This code provides an implementation of the get_diagnostics
API for the libvirt driver. This API is invoked by the
'nova diagnostic' command.
There is no existing formal specification for what data items
to expose from the 'get_diagnostics' API, so the XenAPI driver
is used as an informal guide. The Xen implementation currently
produces the following output:
+----------------+-----------------+
| Property | Value |
+----------------+-----------------+
| cpu0 | 4.3627 |
| memory | 1171088064.0000 |
| memory_target | 1171088064.0000 |
| vbd_xvda_read | 0.0 |
| vbd_xvda_write | 0.0 |
| vif_0_rx | 3223.6870 |
| vif_0_tx | 0.0 |
| vif_1_rx | 104.4955 |
| vif_1_tx | 0.0 |
+----------------+-----------------+
The new libvirt implementation will return a similar set of stats
for guest CPUs, memory, disks and network interfaces, resulting
in the following output:
+------------------+------------+
| Property | Value |
+------------------+------------+
| cpu0_time | 2870000000 |
| memory | 524288 |
| vda_errors | -1 |
| vda_read | 262144 |
| vda_read_req | 112 |
| vda_write | 5606400 |
| vda_write_req | 376 |
| vnet0_rx | 63343 |
| vnet0_rx_drop | 0 |
| vnet0_rx_errors | 0 |
| vnet0_rx_packets | 431 |
| vnet0_tx | 4905 |
| vnet0_tx_drop | 0 |
| vnet0_tx_errors | 0 |
| vnet0_tx_packets | 45 |
+------------------+------------+
The implementation has only been tested with the libvirt KVM
driver.
In a future a formal specification for the required data
format should be written and all virt drivers made to comply.
Finally, the fakelibvirt driver has been updated in order to
support the methods required by the get_diagnostics method.
For those interested, below is the formula to extract the
CPU usage based on the CPU time. I've added it here for reference
since it's not used at the moment and it took quite a while
to locate it in the libvirt documentation.
usage = 100 * (current_cpu_time - last_cpu_time) /
(measurement_interval * (10 ** 9))
Fixes bug 986200.
Change-Id: I92da392f43d0840ed31cfd88f54570ea5830bb4a
29030ec179