openstack快照分析

1. OpenStack快照类别：虚拟机快照（快照存放于glance）：nova image-createcold snapshotlive snapshot with no downtime卷快照（LVM快照存放于本地）: nova volume-snapshot-create对openstack而言，虚拟机的快照即是镜像，虚拟机快

liukuan73

12875人浏览 · 2015-06-11 14:40:42

liukuan73 · 2015-06-11 14:40:42 发布

1.openstack快照

1.1.OpenStack快照类别

注意：不要混淆为卷创建快照和为以卷作为后端存储的虚拟机创建快照

虚拟机快照（快照存放于glance）：nova image-create <server> <name>
cold snapshot:创建snapshot时，需暂停虚拟机。
live snapshot with no downtime:创建snapshot时，无需暂停虚拟机。（目前未实现，存在的问题见此链接：http://blog.chinaunix.net/uid-29718549-id-4346700.html）

volume backed以外类型的虚拟机，在nova-api中调用API.snapshot接口创建快照（最终实现在createImage接口）

volume backed的虚拟机，在nova-api中调用API.snapshot_volume_backed接口创建虚拟机快照

卷快照（LVM快照存放于本地）: nova volume-snapshot-create <volume-id>

在nova-api中调用API.volume_snapshot_create接口创建虚拟机快照

1.2.openstack创建快照命令

nova image-create

usage: nova image-create [--show] [--poll] <server> <name>

Create a new image by taking a snapshot of a running server.

Positional arguments

Name or ID of server.

<name>

Name of snapshot.

Optional arguments

--show

Print image info.

--poll

Report the snapshot progress and poll until image creation is complete.

nova volume-snapshot-create

usage: nova volume-snapshot-create [--force <True|False>]
                                   [--display-name <display-name>]
                                   [--display-description <display-description>]
                                   <volume-id>

Add a new snapshot.

Positional arguments

<volume-id>

ID of the volume to snapshot

Optional arguments

--force <True|False>

Optional flag to indicate whether to snapshot a volume even if its attached to a server. (Default=False)

--display-name <display-name>

Optional snapshot name. (Default=None)

--display-description <display-description>

Optional snapshot description. (Default=None)

1.3.流程

cold snapshot 流程:

# Save the state and stop a running guest, then detach pci devices

  $ virsh managedsave vm (managedsave流程见此链接：http://blog.chinaunix.net/uid-29718549-id-4772180.html)

# Create a qemu internal snapshot

  $ qemu-img snapshot -c snap1   vm_snapshot

# Extract the internal snapshot, convert it to qcow2 and export it a file, then upload to glance

  $ qemu-img convert -f qcow2 vm -O qcow2 vm_snapshot

# Start the guest again

  $ virsh start vm

live snapshot 流程:

# Abort any failed/finished block operations:

$ virsh blockjob vm vda --abort

# Undefine a running domain. (Note: Undefining a running domain does not _kill_ the domain, it just converts it from persistent to transient.)

$ virsh undefine vm

# create a destination image with the original backing file and matching size of the instance root disk.

$ qemu-img create -f qcow2 vm_copy --backing_file=backing_file --size=root_disk_size

#Invoke 'virsh blockcopy' (This will take time, depending on the size of disk image vm1):

   $ virsh blockcopy --domain vm vda vm_copy --wait --verbose

#Abort any failed/finished block operations:

    $ virsh blockjob vm vda --abort

#Define the guest again (to make it persistent):

    $ virsh define vm

#From the obtained new copy, convert the QCOW2 with a backing file to a qcow2 image with no backing file, then upload to glance:

    $ qemu-img convert -f qcow2 -O raw vm_copy vm_convert

2.与传统快照的区别

对openstack而言，虚拟机的快照其实就是镜像。这是因为Openstack是采用创建镜像的方式创建快照（qemu-img convert -f qcow2 -O qcow2 <disk_path> <out_path>），而不是通过正统的virsh或者qemu的live snapshot创建快照函数( virsh snapshot-create-as --live ....)来实现的。虚拟机快照做完后以镜像形式存于glance。 openstack的快照与传统快照有以下区别：

libvirt 主流快照实现：采用virDomainSnapshotCreateXML()函数(CLI为virsh snapshot-create)。新建的快照与虚拟机有关联：若为内置快照，快照信息和虚拟机存在同一个qcow2镜像中；若为外置快照，新建一个qcow2文件，原虚拟机的disk将变为一个read only的模板镜像，新qcow2镜像仅记录与2.模板镜像的差异数据。这种快照包含快照链信息，可保留disk和ram信息，可回滚至快照点。
openstack快照实现：openstack并未采用virDomainSnapshotCreateXML()来实现快照，而是单纯的对虚拟机镜像做转换和拷贝，生成一个与虚拟机无关联的镜像，最后上传至glance中。这种快照不包含快照链信息，只保留disk信息，无法回滚至快照点，只能采用该快照镜像创建一个新的虚拟机。

