Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2021-11-11
14:50:37 gibi lyarwood: I use these in my local conf https://paste.opendev.org/show/810945/
14:51:09 sean-k-mooney i dont think thats actully sufficent now
14:51:26 sean-k-mooney unless they have fixed it recently
14:51:37 gibi sean-k-mooney: I build devstack with this last week I think
14:51:41 sean-k-mooney Q_AGENT=openvswitch did not disable qemu last time i check
14:51:49 sean-k-mooney ok so them might have fixed it then
14:52:14 gibi lyarwood: I added console dumping before and after the sleep that happens before detach https://review.opendev.org/c/openstack/tempest/+/817560/1/tempest/api/compute/base.py
14:52:30 lyarwood gibi: ack cool thanks
14:53:03 lyarwood sean-k-mooney: yeah thanks, I think the local issue was with an older version of ovn as the f34 jobs are passing upstream
14:53:12 lyarwood die 174 'Socket /var/run/openvswitch/ovnnb_db.sock not found'
14:53:14 lyarwood maybe not
14:53:16 lyarwood urgh
14:53:35 sean-k-mooney it might be selinux
14:53:57 sean-k-mooney devstack used to complie form source in the past to make ovn work
14:54:09 lyarwood $ getenforce
14:54:09 lyarwood Permissive
14:55:06 lyarwood https://paste.centos.org/view/2cd83b46 - I get the feeling devstack is just looking for the wrong socket somehow
14:55:54 sean-k-mooney am if i remmeber they create a symlink
14:56:05 sean-k-mooney they bing devstack
14:56:36 sean-k-mooney devstack was simlink /run/ovn to /run/openvswtch
14:56:43 sean-k-mooney so that they were all in one folder
14:56:50 sean-k-mooney so ys that proably is broken
14:58:03 sean-k-mooney https://github.com/openstack/devstack/blob/master/lib/neutron_plugins/ovn_agent#L362
14:59:22 sean-k-mooney lyarwood: is /run a symlink of /var/run on your system or the other way around
14:59:35 sean-k-mooney it shold be but if not that would be the issue
14:59:58 sean-k-mooney devstack is assuming that /run and /var/run are the same location
15:00:08 lyarwood /var/run is a symlink to /run
15:00:43 lyarwood I'll just nuke this env and start again
15:00:49 lyarwood but thanks for the pointers
15:01:02 sean-k-mooney ok then you shoudl see. /run/ovn/ovnnb_db.sock at /run/openvswitch/ovnnb_db.sock
15:01:23 sean-k-mooney this was broken on one of my setup
15:01:39 lyarwood yeah it wasn't there
15:01:41 sean-k-mooney i had to fully uninstall openvswtich and ovn
15:01:47 lyarwood the symlink was missing
15:02:02 lyarwood I'm just going to reimage the vm and start again
15:02:07 sean-k-mooney then remove the /run/openvswtch and /run/ovn dirs and the /var/run versions
15:02:09 sean-k-mooney then it worked
15:02:48 sean-k-mooney well you can but i would jsut remove the directores and packages and try stacking one more time first
15:04:02 sean-k-mooney i had to do this to get vdpa working on my home server on ubuntu about 3 weeks ago but its been fine since then
15:25:27 EugenMayer sean-k-mooney i made ESXi work finally, interestingly, one cannot install it on openstack but i can install it under my local libvirt and thus use a qcow of the prininstalled variant on openstack, see https://stackoverflow.com/questions/69924901/how-to-run-esxi-on-openstack-as-kvm-vm - what wonders me is, that the installer should theoratically
15:25:28 EugenMayer work under openstack too. I mean i use the same e1000/e1000e nics and is use libvirt qemu/kvm locally as for the hv in nove. Odd
15:26:21 EugenMayer Kind of unexpected "i cannot explain" why e1000e works for 7.x when booting the non installer, but not for the installer under openstack.
15:43:14 sean-k-mooney e1000e and e1000 are actully different nics
15:43:25 sean-k-mooney on is a pci nic and the other is pcie
15:43:40 sean-k-mooney but i have never had much luck using e1000e with openstack
15:43:48 sean-k-mooney you need to use the q35 manchine type
15:43:53 sean-k-mooney to get a pcie bus
15:43:59 sean-k-mooney perhaps that is the problem
15:44:09 sean-k-mooney the default machine type is pc-ix440
15:44:19 sean-k-mooney which doen not have pcie only pci
15:48:40 EugenMayer that is an interesting point
15:48:59 EugenMayer but since i do not set this via meta-data it does not make sense that the qcow2 variant works and the other does not, right
15:49:53 kevko sean-k-mooney: hi, maybe I found something weird in my last discussed problem if you remember ...
