Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2021-11-05
10:45:39 EugenMayer is there really no better way or even usual default to start vms after a node reboot then adding resume_guests_state_on_host_boot to the nova conf and/or using virsh to add the autostart flag? There is no GUI nor nova API flag (openstack server create) flag to automatically create an instance with autostart?
10:50:55 opendevreview Balazs Gibizer proposed openstack/nova master: Use ReplaceEngineFacade fixture https://review.opendev.org/c/openstack/nova/+/816820
10:52:08 opendevreview Balazs Gibizer proposed openstack/nova master: Fix interference in db unit test https://review.opendev.org/c/openstack/nova/+/814735
10:53:39 gibi EugenMayer: I think nova considers such automatic recovery as mostly outside of nova scope. An external service can detect the compute host failre and can implement automatic evacuation for some VMs while for others it can wait for the compute host recovery and implement auto startup.
10:55:17 EugenMayer gibi interesting. I understand that this is an good option, but it would be somewhat nice to have this as a possible build in default to, or?
10:55:53 EugenMayer but i understand, if you run hundreds of VMs the strategy to recover could be more suffistacted
10:56:04 gibi EugenMayer: OpenStack has tools to build such behavior top of nova. I.e. https://docs.openstack.org/self-healing-sig/latest/use-cases/heat-mistral-aodh.html
10:56:56 gibi the use case is vaild, I just don't think the implementation needs to be inside nova
10:58:00 EugenMayer thank you for that article, will look into that.
11:07:23 opendevreview Lee Yarwood proposed openstack/nova master: nova-next: Deploy noVNC from source instead of packages https://review.opendev.org/c/openstack/nova/+/816738
11:09:56 opendevreview Kashyap Chamarthy proposed openstack/nova master: libvirt: Introduce config classes for QEMU's "tb-cache" https://review.opendev.org/c/openstack/nova/+/816823
11:10:27 kashyap Huh, plural; it's a single class
11:12:05 opendevreview Kashyap Chamarthy proposed openstack/nova master: libvirt: Introduce config class for QEMU's "tb-cache" https://review.opendev.org/c/openstack/nova/+/816823
11:14:10 kashyap Not sure if the above WIP requires a bp yet ... but I filed one prememptively (https://blueprints.launchpad.net/nova/+spec/control-qemu-tb-cache)
11:35:54 gibi kashyap: I think a specless bp is enough
11:36:44 kashyap gibi: Cool; guessed as much. :)
11:41:54 sean-k-mooney EugenMayer: https://docs.openstack.org/masakari/latest/ is really the service you likely want to have automatic evacuation of ha instnaces
11:42:31 sean-k-mooney you can manually do it with mistal heat and aodh but masakari is a single service desinged to provide instance ha
11:43:09 EugenMayer migrating from proxmox to openstack, i have VMs with multiple disks. I understand that i can import disks as images using 'openstack image create --import' - but how to create a multi-disk VM?
