On 18-07-12 10:40, Sebastian Wiedenroth wrote:
Hi, I'm currently working on a patch to support Sheepog [1] as primary storage. The changes are similar to those proposed by the recently posted RBD patch.
Nice! Shouldn't we team this up? In my RBD patch I already layed a bit of groundwork for Sheepdog, so we should prevent our patches conflicting.
Would be nice if Apache CloudStack 4.0 could have both RBD and Sheepdog! :)
However, instead of ignoring the pool in FirstFitStoragePoolAllocator for system vms, my patch currently creates a temporary local pool (on /tmp) to patch the system vm and then moves the image onto sheepdog. That works, but I'm not too happy with the solution as it requires the virtualization node to have a local disk with enough free space. Is there a better way to do this? Should I just disable Sheepdog for system vms and let them run on a "regular" primary storage?
I don't see a better way at the moment and I've chosen the easy way out by excluding RBD for deployment of System VM's, you probably saw that in the FirstFitAllocator.
As long as the System VM's are being build with all kinds of scripts I really don't see a clean way to do this.
Wido
Best Regards, Sebastian [1] http://www.osrg.net/sheepdog/
