Comment 2 for bug 1422775

Revision history for this message
Chris Friesen (cbf123) wrote : Re: nova libvirt driver assumes qemu support for NUMA pinning

A few things:

1) Currently the "pages" subelement of the "hugepages" element is set even when no NUMA support is requested for the guest. This is a bug. We should be able to use hugepages even if we can't use NUMA.

2) As long as the CPU affinity is set before the memory allocation (should probably check whether this is the case) then the NUMA allocation would be governed by the system-wide allocation rule. So maybe this means that with older qemu we don't allow setting per-flavor hw_numa_mempolicy, you're limited to the system-wide rule.

Even if we decide to totally disallow NUMA support unless qemu supports it "properly", we should still add in a version (or better yet a runtime capabilities) check so that we can fail early with a nice error message.