nova-重启云主机实例

1. nova/api/openstack/compute/servers.py文件中

    def _action_reboot(self, req, id, body):

        reboot_type = body['reboot']['type'].upper()
        context = req.environ['nova.context']
        instance = self._get_server(context, req, id)
        context.can(server_policies.SERVERS % 'reboot',
                    target={'project_id': instance.project_id})

        try:
            #这个是主要调用的函数
            self.compute_api.reboot(context, instance, reboot_type)
        #这里是异常信息检测
        except exception.InstanceIsLocked as e:
            raise exc.HTTPConflict(explanation=e.format_message())
        except exception.InstanceInvalidState as state_error:
            common.raise_http_conflict_for_instance_invalid_state(state_error,
                    'reboot', id)

2. nova/compute/api.py文件中

    @check_instance_lock
    def reboot(self, context, instance, reboot_type):
        """Reboot the given instance."""
        if reboot_type == 'SOFT':
            self._soft_reboot(context, instance)
        else:
            self._hard_reboot(context, instance)

这里有一点小小的问题，因为在步骤1中有云主机状态的检测，但是在这里并没有状态检测的装饰器

self._soft_reboot和self._hard_reboot函数

 

3. 软重启和硬重启，前面调用的函数流程是一样的，只不过reboot_type参数是‘SOFT’和‘HARD’

    def _soft_reboot(self, context, instance):
        expected_task_state = [None]
        instance.task_state = task_states.REBOOTING
        instance.save(expected_task_state=expected_task_state)

        self._record_action_start(context, instance, instance_actions.REBOOT)

        self.compute_rpcapi.reboot_instance(context, instance=instance,
                                            block_device_info=None,
                                            reboot_type='SOFT')

首先记录状态

这里有点不太明白了

compute_utils.notify_about_instance_action和self._record_action_start(context, instance, instance_actions.REBOOT)函数之间有什么区别呢

  def _record_action_start(self, context, instance, action):
        objects.InstanceAction.action_start(context, instance.uuid,
                                            action, want_result=False)


    @base.remotable_classmethod
    def action_start(cls, context, instance_uuid, action_name,
                     want_result=True):
        values = cls.pack_action_start(context, instance_uuid, action_name)
        db_action = db.action_start(context, values)
        if want_result:
            return cls._from_db_object(context, cls(), db_action)

self._record_action_start函数的作用是向数据库中记录云主机的操作！这个函数的调用是在发送RPC消息之前调用的。

compute_utils.notify_about_instance_action函数是在发送RPC消息之后调用的！

4. 

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def reboot_instance(self, context, instance, block_device_info,
                        reboot_type):
        @utils.synchronized(instance.uuid)
        def do_reboot_instance(context, instance, block_device_info,
                               reboot_type):
            self._reboot_instance(context, instance, block_device_info,
                                  reboot_type)
        do_reboot_instance(context, instance, block_device_info, reboot_type)

这里code写的有点绕

主要调用的代码是_reboot_instance

    def _reboot_instance(self, context, instance, block_device_info,
                         reboot_type):
        """Reboot an instance on this host."""
        # acknowledge the request made it to the manager
        #分为软重启和硬重启
        if reboot_type == "SOFT":
            instance.task_state = task_states.REBOOT_PENDING
            expected_states = task_states.soft_reboot_states
        else:
            instance.task_state = task_states.REBOOT_PENDING_HARD
            expected_states = task_states.hard_reboot_states

        context = context.elevated()
        LOG.info("Rebooting instance", instance=instance)

        #bdms信息是从数据库中根据instance的uuid查找得到的块设备挂载信息表
        bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
            context, instance.uuid)
        #然后根据这个bdms信息得到block_device_info表
        block_device_info = self._get_instance_block_device_info(
            context, instance, bdms=bdms)

        network_info = self.network_api.get_instance_nw_info(context, instance)

        accel_info = self._get_accel_info(context, instance)

        self._notify_about_instance_usage(context, instance, "reboot.start")
        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.REBOOT,
            phase=fields.NotificationPhase.START,
            bdms=bdms
        )

        instance.power_state = self._get_power_state(context, instance)
        instance.save(expected_task_state=expected_states)

        if instance.power_state != power_state.RUNNING:
            state = instance.power_state
            running = power_state.RUNNING
            LOG.warning('trying to reboot a non-running instance:'
                        ' (state: %(state)s expected: %(running)s)',
                        {'state': state, 'running': running},
                        instance=instance)

