Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-08-23
15:29:41 mriedem nope
15:29:55 CeeMac that answers that then
15:29:56 mriedem it's come up many a time
15:30:20 CeeMac so, the only option for shared storage backend is to mount the instance disk path to NFS?
15:30:48 mriedem you're talking about different things i think
15:30:56 CeeMac its entirely possible
15:31:12 openstackgerrit Balazs Gibizer proposed openstack/nova master: Make _rever_allocation nested allocation aware https://review.opendev.org/676138
15:31:27 mriedem the instance path files are not the same as the root disk - you can have the root disks in rbd but have the instance files (console log, config drive, etc) be on local disk on the compute node
15:31:29 mriedem it gets confusing
15:31:43 CeeMac i don't use ceph, if that helps
15:31:48 dansmith artom: so, what do you think the timetable is for a respin of the numalm set? you said you've got unit tests fixed locally already, so can you do some swizzling of the top two patches and push that up today to get a run started?
15:31:50 mriedem or you could have the instance path files on local disk or shared NFS/rbd, but the root disk is actually a volume in cinder
15:32:13 CeeMac i was thinking in terms of when an instance is launched without specifying to use a cinder volume
15:32:19 mriedem CeeMac: sure, just saying that's a common shared storage backend
15:32:21 artom dansmith, working on updating the func test currently, since confidence seems to be the big thing
15:32:36 CeeMac mriedem: got you, yeah its all very confusing.
15:33:09 mriedem CeeMac: assuming you're using libvirt, the image backend for the root disk is defined with this config https://docs.openstack.org/nova/latest/configuration/config.html#libvirt.images_type
15:33:25 mriedem so you're probably using raw/flat/qcow2 yeah?
15:33:31 openstackgerrit Balazs Gibizer proposed openstack/nova master: Support reverting migration / resize with bandwidth https://review.opendev.org/676140
15:33:42 CeeMac qcow2
15:33:55 CeeMac but i back images off onto volume, so it converts it raw iirc
15:34:51 mriedem https://docs.openstack.org/nova/latest/configuration/config.html#DEFAULT.instances_path controls where the instance files are going to be then
15:35:14 CeeMac right
15:35:38 CeeMac thats what I was trying to get at i think
15:35:52 openstackgerrit Balazs Gibizer proposed openstack/nova master: Func test for migrate re-schedule with bandwidth https://review.opendev.org/676972
15:36:10 CeeMac when an instance is launched without specifying to use volume backing for root disk, the root disk will go there too by default?
15:36:21 CeeMac i.e. not through horizon
15:36:53 mriedem horizon doesn't have anything to do with it (unless you mean horizon is somehow forcing you to use volume-backed servers)
15:36:55 mriedem but yeah
15:37:21 mriedem https://review.opendev.org/#/c/511965/ has some background details on the whole "configure nova to do volume-backed servers by default" thing which has had lots of discussion in the past like i said
15:37:26 mriedem many attempts, many failures
15:37:35 sean-k-mooney horizon default to createing a volume if cinder is enabled
15:37:43 mriedem sean-k-mooney: ok but it's not required
15:37:46 mriedem nor does everyone use horizon
15:37:51 sean-k-mooney im not really sure why but it does. its not require correct
15:38:10 CeeMac sean-k-mooney: thats my understanding too
15:38:14 openstackgerrit Balazs Gibizer proposed openstack/nova master: Support migrating SRIOV port with bandwidth https://review.opendev.org/676980
15:38:15 openstackgerrit Balazs Gibizer proposed openstack/nova master: Allow migrating server with port resource request https://review.opendev.org/671497
15:38:29 sean-k-mooney CeeMac: it defintly does not require it as i untick it all the time
15:38:44 CeeMac my issue comes when using rancher to deploy VMs, socker-volume isn't block-storage aware
15:39:02 CeeMac I'm happy with horizon defaulting to volume, as thats where I'd prefer for the instance to go
15:39:22 CeeMac trying to work out a way to force all instances to back off to block storage regardless of how they're requested
15:39:29 CeeMac was hoping there was an integrated cinder way
15:39:30 sean-k-mooney CeeMac: i kind of wish it default to delete the volume when you delete the vm too however
15:39:53 CeeMac but I'm guessing I'd need to go with mounting /var/lib/nova/instances to NFS on our SAN
15:40:15 CeeMac which is where instances_path points to right now
15:40:19 CeeMac in nova.config
15:40:20 sean-k-mooney CeeMac: i think we have a config option that will help
15:40:37 sean-k-mooney https://docs.openstack.org/nova/latest/configuration/config.html#DEFAULT.max_local_block_devices
15:40:51 sean-k-mooney if you set that to 0 you have to use cinder volumes
15:40:58 sean-k-mooney not sure if tha tis what you want
15:41:39 CeeMac sean-k-mooney: that is a good start to prevent local disks being created and potentially filling up the (small) local disks on the compute
15:41:49 sean-k-mooney the other options open to you if you nova no/limnited local stoage and have ceph is to enable the RBD image backend
15:41:57 CeeMac but I haven't found a way to instruct rancher, or docker for that matter, to use volumes for images
15:42:09 CeeMac sean-k-mooney: sadly no ceph here
15:42:17 CeeMac iscsi backed cinder
15:42:26 sean-k-mooney ah ok
15:42:52 sean-k-mooney we dont have a cinder or iscis image backend to force all instace to use cinder unfortunetly
15:43:01 CeeMac is NFS the only/best/supported way of backing off the instance_path, or is iscsi an option?
