Best regards

Eli

天涯无处不重逢

a leaf duckweed belongs to the sea, where not to meet in life

On Mon, Mar 06, 2017 at 05:50:43PM +0800, Eli Qiao wrote:

Add new virsh command line `nodecachestats` to expose the cache usage
on a node.

Signed-off-by: Eli Qiao <liyong.qiao@intel.com>
---
src/libvirt_private.syms | 3 ++-
src/qemu/qemu_driver.c | 12 ++++++++++
src/util/virresctrl.c | 62 ++++++++++++++++++++++++++++++++++++++++++++++++
src/util/virresctrl.h | 8 +++++++
tools/virsh-host.c | 49 ++++++++++++++++++++++++++++++++++++++
5 files changed, 133 insertions(+), 1 deletion(-)

It perhaps would be good to also report the largest contiguous region

available: So that if the space is fragmented, management software

can detect the situation beforehand.

Like this perhaps:

# LD_LIBRARY_PATH=/root/git/libvirt/src/.libs/

# /root/git/virshnodegetcachestats

ret=0 nparams=10

L3.0: free=11520, max_contiguous=X.

L3.1: free=11520, max_contiguous=X.

Marclo, thanks for your testing and comments.

Actually, I only expose the contiguous cache which can be allocated to VMs for now.

free == max_contiguous.

Othere will be one case:

schemata 1 = 1000 0000

schemata 2 = 0010 0000

default schemata should be 0001 1111 (schemata should be contiguous, so it can’t be 01011111

and free_size = 5 * min_cache_unit

virsh nodecachestats will report free=5 * min_cache_unit

next if a new vm requires cache of 1 * min_cache_unit

schemata 1 = 1000 0000

schemata 2 = 0010 0000

new VM’s schemata = 0100 0000

default schemata should be 0001 1111

so free_size is still 5 * min_cache_unit.

virsh nodecachestats will report free=5 * min_cache_unit

I am not sure if that will meaningful to user we expose (fragmented) cache which can not be allocated to VMs.

Or:

L3.0.free: 11520

L3.0.max_contiguous=X.

L3.1.free: 11520

L3.1.max_contiguous=X.

Not sure what is the preferred way to do this in libvirt.

Otherwise, testing looks good now.