nova/nova/virt/fake.py

282 lines
10 KiB
Python

# vim: tabstop=4 shiftwidth=4 softtabstop=4
# Copyright 2010 United States Government as represented by the
# Administrator of the National Aeronautics and Space Administration.
# All Rights Reserved.
# Copyright (c) 2010 Citrix Systems, Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License"); you may
# not use this file except in compliance with the License. You may obtain
# a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
# License for the specific language governing permissions and limitations
# under the License.
"""
A fake (in-memory) hypervisor+api.
Allows nova testing w/o a hypervisor. This module also documents the
semantics of real hypervisor connections.
"""
from nova import exception
from nova.compute import power_state
def get_connection(_):
# The read_only parameter is ignored.
return FakeConnection.instance()
class FakeConnection(object):
"""
The interface to this class talks in terms of 'instances' (Amazon EC2 and
internal Nova terminology), by which we mean 'running virtual machine'
(XenAPI terminology) or domain (Xen or libvirt terminology).
An instance has an ID, which is the identifier chosen by Nova to represent
the instance further up the stack. This is unfortunately also called a
'name' elsewhere. As far as this layer is concerned, 'instance ID' and
'instance name' are synonyms.
Note that the instance ID or name is not human-readable or
customer-controlled -- it's an internal ID chosen by Nova. At the
nova.virt layer, instances do not have human-readable names at all -- such
things are only known higher up the stack.
Most virtualization platforms will also have their own identity schemes,
to uniquely identify a VM or domain. These IDs must stay internal to the
platform-specific layer, and never escape the connection interface. The
platform-specific layer is responsible for keeping track of which instance
ID maps to which platform-specific ID, and vice versa.
In contrast, the list_disks and list_interfaces calls may return
platform-specific IDs. These identify a specific virtual disk or specific
virtual network interface, and these IDs are opaque to the rest of Nova.
Some methods here take an instance of nova.compute.service.Instance. This
is the datastructure used by nova.compute to store details regarding an
instance, and pass them into this layer. This layer is responsible for
translating that generic datastructure into terms that are specific to the
virtualization platform.
"""
def __init__(self):
self.instances = {}
@classmethod
def instance(cls):
if not hasattr(cls, '_instance'):
cls._instance = cls()
return cls._instance
def init_host(self):
"""
Initialize anything that is necessary for the driver to function
"""
return
def list_instances(self):
"""
Return the names of all the instances known to the virtualization
layer, as a list.
"""
return self.instances.keys()
def spawn(self, instance):
"""
Create a new instance/VM/domain on the virtualization platform.
The given parameter is an instance of nova.compute.service.Instance.
This function should use the data there to guide the creation of
the new instance.
The work will be done asynchronously. This function returns a
Deferred that allows the caller to detect when it is complete.
Once this successfully completes, the instance should be
running (power_state.RUNNING).
If this fails, any partial instance should be completely
cleaned up, and the virtualization platform should be in the state
that it was before this call began.
"""
fake_instance = FakeInstance()
self.instances[instance.name] = fake_instance
fake_instance._state = power_state.RUNNING
def reboot(self, instance):
"""
Reboot the specified instance.
The given parameter is an instance of nova.compute.service.Instance,
and so the instance is being specified as instance.name.
The work will be done asynchronously. This function returns a
Deferred that allows the caller to detect when it is complete.
"""
pass
def rescue(self, instance):
"""
Rescue the specified instance.
"""
pass
def unrescue(self, instance):
"""
Unrescue the specified instance.
"""
pass
def pause(self, instance, callback):
"""
Pause the specified instance.
"""
pass
def unpause(self, instance, callback):
"""
Unpause the specified instance.
"""
pass
def suspend(self, instance, callback):
"""
suspend the specified instance
"""
pass
def resume(self, instance, callback):
"""
resume the specified instance
"""
pass
def destroy(self, instance):
"""
Destroy (shutdown and delete) the specified instance.
The given parameter is an instance of nova.compute.service.Instance,
and so the instance is being specified as instance.name.
The work will be done asynchronously. This function returns a
Deferred that allows the caller to detect when it is complete.
"""
del self.instances[instance.name]
def attach_volume(self, instance_name, device_path, mountpoint):
"""Attach the disk at device_path to the instance at mountpoint"""
return True
def detach_volume(self, instance_name, mountpoint):
"""Detach the disk attached to the instance at mountpoint"""
return True
def get_info(self, instance_name):
"""
Get a block of information about the given instance. This is returned
as a dictionary containing 'state': The power_state of the instance,
'max_mem': The maximum memory for the instance, in KiB, 'mem': The
current memory the instance has, in KiB, 'num_cpu': The current number
of virtual CPUs the instance has, 'cpu_time': The total CPU time used
by the instance, in nanoseconds.
This method should raise exception.NotFound if the hypervisor has no
knowledge of the instance
"""
if instance_name not in self.instances:
raise exception.NotFound(_("Instance %s Not Found")
% instance_name)
i = self.instances[instance_name]
return {'state': i._state,
'max_mem': 0,
'mem': 0,
'num_cpu': 2,
'cpu_time': 0}
def list_disks(self, instance_name):
"""
Return the IDs of all the virtual disks attached to the specified
instance, as a list. These IDs are opaque to the caller (they are
only useful for giving back to this layer as a parameter to
disk_stats). These IDs only need to be unique for a given instance.
Note that this function takes an instance ID, not a
compute.service.Instance, so that it can be called by compute.monitor.
"""
return ['A_DISK']
def list_interfaces(self, instance_name):
"""
Return the IDs of all the virtual network interfaces attached to the
specified instance, as a list. These IDs are opaque to the caller
(they are only useful for giving back to this layer as a parameter to
interface_stats). These IDs only need to be unique for a given
instance.
Note that this function takes an instance ID, not a
compute.service.Instance, so that it can be called by compute.monitor.
"""
return ['A_VIF']
def block_stats(self, instance_name, disk_id):
"""
Return performance counters associated with the given disk_id on the
given instance_name. These are returned as [rd_req, rd_bytes, wr_req,
wr_bytes, errs], where rd indicates read, wr indicates write, req is
the total number of I/O requests made, bytes is the total number of
bytes transferred, and errs is the number of requests held up due to a
full pipeline.
All counters are long integers.
This method is optional. On some platforms (e.g. XenAPI) performance
statistics can be retrieved directly in aggregate form, without Nova
having to do the aggregation. On those platforms, this method is
unused.
Note that this function takes an instance ID, not a
compute.service.Instance, so that it can be called by compute.monitor.
"""
return [0L, 0L, 0L, 0L, null]
def interface_stats(self, instance_name, iface_id):
"""
Return performance counters associated with the given iface_id on the
given instance_id. These are returned as [rx_bytes, rx_packets,
rx_errs, rx_drop, tx_bytes, tx_packets, tx_errs, tx_drop], where rx
indicates receive, tx indicates transmit, bytes and packets indicate
the total number of bytes or packets transferred, and errs and dropped
is the total number of packets failed / dropped.
All counters are long integers.
This method is optional. On some platforms (e.g. XenAPI) performance
statistics can be retrieved directly in aggregate form, without Nova
having to do the aggregation. On those platforms, this method is
unused.
Note that this function takes an instance ID, not a
compute.service.Instance, so that it can be called by compute.monitor.
"""
return [0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L]
def get_console_output(self, instance):
return 'FAKE CONSOLE OUTPUT'
def get_ajax_console(self, instance):
return 'http://fakeajaxconsole.com/?token=FAKETOKEN'
class FakeInstance(object):
def __init__(self):
self._state = power_state.NOSTATE