Multiple proxmox clusters sharing same CEPH pool

[pieterk]

Hi,

Is it possible to have administrators set the VMID range for a proxmox cluster so that clusters can be made to have their own unique ranges?

For example:
Cluster 1 - VMID 1000 to VMID 1999
Cluster 2 - VMID 2000 to VMID 2999

Reason being that we have a CEPH cluster (separate) and a storage pool that possibly multiple clusters would interact with but if two clusters have the same VMID then they point to the same RBD device which is not good.

If not is this a feature that proxmox could maybe implement in the GUI in a future patch?

Thanks

 

[spirit]

Hi,

Is it possible to have administrators set the VMID range for a proxmox cluster so that clusters can be made to have their own unique ranges?

For example:
Cluster 1 - VMID 1000 to VMID 1999
Cluster 2 - VMID 2000 to VMID 2999

Reason being that we have a CEPH cluster (separate) and a storage pool that possibly multiple clusters would interact with but if two clusters have the same VMID then they point to the same RBD device which is not good.

If not is this a feature that proxmox could maybe implement in the GUI in a future patch?

Thanks

can't you create 2 differents pools on ceph storage ?

 

[pieterk]

I think the concern was about placement group balancing issue that might crop up if pools were to be created willynilly if proxmox clusters are just continuously added on. What if two more clusters are added? Then that would be 2 more CEPH pools and the PG juggling that comes with it...

 

[pieterk]

In addition spirit (I appreciate you help anyway) the CEPH mailing list actually mentions about creating many pools will have a penalty (as such). Lets say we have 5 proxmox clusters which is a reasonably large amount of servers but not too much, if a pool were created for each and each pool had a PG value of 4096 (again not to large) then that would have a total PG value of 20,000 placement groups on your OSD's. This would put extra unnecessary load on the OSD's and the system if the placement group value gets to high.

This is one of the reason we would prefer to share a CEPH pool among clusters and still have clusters not step on each others toes.

Edit: (not a massive ceph cluster)

 

[spirit]

In addition spirit (I appreciate you help anyway) the CEPH mailing list actually mentions about creating many pools will have a penalty (as such). Lets say we have 5 proxmox clusters which is a reasonably large amount of servers but not too much, if a pool were created for each and each pool had a PG value of 4096 (again not to large) then that would have a total PG value of 20,000 placement groups on your OSD's. This would put extra unnecessary load on the OSD's and the system if the placement group value gets to high.

This is one of the reason we would prefer to share a CEPH pool among clusters and still have clusters not step on each others toes.

Edit: (not a massive ceph cluster)

Ok, make sense.
Currently it's not possible to limit vm ids range.
But maybe can you fill a request on proxmox bugzilla. (But I don't think it ll be available soon)