hardware.* metrics fetched via tripleo discovery broken with gnocchi

[gordon chung]

hardware.* metrics can be discovered via tripleo rather than statically defining the host ip. this discovery sets the resource_id to the instance.id[1]. this is not good as now we have host and instance resources with the same resource_id

therefore, when it sends data to gnocchi, it causes chaos as it will build a hybrid resource with metrics from host and instance resources[2]. this may be fine but it falls apart when updating metadata as it may have created resource as instance but it will try to update host resource-type or vice versa.

tl;dr, the overcloud host is the same as the undercloud instance but ceilometer is incapable of figuring out how to handle that.

snmp discovery should probably not use instance.id as the resource_id.

[1] https://github.com/openstack/ceilometer/blob/342feaa285e586903e6b3a148b282cfa38881ea0/ceilometer/hardware/discovery.py#L119
[2] we don't use type when pushing metrics: https://github.com/openstack/ceilometer/blob/342feaa285e586903e6b3a148b282cfa38881ea0/ceilometer/publisher/gnocchi.py#L390