11:43:44 EugenMayer sean-k-mooney interesting. Currently, since we are not planning in cinder, auto-evacuation is not possible anyway
11:44:19 sean-k-mooney EugenMayer: the imporant thing to remember however for the "core" service like nova is that openstack design principal is not based on a declaritve model its inparitive. e.g. it is not intend to take action autonomusly only when you interact with the system. so nova should never alter the state of a vm unless you make an api call
11:44:19 EugenMayer sean-k-mooney beside doing it from backup, which is rather a manual decision to do so. So not as cloudish as masakari would do it
11:45:02 sean-k-mooney EugenMayer: i hesitate to say this cause i dislike this code path but without cinder you can use masikari if you put the instnace state dir on an nfs share
11:45:04 EugenMayer sean-k-mooney understood. So it is more an API wrapper with high-level tasks, but it does not just act, that is for others to implement
11:45:17 sean-k-mooney EugenMayer: correct
11:45:25 EugenMayer sean-k-mooney nfs share is nothing else as cinder on horrible storage :)
11:45:59 sean-k-mooney yep and it used a diffeent code path then the one we normally use so less well tested
11:46:05 sean-k-mooney which is why its better to avoid it
11:46:15 EugenMayer right now, we go for local disks only, at least for the legacy VMs. K8s cluster is not yet decided. So most of the evacuate features / hot/live migration are not for us with the legacy vms
11:46:22 sean-k-mooney and the performance sucks vs a real shared stoage system
11:46:35 sean-k-mooney so back to your multi disk question
11:46:54 sean-k-mooney you can create multi disk vms but its not how nova is typically used
11:46:54 EugenMayer the performance is the main / only reason we are not jumping on ceph or similiar. We have to be realistic, our network is 1GB
11:47:08 EugenMayer with an MTU of 1400, no jumbo or anything
11:47:16 sean-k-mooney in the falvor there are 3 storage options disk, ephemeral and swap
11:47:32 sean-k-mooney disk is the root disk size
11:48:30 EugenMayer can i create VMs manually with any amount of disks since it's virsh in the end? I will only need this for the VMs i migrate - the ones i create from scratch are based on cloud-init and the proper disks you get using nova directly
11:48:32 sean-k-mooney ephemeral is the total amount of addtional ephemeral storage, i belive by default if you dont otherwise say all the ephemeral storage will be created as a singel addtional disk
11:48:52 sean-k-mooney you can however subdevide the ephemeral storage into multiple disk on the command line when you create the vm
11:49:15 EugenMayer using ephemeral storage we stopped, since the bug from 2016 that it cannot be resized :)
11:49:25 sean-k-mooney EugenMayer: no, as an end use you are not allowed to know the hypervior in use
11:49:29 EugenMayer nice, openstack server create is not the tool to go with here i guess
11:49:32 sean-k-mooney so you can just assume its virsh in the end
11:50:22 EugenMayer so using 'nova' is still hyperevisor agnostic, so it would be somewhat right? using virsh is discourage - that's what you mean, right?
11:51:40 sean-k-mooney openstack provides an abstration api over multiple hyperviors like libvirt/kvm, hyperv, vmware so yes the api is hypervior agnostic mostly
11:51:48 sean-k-mooney EugenMayer: using virsh is entirely unsupported
11:52:02 sean-k-mooney with opentask you can use virsh to inspect the xml for debugging
11:52:17 EugenMayer right now, i'am planing on how to migrate my old VMs which might have 1-3 disks. I know how to do it with one-disk setups. openstack image create --import .. the launch from snapshot. With multiple disks it all seems more complicated
11:52:22 sean-k-mooney but beyond that you shoudl never use virsh to modify any vm created by nova
11:53:00 EugenMayer i see, understood
11:53:03 sean-k-mooney with multi disk unfortunetly the only option that is supported via the api would be cinder
11:53:23 EugenMayer which is not an option (i understand, those would be volumes)
11:53:25 sean-k-mooney the unsupproted way woudl be to create a mulit disk vm. stop it and then copy the data
11:53:50 EugenMayer copy the data you mean, fs to fs?
11:53:56 EugenMayer or as an image
11:54:10 sean-k-mooney yes or via a rescue image but avoiding glance
11:54:55 sean-k-mooney so boot a 3 disk vm then put it in rescue mode using a clonzilla image and have clonzilla reimage the disk using the orignal vm as the image source
11:54:58 sean-k-mooney something like that
11:55:06 EugenMayer wow, that sounds quiet complicated. But understood. I somehow create and instance with x disks, then using grml and somehow mount new disks and old disks and sync the filesystem
11:56:05 sean-k-mooney there are some generic tools that might help virt2virt i think is one? i have never used them myself
11:56:21 sean-k-mooney https://libguestfs.org/virt-v2v.1.html
11:58:05 EugenMayer interesting, thanks
11:59:24 EugenMayer sean-k-mooney i assume that the disk if find on my compute is, eventhough named 'disk' nothing else then jsut a qcow disk
12:00:24 EugenMayer so i really consider going full force and trying to neither do nova-api nor virsh api, but replacing the disks. I understand this is not supported, but in the end, i do not run y types of hypevervisors and i need to migrate
12:00:42 sean-k-mooney yes you jsut need the qcow
12:01:05 sean-k-mooney ya an simple scp and overridign the disk might work
12:01:19 sean-k-mooney provided the disk dont have backing files
12:01:25 EugenMayer i will need to either adjust the size via the XML or via the API
12:01:32 sean-k-mooney you might need to first convert the qcow to a flatend one
12:01:50 EugenMayer yeah aware on how to do this, but i usually use flat
12:02:14 EugenMayer thank you a lot, again!
