Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2018-12-11
16:35:21 mriedem it == the api thinking we have old api services and looking up instances from cell0 rather than the actual cell they are in
16:35:54 mriedem unless coincidentally those instances *are* in cell0
16:36:11 mriedem or tempest just doesn't care about a 404
16:37:37 openstackgerrit Jack Ding proposed openstack/nova-specs master: Select cpu model from a list of cpu models https://review.openstack.org/620959
16:39:36 cdent mriedem: ah, okay, so you're going to log all the service creations to fill in some blanks: who is doing this!? roger
16:39:44 mriedem yeah
16:40:30 cdent good luck
16:40:45 cdent i'm trying to get the mysql opportunistic to fail, but struggling
16:43:55 dansmith mriedem: replied
17:03:35 mdbooth mriedem: Here's an interesting thing. You noted the other day that setting created_disks = False doesn't make the evacuate tests fail. After working it through it turns out that's correct behaviour for all 4 cases, and if you want to make that fail something you'd need to make the initial spawn() fail instead... Except that a spawn failure actually calls cleanup *twice*.
17:04:35 mdbooth Basically I added a similar functional test making spawn fail instead, and I was trying to debug why my Rbd disks were being deleted anyway when I was setting created_disks = False.
17:05:23 openstackgerrit Stephen Finucane proposed openstack/nova master: Fix typo https://review.openstack.org/624442
17:05:24 mdbooth It's because we call _shutdown_instance() in ComputeManager, which calls driver.destroy()
17:05:36 mdbooth mriedem: Which leaves that code in a weird limbo.
17:06:52 mdbooth That is, the code ensures the correct behaviour of _cleanup_failed_start(), but it's irrelevant because we call it again anyway.
17:10:14 mdbooth mriedem: Yeah.
17:10:29 mriedem my goal is to add tests that don't pass without the code fix,
17:10:33 mriedem so however that needs to happen...
17:11:25 mdbooth mriedem: That's kinda the thing. I wrote those, but because the code structure is weird they still didn't fail. However, if you took the code out, the resulting code would be fragile.
17:11:29 openstackgerrit Stephen Finucane proposed openstack/nova master: objects: Store InstancePCIRequest.numa_policy in DB https://review.openstack.org/624444
17:11:29 openstackgerrit Stephen Finucane proposed openstack/nova master: conf: Call out where pci.alias should be set https://review.openstack.org/624445
17:11:33 stephenfin sean-k-mooney: ^
17:11:48 mdbooth mriedem: Explaining why you *didn't* include the code would involve understanding ^^^
17:11:56 mdbooth And nobody wants to do that
17:11:57 stephenfin I feel like a moron. I've no idea what I did to miss that the first day
17:12:46 mdbooth mriedem: I'll have another go at it tomorrow.
17:19:44 openstackgerrit Jack Ding proposed openstack/nova master: Improve libvirt image and snapshot handling https://review.openstack.org/616692
17:20:39 openstackgerrit Jack Ding proposed openstack/nova master: Correct instance port binding for rebuilds/reboots https://review.openstack.org/603844
17:23:02 jackding hi mriedem: could you revisit https://review.openstack.org/#/c/616692/ and https://review.openstack.org/#/c/603844/ when you got a chance?
17:23:41 sean-k-mooney stephenfin: isnt the alais ment to be set in the schduler not the api
17:26:55 stephenfin sean-k-mooney: Nope, https://github.com/openstack/nova/blob/master/nova/compute/api.py#L815-L816
17:27:11 stephenfin It's nova-api that converts the flavor extra spec to an InstancePCIRequests object
17:27:56 sean-k-mooney ok
17:28:03 stephenfin and unfortunately also the resource tracker, hence nova-compute https://github.com/openstack/nova/blob/master/nova/compute/resource_tracker.py#L307-L309
17:28:25 sean-k-mooney i knew this had to be set on the controller but i jsut assumed it was teh scduelr that read it
17:28:39 sean-k-mooney if its the api that is fine as long as operators know that too
17:29:04 sean-k-mooney the resouce tracker should not need the alias definion
17:29:16 sean-k-mooney at least not in the compute agent config
17:30:17 sean-k-mooney uhg
17:30:30 stephenfin Yup, it's icky
17:30:34 sean-k-mooney damit it does
17:30:44 sean-k-mooney that is defintely not documented
17:31:21 stephenfin It actually is https://docs.openstack.org/nova/latest/admin/pci-passthrough.html#configure-pci-devices-compute
17:31:26 stephenfin But it's hidden
17:31:30 stephenfin sort of
17:31:36 stephenfin "From the Newton release, to resize guest with PCI device, configure the PCI alias on the compute node as well."
17:31:57 sean-k-mooney stephenfin: the whitelist should be set on the compute and alais on the contoller that is how i have understood it was ment to work
17:32:07 stephenfin Me too
17:33:01 sean-k-mooney your updating the config docs mind patching https://docs.openstack.org/nova/latest/admin/pci-passthrough.html#configure-pci-devices-compute to make that bold
17:34:23 sean-k-mooney luckally i was seting it on all nodes just because i adapted my controler local.conf into my compute node one
17:34:56 sean-k-mooney stephenfin: speaking of which do you have 2 node capably of sriov or one?
