nova创建instance流程

###################################################################

# Start: Jeffrey 的话
###############################################################################################

具体请看官方的介绍：http://docs.openstack.org/developer/nova/conductor.html

Conductor as a place for orchestrating tasks

1. nova-conductor作用：

nova-conductor除了用作一个是DB proxy之外，它还有一个作用就是可以对任务进行编排。比如：重新创建一个Instance, insance 迁移，Instance创建等工作都是由于来负责“分配工作”（实际它并不负责执行相关的动作，它只是进行任务的编排）。

2. 好处：

可以把nova-scheduler的工作与nova-compute的工作完全分开，各司其职。

3. 为什么选择nova-conductor做这件事情

可以根据response同步请求 nova-scheduler.

###################################################################

# End Jeffrey 的话
################################################################### http://www.aboutyun.com/thread-17436-1-1.html

综述

启动一个新的instance涉及到很多openstack nova里面的组件

API server：处理客户端的请求，并且转发到cloud control
Cloud control：处理compute节点，网络控制节点，API server和scheduler中间连接
Scheduler：选择一个host去执行命令
compute worker：启动和停止实例，附加和删除卷 等操作
network controller：管理网络资源，分配固定IP，配置vlans



1.API server将消息发送到Cloud Controller
2. Authertication 保用户有权限，然后Cloud Controller将消息发送给Scheduler
3. Scheduler caste 一个消息给一个选择好的host要求他启动一个实例
4.compute worker（选择的那个host）获取到消息
5.6.7.8 compute worker需要一个固定的ip去启动一个实例，所以向network controller发送消息

下面我将详细说明一下：

API

1.可以在dashboard网页上面进行
2.可以用命令行 euca-add-keypair euca-run-instances

用户的请求发送到nova-api，有两种方式
第一种：通过openstack api (nova/api/servers.py 类 class Controller(object)）create方法

def create(self, req, body):
""" Creates a new server for a given user """
if 'server' in body:
body['server']['key_name'] = self._get_key_name(req, body)

extra_values = None
extra_values, instances = self.helper.create_instance(
req, body, self.compute_api.create)

第二种：通过ec2 api (nova/api/cloud.py 中类 CloudController )
调用def run_instances(self, context, **kwargs):
...
(instances, resv_id) = self.compute_api.create(context,
instance_type=instance_types.get_instance_type_by_name(
kwargs.get('instance_type', None)),
...

最终调用的Compute API create()：

查看这种类型的instance是否达到最大值
如果不存在安全组，就创建个
生成MAC地址和hostnames
给scheduler发送一个消息去运行这个实例

CAST

当然maxCount为1（默认值为1）的时候 调用RPC.cast方法向scheduler发送运行实例的消息



在openstack中通过RPC.cast来发送消息，消息的分发通过RabbitMQ。消息发送方(Compute API)往
topic exchange(scheduler topic)发送一个消息，消息消费者(Scheduler worker)从队列中获得消息，
cast调用不需要返回值。

[python] view
plain copy

def _schedule_run_instance(self,

...

return rpc_method(context,

FLAGS.scheduler_topic,

{"method": "run_instance",

"args": {"topic": FLAGS.compute_topic,

"request_spec": request_spec,

"admin_password": admin_password,

"injected_files": injected_files,

"requested_networks": requested_networks,

"is_first_time": True,

"filter_properties": filter_properties}})

Scheduler

scheduler接收到消息，然后通过设定的scheduler策略选择一个目的host，如：zone scheduler
选择一个主机在特定的可获取的zone上面。最后发送一个cast消息到特定的host上面

[python] view
plain copy

def cast_to_compute_host(context, host, method, update_db=True, **kwargs):

"""Cast request to a compute host queue"""

if update_db:

# fall back on the id if the uuid is not present

instance_id = kwargs.get('instance_id', None)

instance_uuid = kwargs.get('instance_uuid', instance_id)

if instance_uuid is not None:

now = utils.utcnow()

db.instance_update(context, instance_uuid,

{'host': host, 'scheduled_at': now})

rpc.cast(context,

db.queue_get_for(context, 'compute', host),

{"method": method, "args": kwargs})

LOG.debug(_("Casted '%(method)s' to compute '%(host)s'") % locals())

Compute

compute worker进程接收到消息执行方法(nova/compute/manager.py)

[python] view
plain copy

def _run_instance(self, context, instance_uuid,

requested_networks=None,

injected_files=[],

admin_password=None,

is_first_time=False,

**kwargs):

"""Launch a new instance with specified options."""

context = context.elevated()

try:

instance = self.db.instance_get_by_uuid(context, instance_uuid)

self._check_instance_not_already_created(context, instance)

image_meta = self._check_image_size(context, instance)

self._start_building(context, instance)

self._notify_about_instance_usage(instance, "create.start")

network_info = self._allocate_network(context, instance,

requested_networks)

try:

block_device_info = self._prep_block_device(context, instance)

instance = self._spawn(context, instance, image_meta,

network_info, block_device_info,

injected_files, admin_password)

...

检查instance是否已经在运行
分配一个固定的ip地址
如果没有设置vlan和网桥，设置一下
最后通过虚拟化的driver spawn一个instance

network controller

network_info = self._allocate_network(context, instance,
requested_networks)

调用network的API的allocate_for_instance方法

[python] view
plain copy

def allocate_for_instance(self, context, instance, **kwargs):

"""Allocates all network structures for an instance.

:returns: network info as from get_instance_nw_info() below

"""

args = kwargs

args['instance_id'] = instance['id']

args['instance_uuid'] = instance['uuid']

args['project_id'] = instance['project_id']

args['host'] = instance['host']

args['rxtx_factor'] = instance['instance_type']['rxtx_factor']

nw_info = rpc.call(context, FLAGS.network_topic,

{'method': 'allocate_for_instance',

'args': args})

RPC.call 与RPC.cast最大的不同 就是call方法需要一个response

Spawn instance

接下来我要说的就是虚拟化的driver spawn instance，我们这里使用的是libvirt（nova/virt/libvirt/lconnection.py）

[python] view
plain copy

def spawn(self, context, instance, image_meta, network_info,

block_device_info=None):

xml = self.to_xml(instance, network_info, image_meta, False,

block_device_info=block_device_info)

self.firewall_driver.setup_basic_filtering(instance, network_info)

self.firewall_driver.prepare_instance_filter(instance, network_info)

self._create_image(context, instance, xml, network_info=network_info,

block_device_info=block_device_info)

self._create_new_domain(xml)

LOG.debug(_("Instance is running"), instance=instance)

self._enable_hairpin(instance)

self.firewall_driver.apply_instance_filter(instance, network_info)

def _wait_for_boot():

"""Called at an interval until the VM is running."""

try:

state = self.get_info(instance)['state']

except exception.NotFound:

LOG.error(_("During reboot, instance disappeared."),

instance=instance)

raise utils.LoopingCallDone

if state == power_state.RUNNING:

LOG.info(_("Instance spawned successfully."),

instance=instance)

raise utils.LoopingCallDone

timer = utils.LoopingCall(_wait_for_boot)

return timer.start(interval=0.5, now=True)

通过libvirt xml文件，然后根据xml文件生成instance
准备network filter，默认的fierwall driver是iptables
image的创建(详细情况以后再介绍)

def _create_image(self, context, instance, libvirt_xml, suffix='',

disk_images=None, network_info=None,
block_device_info=None):

...

最后虚拟化driver的spawn()方法中调用driver 的creatXML()