15:43:27 CeeMac haha, i think you answered that question then.
15:43:33 sean-k-mooney i think you would be stuck with nfs
15:44:12 CeeMac you could be right, at least for the short term until I can get/hack a workaround for rancher to be able to call for volume on instance launch
15:44:48 CeeMac thanks for the help mriedem and sean-k-mooney appreciate it
15:44:49 sean-k-mooney rancher use docker to run all service rihgt
15:44:58 sean-k-mooney i have not looked at rahcner os much
15:45:12 CeeMac yeah
15:45:16 sean-k-mooney but i think you can confugre docer to use remote sotorage
15:45:20 CeeMac and docker-volume isn't block storage aware
15:45:32 CeeMac you feed it a flavour and image effectively
15:45:59 sean-k-mooney e.g. i think you should be able to have the non system docker use a volume plugin and mont /var/libvirt/instace... or what ever on a network share
15:46:44 CeeMac if rancher was running inside openstack yeah, but we're connecting to openstack through api to a project from outside
15:47:07 CeeMac i'll have a ponder
15:47:40 sean-k-mooney i know there used to be a tone of voluem plugins https://docs.docker.com/engine/extend/legacy_plugins/#volume-plugins but i dont know if any of thos worked with racner/still exists
15:48:16 CeeMac i've seen a cinder driver for docker on github, but havent had a chance to test that yet
15:49:01 mriedem sean-k-mooney: https://docs.openstack.org/nova/latest/configuration/config.html#DEFAULT.max_local_block_devices is not what you think it is
15:49:31 mriedem sean-k-mooney: that's only if the bdm has destination_type='local'
15:49:45 mriedem unless you mean for the implicit image-defined bdm
15:49:48 sean-k-mooney it prevent you spaning flavor tha trequest ephmeral/swap or root disk if you are using qcow
15:49:56 mriedem source_type='image', destination_type='local'
15:50:14 sean-k-mooney ya i ment you can use that to ensure only bfv guest can boot on that node
15:50:29 mriedem *any* node
15:50:36 mriedem it's config-driven api
15:50:49 sean-k-mooney yeah thos are awsome... but ya
15:51:52 sean-k-mooney i was looking to do that in the past but since i was deploying ceph for cinder i just deploy with the rbd image backend instead
15:57:51 sean-k-mooney mriedem: im going to step away for an hour or so to have breakfast/lunch but i plan to adress your comments on https://review.opendev.org/#/c/669738/ when i get back since they have been waiting a while.
15:58:49 mriedem breakfast/lunch?
15:58:59 mriedem it's 5pm
15:59:00 sean-k-mooney lyarwood: mriedem if we can land https://review.opendev.org/#/c/678098/ on stable stine that would make mnaser happy
15:59:37 sean-k-mooney yes.. i normally eat once to twice a day. usually i try to grab something at around 3
15:59:48 sean-k-mooney and then around 8
16:00:28 sean-k-mooney or i jsut have one big meal around 6 and then snack on fruit/nuts/small things
16:02:48 mriedem i just got the image of you as the giant in jack and the beanstalk crushing human bones to make your bread
16:02:50 mriedem "small things"
16:04:12 dansmith artom: I looked at the top two patches with plans to prescribe an ordering and squash/split instructions, but I think it probably just needs to be fully squashed
16:04:53 artom dansmith, yeah? I was hoping to avoid that, as it makes reviewing difficult
16:04:56 dansmith I know
16:05:10 dansmith I kept calling out lines as "this could be separate" and then removed those comments
16:05:26 artom But you're most likely right (duh!), as a whole bunch of things need to happen in one chunk
16:06:17 dansmith yup, any ordering that I could come up with was dependent on a patch ahead of everything that was like "add a bunch of single-line unrelated changes to various places" followed by a bigger one that hooked them all up
16:06:21 dansmith and it just doesn't make sense

Earlier   Later