第9章

This option does not affect `PCPU` inventory, which cannot be
overcommitted.

If this option is set to something *other than* `None` or `0.0`, the
allocation ratio will be overwritten by the value of this option, otherwise,
the allocation ratio will not change. Once set to a non-default value, it is
not possible to "unset" the config to get back to the default behavior. If
you want to reset back to the initial value, explicitly specify it to the
value of `initial_cpu_allocation_ratio`.

If the value is set to `>1`, we recommend keeping track of the free disk
space, as the value approaching `0` may result in the incorrect
functioning of instances using it at the moment.

If this option is set to something *other than* `None` or `0.0`, the
allocation ratio will be overwritten by the value of this option, otherwise,
the allocation ratio will not change. Once set to a non-default value, it is
not possible to "unset" the config to get back to the default behavior. If
you want to reset back to the initial value, explicitly specify it to the
value of `initial_disk_allocation_ratio`.

The following image properties are *never* inherited regardless of
whether they are listed in this configuration option or not:

If this option is set to something *other than* `None` or `0.0`, the
allocation ratio will be overwritten by the value of this option, otherwise,
the allocation ratio will not change. Once set to a non-default value, it is
not possible to "unset" the config to get back to the default behavior. If
you want to reset back to the initial value, explicitly specify it to the
value of `initial_ram_allocation_ratio`.

The compute service cannot reliably determine which types of virtual
interfaces (`port.binding:vif_type`) will send `network-vif-plugged`
events without an accompanying port `binding:host_id` change.
Open vSwitch and linuxbridge should be OK, but OpenDaylight is at least
one known backend that will not currently work in this case, see bug
https://launchpad.net/bugs/1755890 for more details.

In a multi-cell (v2) setup where the cell MQ is separated from the
top-level, computes cannot directly communicate with the scheduler. Thus,
this option cannot be enabled in that scenario. See also the
`[workarounds] disable_group_policy_check_upcall` option.

It is recommended to read :ref:`the deployment documentation's
section on this option <num_memory_encrypted_guests>` before
deciding whether to configure this setting or leave it at the
default.