限制与缺点：

没有快照链信息，不支持revert恢复虚拟机到某一个快照点
只对系统盘进行快照，
不支持内存快照，不支持同时对虚拟机和磁盘做快照
需要用户进行一致性操作
不支持含元数据导出，
不支持含元数据导入
只支持虚拟机全量数据快照（与快照的实现方式有关，因为是通过image进行保存的）
过程较长（需要先通过存储快照，然后抽取并上传至glance)
快照以Image方式保存，而非以cinder卷方式保存，无法充分利用存储本身能力加快快照的创建和使用
当前限制openstack的虚拟机快照只快照root盘，不快照内存/CPU状态以及挂载磁盘。挂载磁盘需要事先卸载磁盘(数据盘），然后进行快照，然后再挂载磁盘
没有快照链信息

3.OpenStack快照源代码流程分析

3.1 nova/compute/api.py

 
   # NOTE(melwitt): We don't check instance lock for snapshot because lock is
 
     # intended to prevent accidental change/delete of instances
 
     @wrap_check_policy
 
     @check_instance_cell
 
     @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.STOPPED,
 
                                     vm_states.PAUSED, vm_states.SUSPENDED])
 
     def snapshot(self, context, instance, name, extra_properties=None):
 
         """Snapshot the given instance.
 
         :param instance: nova.db.sqlalchemy.models.Instance
 
         :param name: name of the snapshot
 
         :param extra_properties: dict of extra image properties to include
 
                                  when creating the image.
 
         :returns: A dict containing image metadata
 
         """
        #_create_image函数中调用glance api创建image entry，为后将snapshot上传为镜像做准备，虽然镜像和snapshot在可以上传到glance作为镜像启动虚拟机，
        #但是为了区分二者的不同，glance将镜像和snapshot标记为不同的类型：type=image 和 type=snapshot
 
         image_meta = self._create_image(context, instance, name,
 
                                         'snapshot',
 
                                         extra_properties=extra_properties)
 
         # NOTE(comstud): Any changes to this method should also be made
 
         # to the snapshot_instance() method in nova/cells/messaging.py
         # 将任务状态(task state) 设置为:image_snapshot_pending
 
         instance.task_state = task_states.IMAGE_SNAPSHOT_PENDING
 
         instance.save(expected_task_state=[None])
 
         #通过rpc调用nova/compute/rpcapi.py的snapshot_instance函数
 
         self.compute_rpcapi.snapshot_instance(context, instance,
 
                                               image_meta['id

3.2 nova/compute/rpcapi.py

 
   #梳理下流程： 
 （1）用户发起create snapshot的请求; 
 （2）nova-api服务接收到这个请求并进行前期处理，即3.1中代码的处理流程;
 （3）真正的snapshot操作是需要在nova-compute节点上执行的，所以nova-api需要向nova-compute发送message
 #由于OpenStack环境中会有很多个nova-compute,所以需要通过server=_compute_host(None, instance)来获取虚拟机所在的host，并向其发送message。
 def snapshot_instance(self, ctxt, instance, image_id):
 
         version = '3.0'
 
         cctxt = self.client.prepare(server=_compute_host(None, instance),
 
                 version=version)
 
         cctxt.cast(ctxt, 'snapshot_instance',
 
                    instance=instance,
 
                    image_id=image_id)

3.3 nova/virt/libvirt/driver.py

 
    def snapshot(self, context, instance, image_id, update_task_state):
 
         """Create snapshot from a running VM instance.
 
         This command only works with qemu 0.14+
 
         """
 
         try:
 
             virt_dom = self._get_domain(instance)
 
         except exception.InstanceNotFound:
 
             raise exception.InstanceNotRunning(instance_id=instance['uuid'])
 
         base_image_ref = instance['image_ref']
 
         base = compute_utils.get_image_metadata(
 
             context, self._image_api, base_image_ref, instance)
 
         snapshot = self._image_api.get(context, image_id)
 
         disk_path = libvirt_utils.find_disk(virt_dom)
 
         source_format = libvirt_utils.get_disk_type(disk_path)
 
         image_format = CONF.libvirt.snapshot_image_format or source_format
 
         # NOTE(bfilippov): save lvm and rbd as raw
 
         if image_format == 'lvm' or image_format == 'rbd':
 
             image_format = 'raw'
 
         metadata = self._create_snapshot_metadata(base,
 
                                                   instance,
 
                                                   image_format,
 
                                                   snapshot['name'])
 
         snapshot_name = uuid.uuid4().hex
 
         state = LIBVIRT_POWER_STATE[virt_dom.info()[0]]
 
