scheduler doesn't consume space properly for create_volume

Hey Walt, I hope this message reaches you well

Would you please let me know the following information:
- Are you using master or any other version
- Do you have any logs in c-sch log or any other place
- Please let us know the step so the team can reproduce the problem.

Thanks a lot

This is running on master.

I think I have finally tracked the source of the problem here.

The allocated_capacity_gb is updated in the backend_state.allocated_capacity_gb. The problem is that the get_pools only looks in the PoolState's capabilities dictionary, which is only updated every get_volume_stats time.

cinder get-pools --detail eventually boils down to a call to the host_manager's get_pools().
This builds a readonly dictionary ONLY from the PoolState class's capabilities here:
https://github.com/openstack/cinder/blob/master/cinder/scheduler/host_manager.py#L818

This is completely out of date.
What the scheduler actually uses is the PoolState's attributes such as
https://github.com/openstack/cinder/blob/master/cinder/scheduler/host_manager.py#L127-L129

Those are updated, but the capabilities dictionary is not.

Here you can see the method
consume_from_volume, which updates the attributes for the pool (PoolState) object.
This updates the allocated_capacity_gb, free_capacity_gb.

https://github.com/openstack/cinder/blob/master/cinder/scheduler/host_manager.py#L336-L353

The pool's capabilities is NOT updated, and hence the get_pools call returns out of date values
for the pool between get_volume_stats() calls.