15:50:10 EugenMayer if using q35 fixes it, i must have used it on my local as on the openstack libvirt. For the latter i did not. The properties i used for running the qcow or iso image where the same:
15:50:10 EugenMayer --property hw_disk_bus=sata --property hw_cpu_policy=dedicated --property img_config_drive=mandatory --property hw_cdrom_bus=ide --property hw_vif_model=e1000 --property hw_boot_menu=true --property hw_qemu_guest_agent=no
15:50:20 EugenMayer or for 7.x
15:50:25 EugenMayer --property hw_disk_bus=sata --property hw_cpu_policy=dedicated --property img_config_drive=mandatory --property hw_cdrom_bus=ide --property hw_vif_model=e1000e --property hw_boot_menu=true --property hw_qemu_guest_agent=no
15:52:57 opendevreview mitya-eremeev-2 proposed openstack/nova master: Close Glance image if downloading failed. https://review.opendev.org/c/openstack/nova/+/815347
16:20:03 gmann lyarwood: gibi sean-k-mooney on tempest things to ssh things. we have two things here 1. all API tests are configurable to pause SSHing via run_validation config option 2. all scenario tests run SSH validation and does not care of run_validation config as most of scenario tests need to ssh
16:21:24 lyarwood gmann: I'm not sure what you mean by 1. do you have an example?
16:22:13 lyarwood what we are looking for here is a check during creation before any additional actions or validations actually happen
16:22:42 lyarwood I've got a WIP locally that I can finish up this evening to show what I mean
16:24:36 gmann lyarwood: we do access remote client only if run_validation is true in API tests
16:24:41 gmann lyarwood: but I think this does not check https://github.com/openstack/tempest/blob/7e96c8e854386f43604ad098a6ec7606ee676145/tempest/common/compute.py#L270
16:25:08 gmann lyarwood: we should have this with condition CONF.validation.run_validation
16:25:12 lyarwood yeah I'm nuking that implementation FWIW
16:25:27 lyarwood parsing the console it's valid IMHO
16:25:46 opendevreview Dan Smith proposed openstack/nova master: WIP: Revert project-specific APIs for servers https://review.opendev.org/c/openstack/nova/+/816206
16:26:09 dansmith gmann: johnthetubaguy[m] ^ wasn't too bad to add that new scenario, and it helps to be able to see what works and how in that case
16:26:48 gmann lyarwood: +1 thanks, ping me once you are ready with that or even we can add new config option also in case we do not want all tests doing ssh things and do not want to wait for console ?
16:27:06 gmann dansmith: thanks, I will check it after lunch today.
16:27:39 lyarwood gmann: I'm overloading wait_until to accept PINGABLE and SSHABLE, both basically fallback to ACTIVE if validations aren't enabled
16:28:12 lyarwood gmann: but when they are we wait first until the server is ACTIVE and then pingable or sshable etc.
16:32:01 gmann lyarwood: yeah, currently this method make server pingable or sshable if validatable=True but not do actual ping/ssh and with the new way you are making wait_until to do actual ssh/ping and after Active state ?
16:42:34 lyarwood gmann: sorry on a call, yeah correct so wait_until=PINGABLE would wait until the server is ACTIVE and will then wait until it is pingable *if* validatable=True etc
16:42:56 lyarwood gmann: if not it just waits until it's ACTIVE and continues
16:43:12 gmann lyarwood: +1
17:01:44 opendevreview mitya-eremeev-2 proposed openstack/nova master: Close Glance image if downloading failed. https://review.opendev.org/c/openstack/nova/+/815347
17:13:54 gibi lyarwood, sean-k-mooney: the first run with the 30 sec sleep before detach finished and did not generated any kernel panic https://zuul.opendev.org/t/openstack/build/06cb831dd6c94a2fa8ca373662014dd2/logs
17:14:25 gibi but also the guest did not print anything to the console during that extra 30 seconds
17:14:36 gibi so no hard proof yet that we need to wait
17:14:54 gibi I will recheck it
17:15:26 sean-k-mooney ack
17:15:59 sean-k-mooney so you wait for it to be active then sleep 30 right
17:16:03 sean-k-mooney then attach and detach
17:17:10 gibi nope, the failures are happening during test class cleanup
17:17:13 sean-k-mooney oh you put it in the detach
17:17:18 gibi so I just wait before the detach
17:18:00 sean-k-mooney oh its not in the actual detach test
17:18:14 sean-k-mooney its in the cleanup of a test that did attach?
17:18:45 gibi yes
17:18:56 sean-k-mooney i see why not just remove https://review.opendev.org/c/openstack/tempest/+/817560/1/tempest/api/compute/base.py#619
17:18:58 gibi created server, attached volume, did a live migration and then cleanup
17:20:01 sean-k-mooney we want to delete the voluem after the test sure but we dont need to detach it just delete the vm
17:22:00 gibi I'm not that familiar with tempest that I can say it is always OK not to detach as volume create and volume attach are two seperate action from the test case
17:26:32 lyarwood yeah volume attach always adds in the detach cleanup in tempest
17:26:53 lyarwood I've talked about removing that in the past but people didn't want to drop the coverage of detach etc
17:28:25 sean-k-mooney which given we have detach test i dont really get
17:28:41 sean-k-mooney we did talk about it for the compute test before definetly
18:12:01 opendevreview Balazs Gibizer proposed openstack/nova master: Add debug log for scheduler weight calculation https://review.opendev.org/c/openstack/nova/+/817632

Earlier   Later