[ops][nova] How to make sure that on a hypervisor there is room for a VM with a specific flavor
Dear all I have the following use case: I defined a flavor F for VMS with 10 VCPUs, 20 GB of RAM, 30 GB of root disk, 1 GPU XYZ. I use PCI passthrough to manage GPUs. I have only a XYZ GPU on my cluster (on a H hypervisor), so a VM using this flavor F must be "placed" to the compute node H (and this is not a problem: I know how to do that). H is very large: it has a lot of RAM and cores, so it can host other VMs that do not use GPUs. How can I make it so that VMs without GPU can go on H, but there must be room for a VM with flavor F (if it doesn't already exist)? Thanks, and happy new year ! Cheers, Massimo
Hi dear If you wanna handle schedule of the instance you can use aggregate For example you can create two aggregate for gpu and general then add computes to aggregate then create flavor for aggregates and add policy in property after that when create instance with flavor of gpu the instace schdule just to gpu computes https://docs.openstack.org/python-openstackclient/pike/cli/command-objects/a... On Tue, Jan 7, 2025, 4:34 PM Massimo Sgaravatto < massimo.sgaravatto@gmail.com> wrote:
Dear all
I have the following use case:
I defined a flavor F for VMS with 10 VCPUs, 20 GB of RAM, 30 GB of root disk, 1 GPU XYZ.
I use PCI passthrough to manage GPUs.
I have only a XYZ GPU on my cluster (on a H hypervisor), so a VM using this flavor F must be "placed" to the compute node H (and this is not a problem: I know how to do that).
H is very large: it has a lot of RAM and cores, so it can host other VMs that do not use GPUs.
How can I make it so that VMs without GPU can go on H, but there must be room for a VM with flavor F (if it doesn't already exist)?
Thanks, and happy new year !
Cheers, Massimo
I am already using aggregates to manage the scheduling, but I can't see how this can address my issue (I want that there is always room for the instantiation of a VM with flavor F) Cheers, Massimo On Tue, Jan 7, 2025 at 2:14 PM Ali Basafa <alibasafa18@gmail.com> wrote:
Hi dear If you wanna handle schedule of the instance you can use aggregate For example you can create two aggregate for gpu and general then add computes to aggregate then create flavor for aggregates and add policy in property after that when create instance with flavor of gpu the instace schdule just to gpu computes
https://docs.openstack.org/python-openstackclient/pike/cli/command-objects/a...
On Tue, Jan 7, 2025, 4:34 PM Massimo Sgaravatto < massimo.sgaravatto@gmail.com> wrote:
Dear all
I have the following use case:
I defined a flavor F for VMS with 10 VCPUs, 20 GB of RAM, 30 GB of root disk, 1 GPU XYZ.
I use PCI passthrough to manage GPUs.
I have only a XYZ GPU on my cluster (on a H hypervisor), so a VM using this flavor F must be "placed" to the compute node H (and this is not a problem: I know how to do that).
H is very large: it has a lot of RAM and cores, so it can host other VMs that do not use GPUs.
How can I make it so that VMs without GPU can go on H, but there must be room for a VM with flavor F (if it doesn't already exist)?
Thanks, and happy new year !
Cheers, Massimo
Hi, Maybe you could do this using the Blazar service[1]. That way, you can create a lease to reserve your desired resources and ensure that the VM will be scheduled. [1]: https://docs.openstack.org/blazar/latest/ Em ter., 7 de jan. de 2025 às 10:22, Massimo Sgaravatto < massimo.sgaravatto@gmail.com> escreveu:
I am already using aggregates to manage the scheduling, but I can't see how this can address my issue (I want that there is always room for the instantiation of a VM with flavor F)
Cheers, Massimo
On Tue, Jan 7, 2025 at 2:14 PM Ali Basafa <alibasafa18@gmail.com> wrote:
Hi dear If you wanna handle schedule of the instance you can use aggregate For example you can create two aggregate for gpu and general then add computes to aggregate then create flavor for aggregates and add policy in property after that when create instance with flavor of gpu the instace schdule just to gpu computes
https://docs.openstack.org/python-openstackclient/pike/cli/command-objects/a...
On Tue, Jan 7, 2025, 4:34 PM Massimo Sgaravatto < massimo.sgaravatto@gmail.com> wrote:
Dear all
I have the following use case:
I defined a flavor F for VMS with 10 VCPUs, 20 GB of RAM, 30 GB of root disk, 1 GPU XYZ.
I use PCI passthrough to manage GPUs.
I have only a XYZ GPU on my cluster (on a H hypervisor), so a VM using this flavor F must be "placed" to the compute node H (and this is not a problem: I know how to do that).
H is very large: it has a lot of RAM and cores, so it can host other VMs that do not use GPUs.
How can I make it so that VMs without GPU can go on H, but there must be room for a VM with flavor F (if it doesn't already exist)?
Thanks, and happy new year !
Cheers, Massimo
participants (3)
-
Ali Basafa
-
Massimo Sgaravatto
-
Winicius Allan