17:16:05 | sean-k-mooney | Zer0Byte: have you tried openstack resource provider list --name tc-m2mem-compute2-1210 | |
| 17:16:31 | Zer0Byte | got a empty response | |
| 17:16:53 | sean-k-mooney | Zer0Byte: assuming this is libvirt what does virsh hostname return | |
| 17:17:12 | sean-k-mooney | does it have a domain appended to it? | |
| 17:17:18 | Zer0Byte | no | |
| 17:17:29 | Zer0Byte | im using juju to install the compute nodes | |
| 17:17:43 | Zer0Byte | so the configuration is always persistent | |
| 17:18:05 | sean-k-mooney | ok libvirt via the python bindign provides the hostname to nova which we use for the RP name | |
| 17:18:22 | Zer0Byte | in this case we should have a empty record with id right? | |
| 17:18:27 | Zer0Byte | uuid | |
| 17:19:01 | sean-k-mooney | well we should have a resouce provider with the hostname as the name and the compute node uuid( not the compute service uuid) as the uuid | |
| 17:19:18 | sean-k-mooney | and then if you do an inventory list you shoudl see the reouces that are tracked | |
| 17:19:25 | sean-k-mooney | so cpu ram disk ectra | |
| 17:20:09 | sean-k-mooney | Zer0Byte: if you are not seeing the RP in placment then you should be sing an error in the compute agent log trying to create the RP at some point | |
| 17:20:26 | Zer0Byte | yeah from the other nodes yes | |
| 17:21:01 | Zer0Byte | i can see the usage form inventory | |
| 17:23:10 | Zer0Byte | my doubt here is how nova is "reporting allocations" to placement without a RP uuid | |
| 17:23:22 | Zer0Byte | or to a non visible RP | |
| 17:23:40 | Zer0Byte | Instance 21a17b74-400f-4dce-99ed-555e1ae80512 actively managed on this compute host and has allocations in placement: {'resources': {'VCPU': 8, 'MEMORY_MB': 65536}}. _remove_deleted_instances_allocations /usr/lib/python3/dist-packages/nova/compute/resource_tracker.py:1538 | |
| 17:25:16 | EugenMayer | sean-k-mooney thank you for that metric trick with cloud-init. Surely not perfect since it makes the entire instance very topology aware, but i guess that info leaked into it never the less | |
| 17:27:54 | sean-k-mooney | well normally the netowrks routers and isntance are created by the tenant | |
| 17:28:01 | sean-k-mooney | so they are creating that toplogy themselves | |
| 17:28:21 | sean-k-mooney | in your case its sligly different since you are migrating/addpoting workload form your previous enviornment | |
| 17:28:55 | sean-k-mooney | i wonder if dhcp has the concept of a metiric it can pass | |
| 17:29:11 | sean-k-mooney | if so it might be worth wile asking neutron to support setting that via the subnet | |
| 17:29:33 | Zer0Byte | i will try something @sean-k-mooney the uuid from RP is auto generated or is from the ocmpute host? | |
| 17:29:34 | sean-k-mooney | it would be nice if you could jsut declare that via the subnet and not need to do it per instance | |
| 17:29:56 | sean-k-mooney | Zer0Byte: its autogenerate when the compute agent first connect to the db | |
| 17:30:25 | sean-k-mooney | Zer0Byte: you can fined it if you have access to the db by looking at the compute_nodes table in the cell db | |
| 17:30:35 | Zer0Byte | ok i wil try that | |
| 17:30:36 | sean-k-mooney | or you can get it via the rest api vai the hyperviors api | |
| 17:30:50 | Zer0Byte | recreate manually the rp | |
| 17:31:13 | Zer0Byte | to see if after i generate it i can get the allocations that the compute node is reporting | |
| 17:31:47 | EugenMayer | i'll try sean-k-mooney | |
| 17:33:03 | sean-k-mooney | to create the allocation i think you whould have to run the allcoation audit or heal allocation nova/placment manage commands | |
| 17:38:03 | EugenMayer | interesting, there is extra_dhcp_option on the port | |