        def bad_volumes_callback(bad_devices):
            self._handle_bad_volumes_detached(
                    context, instance, bad_devices, block_device_info)

        try:
            # Don't change it out of rescue mode
            if instance.vm_state == vm_states.RESCUED:
                new_vm_state = vm_states.RESCUED
            else:
                new_vm_state = vm_states.ACTIVE
            new_power_state = None
            if reboot_type == "SOFT":
                instance.task_state = task_states.REBOOT_STARTED
                expected_state = task_states.REBOOT_PENDING
            else:
                instance.task_state = task_states.REBOOT_STARTED_HARD
                expected_state = task_states.REBOOT_PENDING_HARD
            instance.save(expected_task_state=expected_state)
            self.driver.reboot(context, instance,
                               network_info,
                               reboot_type,
                               block_device_info=block_device_info,
                               accel_info=accel_info,
                               bad_volumes_callback=bad_volumes_callback)

        except Exception as error:
            with excutils.save_and_reraise_exception() as ctxt:
                exc_info = sys.exc_info()
                # if the reboot failed but the VM is running don't
                # put it into an error state
                new_power_state = self._get_power_state(context, instance)
                if new_power_state == power_state.RUNNING:
                    LOG.warning('Reboot failed but instance is running',
                                instance=instance)
                    compute_utils.add_instance_fault_from_exc(context,
                            instance, error, exc_info)
                    self._notify_about_instance_usage(context, instance,
                            'reboot.error', fault=error)
                    compute_utils.notify_about_instance_action(
                        context, instance, self.host,
                        action=fields.NotificationAction.REBOOT,
                        phase=fields.NotificationPhase.ERROR,
                        exception=error, bdms=bdms
                    )
                    ctxt.reraise = False
                else:
                    LOG.error('Cannot reboot instance: %s', error,
                              instance=instance)
                    self._set_instance_obj_error_state(context, instance)
        
        #重启成功之后，保存云主机的状态
        if not new_power_state:
            new_power_state = self._get_power_state(context, instance)
        try:
            instance.power_state = new_power_state
            instance.vm_state = new_vm_state
            instance.task_state = None
            instance.save()
        except exception.InstanceNotFound:
            LOG.warning("Instance disappeared during reboot",
                        instance=instance)

        self._notify_about_instance_usage(context, instance, "reboot.end")
        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.REBOOT,
            phase=fields.NotificationPhase.END,
            bdms=bdms
        )

4.1  _get_instance_block_device_info函数是如何调用的？

    def _get_instance_block_device_info(self, context, instance,
                                        refresh_conn_info=False,
                                        bdms=None):
        """Transform block devices to the driver block_device format."""

        if bdms is None:
            bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                    context, instance.uuid)
        block_device_info = driver.get_block_device_info(instance, bdms)

        if not refresh_conn_info:
            # if the block_device_mapping has no value in connection_info
            # (returned as None), don't include in the mapping
            block_device_info['block_device_mapping'] = [
                bdm for bdm in driver.block_device_info_get_mapping(
                                    block_device_info)
                if bdm.get('connection_info')]
        else:
            driver_block_device.refresh_conn_infos(
                driver.block_device_info_get_mapping(block_device_info),
                context, instance, self.volume_api, self.driver)

        self._block_device_info_to_legacy(block_device_info)

        return block_device_info

 

 

 

5.soft_reboot

    def _soft_reboot(self, instance):
        """Attempt to shutdown and restart the instance gracefully."""
        guest = self._host.get_guest(instance)

        state = guest.get_power_state(self._host)
        old_domid = guest.id
        # NOTE(vish): This check allows us to reboot an instance that
        #             is already shutdown.
        if state == power_state.RUNNING:
            guest.shutdown()
        # NOTE(vish): This actually could take slightly longer than the
        #             FLAG defines depending on how long the get_info
        #             call takes to return.
        self._prepare_pci_devices_for_use(
            pci_manager.get_instance_pci_devs(instance, 'all'))
        for x in range(CONF.libvirt.wait_soft_reboot_seconds):
            guest = self._host.get_guest(instance)

            state = guest.get_power_state(self._host)
            new_domid = guest.id

            # NOTE(ivoks): By checking domain IDs, we make sure we are
            #              not recreating domain that's already running.
            if old_domid != new_domid:
                if state in [power_state.SHUTDOWN,
                             power_state.CRASHED]:
                    LOG.info("Instance shutdown successfully.",
                             instance=instance)
                    self._create_domain(domain=guest._domain)
                    timer = loopingcall.FixedIntervalLoopingCall(
                        self._wait_for_running, instance)
                    timer.start(interval=0.5).wait()
                    return True
                else:
                    LOG.info("Instance may have been rebooted during soft "
                             "reboot, so return now.", instance=instance)
                    return True
            greenthread.sleep(1)
        return False