12:05:07 EugenMayer One last question, is there an API based way to activate nested-virtualization on a VM? I use host-passthrough already, but not sure that is enough. I need this to run an ESXi which i soley run for image disk conversion
12:05:41 sean-k-mooney no nested virt will be avaiable to the vm automatically if the host is configured to allow it
12:06:23 sean-k-mooney on older kernel you use to have to set teh nested virt flag in the intel-kvm or amd-kvm kernel module options
12:06:39 sean-k-mooney in newer kernels it default to enabled so that is no longer required
12:07:10 sean-k-mooney EugenMayer: you can check by installing linbvirt in the guest and runnign virt-host-validate
12:07:32 sean-k-mooney or just look for vmx or svm on the amd side in lscpu in the guest i think
12:08:54 sean-k-mooney so ya if your host is set up correctly then it shoudl just work
12:10:06 sean-k-mooney with that said my expeirnce with nested virt is mainly kvm on kvm and limited expericne with windows(docker/linux subsystem) usecases
12:10:27 sean-k-mooney both of those can be made work in openstack but have never tired esxi
12:17:39 kashyap EugenMayer: sean-k-mooney: Nested is enabled only in the newer _upstream_ kernels, though.
12:17:50 kashyap Some distributions might not enable it by default
12:18:05 sean-k-mooney well its enabled on rhel for intel by default now but not for amd
12:18:19 kashyap EugenMayer: So check before you run. A handy tool is `virt-host-validate` (it'll check for a whole bunch of things, including /dev/kvm)
12:18:23 kashyap You can run it on a VM too
12:18:54 sean-k-mooney yep also said that above its pretty handy but not well advertised in my experince
12:19:00 kashyap sean-k-mooney: No, not even for RHEL for Intel
12:19:15 kashyap sean-k-mooney: RHEL made it tech preview for AMD *and* Intel.
12:19:50 kashyap There it is, public docs:
12:19:51 kashyap https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/8/html/configuring_and_managing_virtualization/creating-nested-virtual-machines_configuring-and-managing-virtualization
12:20:33 kashyap sean-k-mooney: Ah, I missed your virt-host-validate reference earlier; maybe we should include it in Nova docs somewhere too
12:20:51 sean-k-mooney well we retoactivly made it tech preview it was GA'd then we found issue on the intel support
12:21:36 EugenMayer thank you both - sorry was at dinner
12:21:59 kashyap Yes, I'm saying about the current status. "Retroactie" == well, if you spot blocker issues, then you must put it back to tech-preview
12:22:18 kashyap EugenMayer: No prob. In general, don't feel pressure to respond "instantly" on IRC
12:22:29 kashyap It's called "instant messaging"; not "instant responding" :D
12:23:36 EugenMayer if i get helped i tend to keep myself as 'instant responding' - but not anybody else. If find this somewhat respecting the time of the answer-giver (what ever that word is:) )
12:24:09 kashyap Yes, sure. That's contextual
12:24:13 EugenMayer i currently run debian:bullseye with the stable kernel 5.10, will check what status it has and maybe upgrade to backports or spin up an compute on a different OS
12:24:42 kashyap EugenMayer: Some 7 years ago I wrote this thing, because we used to a see a lot of "naked pings" - https://www.rdoproject.org/contribute/irc-etiquette/

Earlier   Later