| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-08-29 | |||
| 07:46:09 | jaypipes | ratailor|Lunch: k, reviewed. left instructions on the patch. | |
| 07:56:33 | yushiro | Hi, I'd like to give some help for registering host aggregate in stable/ocata with devstack environment. I deployed 2 nodes. 1st is controller which compute driver is 'ironic'. The other is compute node which compute driver is 'libvirt.LibvirtDriver' | |
| 07:56:39 | yushiro | I'd like to separate 'virtual' and 'baremetal' by using host aggregate. However, if I specify controller's hostname as aggregate host, it returns 404(Not Found) | |
| 07:56:45 | yushiro | Here is an operation log: http://paste.openstack.org/show/619721/ | |
| 07:56:51 | yushiro | What can I specify as <host> for 'openstack aggregate add host <aggregate> <host>'? | |
| 08:05:50 | jaypipes | yushiro: generally best to ask usage questions like this on the openstack@ mailing list or ask.openstack.org (see channel /topic). but, short answer is that you want to use the *compute host* as the host in your aggregate, not the controller. | |
| 08:06:50 | jaypipes | yushiro: your ironic baremetal nodes will be managed by one or more nova-compute services. likewise, your VMs will be managed by nova-compute services running on hypervisor hosts. It is those nova-compute host names that you will use in doing the host-aggregate assignments. | |
| 08:08:27 | yushiro | jaypipes, Thanks. Ah, sorry. I deployed 2 nodes( 1:controller with compute(including ir-cond, ir-api), 2:compute) | |
| 08:09:09 | openstackgerrit | Elod Illes proposed openstack/nova master: Functional test: evacuate with no compute https://review.openstack.org/498482 | |
| 08:09:23 | yushiro | jaypipes, Yes, I think so too. However, controller-hostname (furukawa-ctrl) cannot found ;( ... | |
| 08:10:35 | jaypipes | yushiro: you will still need a nova-compute service on that controller node, though. if you do a nova-manage service list, what shows up? | |
| 08:11:56 | openstackgerrit | Gábor Antal proposed openstack/nova master: Transform instance.resize_prep notification https://review.openstack.org/465081 | |
| 08:12:08 | yushiro | jaypipes, hmm, my goal is to deploy instance both ironic and nova(libvirt) in 1 controller. Let me check a result... | |
| 08:13:14 | jaypipes | yushiro: that won't work :) you'll need to set up a container running on the controller that has a separate hostname for the nova-compute service managing the Ironic nodes. | |
| 08:16:58 | yushiro | jaypipes, ouch!! oh, OK. So, you mean it should be separated between nova-compute and ironic-conductor ? | |
| 08:19:20 | jaypipes | yushiro: the same nova-compute service cannot run *both* the libvirt *and* the ironic virt driver. therefore you will need to have the nova-compute service worker that manages the Ironic baremetal nodes on a separate installation than the nova-compute service running the libvirt virt driver and managing virtual resources. | |
| 08:19:46 | jaypipes | yushiro: you can easily do so by just making a container that runs the nova-compute service with the ironic virt driver. | |
| 08:20:30 | jaypipes | yushiro: that container can run on the same physical host that the other nova-compute service (that manages the virtual resources) is running on. But, it needs a separate hostname and separate process (because the nova.conf will be different) | |
| 08:21:17 | yushiro | jaypipes, Thank you. Well, here is a result 'nova-manage host list' http://paste.openstack.org/show/619730/ Please refer at the end of line. | |
| 08:24:21 | yushiro | jaypipes, Let me explain again.. furukawa-ctrl includes n-cpu,n-csh,n-cond,n-api,n-cell-region,n-cell-child,placement-api,ir-api,ir-cond..., and compute driver is 'ironic.IronicDriver', furukawa-com includes n-cond,n-api-meta,n-cpu and compute driver is 'libvirt.LibvirtDriver' | |
| 08:25:41 | jaypipes | yushiro: you are using cellsv1 functionality? | |
| 08:26:46 | jaypipes | yushiro: you should not use cellsv1 (nova-cells service). what version of OpenStack are you running? | |
| 08:27:02 | jaypipes | yushiro: and how are you installing? are you installing a distro like RDO? | |
| 08:27:09 | openstackgerrit | Gábor Antal proposed openstack/nova master: Transform libvirt.error notification https://review.openstack.org/484851 | |
| 08:27:15 | yushiro | jaypipes, oops, sorry. I didn't tell you a version. The version is stable/ocata with devstack. | |
| 08:28:26 | yushiro | jaypipes, I executed 'nova-manage cell_v2 list_cells' and returned named 'cell0' which UUID is 00000000-0000-0000-0000-000000000000. | |
| 08:28:48 | jaypipes | yushiro: yes, it's confusing. nova-cells service is cellsv1... | |
| 08:29:03 | jaypipes | yushiro: in your localrc, did you enable NOVA_CELLS or something? | |