软重启的过程就是调用domain中的shutdown和start，由于domain.id只有虚拟机在运行的时候才会存在。从代码中看一下，重启之后其domain.id也应该会变化。

6. hard_reboot

    def _hard_reboot(self, context, instance, network_info,
                     block_device_info=None, accel_info=None):
        """Reboot a virtual machine, given an instance reference.

        Performs a Libvirt reset (if supported) on the domain.

        If Libvirt reset is unavailable this method actually destroys and
        re-creates the domain to ensure the reboot happens, as the guest
        OS cannot ignore this action.
        """
        # NOTE(sbauza): Since we undefine the guest XML when destroying, we
        # need to remember the existing mdevs for reusing them.
        mdevs = self._get_all_assigned_mediated_devices(instance)
        mdevs = list(mdevs.keys())
        # NOTE(mdbooth): In addition to performing a hard reboot of the domain,
        # the hard reboot operation is relied upon by operators to be an
        # automated attempt to fix as many things as possible about a
        # non-functioning instance before resorting to manual intervention.
        # With this goal in mind, we tear down all the aspects of an instance
        # we can here without losing data. This allows us to re-initialise from
        # scratch, and hopefully fix, most aspects of a non-functioning guest.
        self.destroy(context, instance, network_info, destroy_disks=False,
                     block_device_info=block_device_info)

        # Convert the system metadata to image metadata
        # NOTE(mdbooth): This is a workaround for stateless Nova compute
        #                https://bugs.launchpad.net/nova/+bug/1349978
        instance_dir = libvirt_utils.get_instance_path(instance)
        fileutils.ensure_tree(instance_dir)

        disk_info = blockinfo.get_disk_info(CONF.libvirt.virt_type,
                                            instance,
                                            instance.image_meta,
                                            block_device_info)
        # NOTE(vish): This could generate the wrong device_format if we are
        #             using the raw backend and the images don't exist yet.
        #             The create_images_and_backing below doesn't properly
        #             regenerate raw backend images, however, so when it
        #             does we need to (re)generate the xml after the images
        #             are in place.
        xml = self._get_guest_xml(context, instance, network_info, disk_info,
                                  instance.image_meta,
                                  block_device_info=block_device_info,
                                  mdevs=mdevs, accel_info=accel_info)

        # NOTE(mdbooth): context.auth_token will not be set when we call
        #                _hard_reboot from resume_state_on_host_boot()
        if context.auth_token is not None:
            # NOTE (rmk): Re-populate any missing backing files.
            config = vconfig.LibvirtConfigGuest()
            config.parse_str(xml)
            backing_disk_info = self._get_instance_disk_info_from_config(
                config, block_device_info)
            self._create_images_and_backing(context, instance, instance_dir,
                                            backing_disk_info)

        # Initialize all the necessary networking, block devices and
        # start the instance.
        # NOTE(melwitt): Pass vifs_already_plugged=True here even though we've
        # unplugged vifs earlier. The behavior of neutron plug events depends
        # on which vif type we're using and we are working with a stale network
        # info cache here, so won't rely on waiting for neutron plug events.
        # vifs_already_plugged=True means "do not wait for neutron plug events"
        self._create_domain_and_network(context, xml, instance, network_info,
                                        block_device_info=block_device_info,
                                        vifs_already_plugged=True)
        self._prepare_pci_devices_for_use(
            pci_manager.get_instance_pci_devs(instance, 'all'))

        def _wait_for_reboot():
            """Called at an interval until the VM is running again."""
            state = self.get_info(instance).state

            if state == power_state.RUNNING:
                LOG.info("Instance rebooted successfully.",
                         instance=instance)
                raise loopingcall.LoopingCallDone()

        timer = loopingcall.FixedIntervalLoopingCall(_wait_for_reboot)
        timer.start(interval=0.5).wait()

这个直接就destroy了，属于直接把虚拟机断电了，然后将相应的外设信息清除；再建立新的xml文件的信息，重建虚拟机