17:35:20 stephenfin sean-k-mooney: Can do, and I only have one
17:35:53 sean-k-mooney ok i have spent most of the day debuging a netwrokg issue
17:36:29 sean-k-mooney im going to need to use a node inbetween my current 2 nodes to sniff packets and see which side is broken
17:37:46 sean-k-mooney adrianc: FYI at least on my hardware the sriov live migration patches are not working correctly
17:38:58 adrianc sean-k-mooney: what are you seeing ?
17:39:22 sean-k-mooney adrianc: after migrating the vm unicast packet from the source node to the vm are droped
17:39:58 sean-k-mooney in my case i have an all in one node and a compute node
17:40:17 sean-k-mooney the vm started on teh all-in-one node and i migrated it to the compute-only node
17:40:41 openstackgerrit Matt Riedemann proposed openstack/nova master: Handle tags in _bury_in_cell0 https://review.openstack.org/621856
17:40:57 adrianc sean-k-mooney: on the source the VF used for macVtap has MAC address of zero ?
17:41:01 sean-k-mooney i can see the packets being transmited with tshark on teh all-in-one node but they are not recived on the destination
17:41:30 sean-k-mooney adrianc: ill check it not set to the vm mac any more but not sure its 0
17:42:24 sean-k-mooney adrianc: its set to ba:2a:94:20:f7:ba
17:42:53 sean-k-mooney the vm mac is fa:16:3e:0e:87:08
17:42:54 adrianc sean-k-mooney: so not an openstack VM MAC
17:43:01 sean-k-mooney correct
17:43:29 adrianc sean-k-mooney: mac appears well set on destination VF ?
17:43:50 sean-k-mooney yep both the macvtap and the dest vf have the mac set
17:44:20 sean-k-mooney if i reboot the vm the dhcp request from the vm is recived on the other node and it replys
17:44:40 adrianc sean-k-mooney: so the code does what its suppose to do :) now need to figure out the issue
17:44:48 sean-k-mooney however the unicast dhcp offer is transmitted on the all-in-one node but not recived on the compute
17:45:08 sean-k-mooney well you can live migrate back and forth
17:45:53 sean-k-mooney if i migate the vm back the networking works agiag
17:45:53 adrianc sean-k-mooney: mmmm, maybe mac is still in FDB on the source ?
17:46:09 sean-k-mooney adrianc: i checked with bridge fdb show and no
17:46:37 sean-k-mooney it is actully never there
17:46:57 sean-k-mooney the ipv6 multicast mac is set on the host that has the vm 33:33:ff:0e:87:08
17:48:03 sean-k-mooney but fa:16:3e:0e:87:08 never show up in bridge fdb show either when the vm is sawned or before or after migration
17:48:32 adrianc sean-k-mooney: let me try to repro on my setup
17:52:50 sean-k-mooney adrianc: sure no worries im going to grab something to eat
17:54:18 adrianc sean-k-mooney: probably some minor tweaking needed (i hope) my setup seems busted, ill take a look tomorrow (20:00 here)
17:54:39 sean-k-mooney adrianc: fyi if i assign an ip on the same subnet to the pf i can ping the vm on the host its on
17:54:50 sean-k-mooney adrianc: sure no worries
17:54:57 openstackgerrit Chris Dent proposed openstack/os-resource-classes master: Initial automated testing for os-resource-classes https://review.openstack.org/624366
17:54:58 openstackgerrit Chris Dent proposed openstack/os-resource-classes master: Rename ORDERED_CLASSES to STANDARDS https://review.openstack.org/624369
17:54:59 openstackgerrit Chris Dent proposed openstack/os-resource-classes master: Sync STANDARDS with placement/nova rc_fields https://review.openstack.org/624370
17:55:00 openstackgerrit Chris Dent proposed openstack/os-resource-classes master: Tune up documentation to be more useful https://review.openstack.org/624384
17:55:01 adrianc bon appetit
19:12:14 melwitt mriedem: not sure what you'll think of this, workaround for non-shared instance dirs not being cleaned up during evacuate with ceph https://review.openstack.org/618478
19:31:32 mriedem melwitt: i think it's ugly
19:32:13 mriedem comments inline anyway
19:32:24 melwitt I think it's ugly too, which is why I've struggled with it
19:32:55 melwitt I can't think of another way to handle it in a backportable way
19:47:24 openstackgerrit Merged openstack/nova stable/queens: Create BDMs/tags in cell with instance when over-quota https://review.openstack.org/623934
19:50:43 openstackgerrit Matt Riedemann proposed openstack/nova master: DNM: add debug logging for bug 1807723 https://review.openstack.org/624181
19:50:43 openstack bug 1807723 in OpenStack Compute (nova) "swap multiattach volume intermittently fails when servers are on different hosts" [Medium,Confirmed] https://launchpad.net/bugs/1807723
20:12:55 melwitt mriedem, dansmith, gibi: there's a proposal to switch the os-traits, os-resource-classes, and osc-placement projects to use the placement.config gerrit acls here https://review.openstack.org/624387 to add your opinion if you have one
20:13:31 codyhill Hello folks. I was working with the #openstack-ironic folks about trying to get vendor_data.json to work with baremetal deployments with Nova and Ironic. So far the issue i’m having is that cloud-init doesn’t seem to “Discover” the baremetal node as an OpenStack Datasource. So it runs the generice EC2 setup. Which doesn’t include vendor_data.json. The recomendation from the folks in #openstack-ironic was to try and use config drive. And it
20:13:31 codyhill looks like the config-drive doesn’t write the vendor_data.json to the config-drive. Any help would be great.

Earlier   Later