| 08:30:19 | yushiro | jaypipes, Yes. I added enable_service n-cell in my local.conf because it was got an error if I didn't add this configuration. | |
| 08:31:16 | jaypipes | yushiro: ok, I believe that is the primary source of the problems... | |
| 08:31:36 | jaypipes | yushiro: I'm afraid you will likely need to start over with your devstack installation :( | |
| 08:31:57 | jaypipes | yushiro: you probably hit an error related to "cellsv2 not setup" or similar, yes? | |
| 08:32:38 | jaypipes | yushiro: it is quite unfortunate that cellsv1 and cellsv2 are so confusing. :( | |
| 08:32:47 | yushiro | jaypipes, wow, really? I didn't test 'cellsv2 not setup' but .... OK | |
| 08:33:13 | jaypipes | yushiro: the nova-cells service is cellsv1 and deprecated. cellsv2 is entirely transparent and does not have separate service workers. | |
| 08:33:41 | jaypipes | yushiro: what was the error you got? | |
| 08:34:06 | jaypipes | yushiro: you said above "got an error if I didn't add this configuration.". what specific error did you get? do you remember? | |
| 08:35:50 | yushiro | jaypipes, Sorry. I deployed this env for 2 month ago, log message has been removed ;( | |
| 08:37:54 | jaypipes | yushiro: understood :) I think the error *probably* said something like "cellsv2 setup not complete" or something like that. | |
| 08:38:47 | jaypipes | yushiro: I am 90% sure, however, that the source of your issue here is that cellsv1 is enabled in that environment and unfortunately, the only way to fix this is to start over and not have n-cells enabled. | |
| 08:39:02 | jaypipes | yushiro: apologies for the bad news :( | |
| 08:39:28 | yushiro | jaypipes, no-no. I appreciate with your kindly explanation :) Thank you so much | |
| 08:42:15 | openstackgerrit | Lei Zhang proposed openstack/nova master: placement: extract traits from flavor extra spec https://review.openstack.org/492026 | |
| 08:55:14 | openstackgerrit | Naichuan Sun proposed openstack/nova master: xenapi: cached images should be cleaned up by time https://review.openstack.org/465954 | |
| 09:00:49 | openstackgerrit | Naichuan Sun proposed openstack/nova master: xenapi: cached images should be cleaned up by time https://review.openstack.org/465954 | |
| 09:11:29 | openstackgerrit | Naichuan Sun proposed openstack/nova master: xenapi: Support live migration in pooled multi-nodes environment https://review.openstack.org/489451 | |
| 09:34:00 | openstackgerrit | Viktor Varga proposed openstack/nova master: Ensure image metadata when unshelving BFV instance https://review.openstack.org/498729 | |
| 09:34:00 | openstackgerrit | Viktor Varga proposed openstack/nova master: Remove mox from unit/conductor/test_conductor.py (1) https://review.openstack.org/496691 | |
| 09:53:38 | bhagyashri_s | ratailor: Hi, There is one alternative option to debug unit test case ref: http://superuser.openstack.org/articles/a-guide-to-testing-in-openstack/ not sure that is working or not. | |
| 09:54:22 | ratailor | bhagyashri_s, Thanks! btw I was looking for functional tests. :0 | |
| 09:54:25 | ratailor | :) | |
| 09:54:58 | ratailor | bhagyashri_s, And I got that by this way.. python -m testtools.run discover nova.tests.functional.db.api.test_migrations | |
| 09:55:36 | bhagyashri_s | ratailor: ok. :) | |
| 09:55:45 | openstackgerrit | Alex Xu proposed openstack/nova master: placement: functional tests for multiple shared resource providers https://review.openstack.org/498737 | |
| 09:55:45 | openstackgerrit | Alex Xu proposed openstack/nova master: placement: ensure RP maps to those RPs that share with it https://review.openstack.org/480379 | |
| 10:01:57 | openstackgerrit | Alex Xu proposed openstack/nova master: placement: ensure RP maps to those RPs that share with it https://review.openstack.org/480379 | |
| 10:17:20 | openstackgerrit | Jianghua Wang proposed openstack/nova master: XenAPI: deprecate image handler class path option https://review.openstack.org/497201 | |
| 11:06:35 | VAhl | WARNING: cell0 mapping not found - not syncing cell0. How do I recreate the cell0? OS Pike | |
| 11:34:41 | openstackgerrit | Merged openstack/nova master: tests: De-duplicate some graphics tests https://review.openstack.org/497969 | |
| 11:36:24 | openstackgerrit | Merged openstack/nova master: Updates to scheduling workflow doc https://review.openstack.org/498613 | |
| 11:57:09 | openstackgerrit | Jianghua Wang proposed openstack/nova master: XenAPI: define a new image handler to use vdi streaming https://review.openstack.org/486475 | |
| 13:04:57 | openstackgerrit | Balazs Gibizer proposed openstack/nova master: cover migration cases with functional tests https://review.openstack.org/493865 | |
| 13:05:58 | bhagyashri_s | mriedem, jaypipes, dansmith: Hi, Just wanted to ask the future plans regarding resources usages related (cpu/disk/memory used) information in the compute_nodes db table, Is there any plan to remove this resource usages related information from compute_nodes db table, because till now what I observed is: resource usages are stored in allocations db table at placement side, so same thing needs to be store again in compute_nodes table as well. | |
| 13:09:20 | jaypipes | bhagyashri_s: excellent question :) I would really like to remove the usage fields from the compute_nodes table for those resource classes. however, we still rely on those fields unfortunately for some Compute REST API results. | |
| 13:09:41 | jaypipes | bhagyashri_s: until we rewrite those REST API pieces to use placement, we can't get rid of them. | |
| 13:10:01 | jaypipes | bhagyashri_s: lemme grab some links for you... one sec | |
| 13:11:37 | claudiub|2 | artom: hello. around? | |
| 13:11:39 | jaypipes | bhagyashri_s: tada: https://github.com/openstack/nova/blob/master/nova/api/openstack/compute/hypervisors.py#L54 | |
| 13:12:00 | jaypipes | bhagyashri_s: that compute REST API endpoint (/os-hypervisors) need to just go away... :) | |
| 13:12:20 | jaypipes | bhagyashri_s: and essentially deprecated in favor of the placement API. | |
| 13:12:32 | bhagyashri_s | jaypipes: ohh ok. | |
| 13:12:36 | jaypipes | bhagyashri_s: which has the /usages endpoint | |
| 13:12:40 | jaypipes | bhagyashri_s: make sense? | |
| 13:12:52 | bhagyashri_s | jaypipes: yes | |
| 13:13:00 | jaypipes | bhagyashri_s: note that the compute API's GET /os-hypervisors endpoint is an admin-only hting | |
| 13:13:02 | jaypipes | thing | |
| 13:13:12 | jaypipes | as is the placement API. | |
| 13:16:07 | bhagyashri_s | jaypipes: so is there any plan in future to deprecate that | |
| 13:19:50 | bhagyashri_s | jaypipes: and as you mentioned for compute rest api endpoint like /os-hypervisors we will get info from placement API. | |
| 13:19:56 | artom | claudiub|2, yeah - give me, like, 30 minutes to drive kids to daycare | |
| 13:21:48 | jaypipes | bhagyashri_s: it's not been thoroughly discussed, no... probably a good thing to bring up on the mailing list and discuss in Denver PTG | |
| 13:22:10 | bhagyashri_s | jaypipes: so that resource usages related columns in the compute_node table will not be needed any more | |
| 13:22:11 | mriedem | gibi: i don't see much advantage in mocking out the FakeDriver here rather than just stubbing out the manager method https://review.openstack.org/#/c/498627/1/nova/tests/functional/test_servers.py | |
| 13:23:56 | bhagyashri_s | jaypipes: ok. Thank you for info and sure will discussed in Denver PTG. | |
| 13:24:04 | bhagyashri_s | jaypipes: :) | |
| 13:33:07 | lpetrut | Hi, we have an issue with nova assisted volume snapshots when a separate DB is being used for each cell. The API only accepts a volume id, fetching the instance id from the BDM. The issue is that the BDM object cannot be retrieved as the nova api service is looking in the wrong DB (not knowing which cell to use). | |
| 13:34:26 | lpetrut | I thought you guys may provide some input :) relevant nova api code snippet: https://github.com/openstack/nova/blob/5d3a11b9c9a6a5aecd46ad7ecc635215184d930e/nova/compute/api.py#L4076-L4078 | |
| 13:35:24 | gibi | mriedem: mocking the manager method changes that part of nova we try to verify. While mocking the FakeDriver would only change the behavior of the test env. I also OK with way you did in your refactor patch by adding FakeDriver variants with a good class doc | |
| 13:35:26 | mriedem | lpetrut: ah very possible. this doesn't grab the instance first, but if it did, the _get_instance method in the compute API would set the cell mapping in the context | |
| 13:36:13 | mriedem | lpetrut: can you report a bug? | |
| 13:36:37 | gibi | mriedem: ahh I see why you didn't added a FakeDriver variant. | |
| 13:36:40 | lpetrut | mriedem: sure. the issue is that at this point, the instance id is unknown as it's not passed in the api call. | |
| 13:36:59 | mriedem | lpetrut: yeah, which sucks, so we'll likely have to iterate the cells looking for the bdm | |
| 13:37:12 | mriedem | and once found, target the context for the rest of the calls | |
| 13:37:31 | mriedem | we wouldn't have hit this in CI because only certain drivers, like NFS, test this API | |
| 13:37:42 | mriedem | there is an NFS job, but it's non-voting and in the experimental queue i think | |
| 13:38:09 | mriedem | gibi: right i'd have to create a new test class to use a new virt driver | |
| 13:38:16 | openstackgerrit | Viktor Varga proposed openstack/nova master: Ensure image metadata when unshelving BFV instance https://review.openstack.org/498729 | |
| 13:38:18 | mriedem | gibi: i can see what mocking out the driver method instead looks like | |
| 13:38:38 | lpetrut | mriedem: yep, that one's failing because of this: http://logs.openstack.org/21/498321/5/check/gate-tempest-dsvm-full-devstack-plugin-nfs-nv/b8bca96/logs/screen-c-vol.txt.gz?level=TRACE | |