         # NOTE(rmk): Live snapshots require QEMU 1.3 and Libvirt 1.0.0.
 
         # These restrictions can be relaxed as other configurations
 
         # can be validated.
 
         # NOTE(dgenin): Instances with LVM encrypted ephemeral storage require
 
         # cold snapshots. Currently, checking for encryption is
 
         # redundant because LVM supports only cold snapshots.
 
         # It is necessary in case this situation changes in the
 
         # future.
         #这里需要注意，解释了为啥现在是cold snapshot而不是live snapshot:
         # 有人提过live snapshot的bug，社区认为live snapshot目前不稳定，所以默认条件下采用cold snapshot，并且是通过硬编码来实现的
         # 看下面这个判断条件，成立的时候将live_snapshot = true,其中MIN_LIBVIRT_LIVESNAPSHOT_VERSION=1.3.0, 其实现在libvirt的最新版本
 
         # 才到1.2.11， 所以这个live_snapshot的条件不满足，就变成了cold_snapshot
 
         if (self._host.has_min_version(MIN_LIBVIRT_LIVESNAPSHOT_VERSION,
 
                                        MIN_QEMU_LIVESNAPSHOT_VERSION,
 
                                        REQ_HYPERVISOR_LIVESNAPSHOT)
 
              and source_format not in ('lvm', 'rbd')
 
              and not CONF.ephemeral_storage_encryption.enabled):
 
             live_snapshot = True
 
             # Abort is an idempotent operation, so make sure any block
 
             # jobs which may have failed are ended. This operation also
 
             # confirms the running instance, as opposed to the system as a
 
             # whole, has a new enough version of the hypervisor (bug 1193146).
 
             try:
 
                 virt_dom.blockJobAbort(disk_path, 0)#live第一步
 
             except libvirt.libvirtError as ex:
 
                 error_code = ex.get_error_code()
 
                 if error_code == libvirt.VIR_ERR_CONFIG_UNSUPPORTED:
 
                     live_snapshot = False
 
                 else:
 
                     pass
 
         else:
 
             live_snapshot = False
 
         # NOTE(rmk): We cannot perform live snapshots when a managedSave
 
         # file is present, so we will use the cold/legacy method
 
         # for instances which are shutdown.
 
         if state == power_state.SHUTDOWN:
 
             live_snapshot = False
 
         # NOTE(dkang): managedSave does not work for LXC
         #注意这里，如果live_snashot目前是false,在做snapshot之前先要执行：
         #（1）_detach_pci_devices， 卸载虚拟机挂载的pci设备，比如数据盘
         #（2）self._detach_sriov_ports， 卸载虚拟机挂载的SRIOV设备，比如支持SRIOV的网卡设备
 
         if CONF.libvirt.virt_type != 'lxc' and not live_snapshot:
 
             if state == power_state.RUNNING or state == power_state.PAUSED:
 
                 self._detach_pci_devices(virt_dom,
 
                     pci_manager.get_instance_pci_devs(instance))
 
                 self._detach_sriov_ports(instance, virt_dom)
 
                 virt_dom.managedSave(0)#cold第一步
 
         #判断虚拟机的后端存储是什么，不同的后端存储做snapshot是不同的，本地文件系统的话，默认qcow2
 
         snapshot_backend = self.image_backend.snapshot(instance,
 
                 disk_path,
 
                 image_type=source_format)
 
         if live_snapshot:
 
             LOG.info(_LI("Beginning live snapshot process"),
 
                      instance=instance)
 
         else:
 
             LOG.info(_LI("Beginning cold snapshot process"),
 
                      instance=instance)
 
         #更新任务的状态为：image_pending_upload, 大家都知道做完snapshot要上传
 
         update_task_state(task_state=task_states.IMAGE_PENDING_UPLOAD)
         #目前做快照的过程是：
         #（1）先在../data/nova/instance/snapshots目录下生成临时目录，比如nova/instances/snapshots/tmptHr585
         #（2）然后将快照生成到这个目录，具体参见snapshot_backend.snapshot_extract(out_path, image_format)这个函数
         #（3）生成完成后，通过glance api上传，具体参见 self._image_api.update
         snapshot_directory = CONF.libvirt.snapshots_directory
 
         fileutils.ensure_tree(snapshot_directory)
 
         with utils.tempdir(dir=snapshot_directory) as tmpdir:
 
             try:
 
                 out_path = os.path.join(tmpdir, snapshot_name)
 
                 if live_snapshot:
 
                     # NOTE(xqueralt): libvirt needs o+x in the temp directory
 
                     os.chmod(tmpdir, 0o701)
 
                     self._live_snapshot(context, instance, virt_dom, disk_path,
 
                                         out_path, image_format, base)#热迁移
 
                 else:
                     #这个函数实际执行了一条命令： qemu-img convert -f qcow2 -O qcow2  disk_path out_path,算是生成了快照
 
                     snapshot_backend.snapshot_extract(out_path, image_format)
 
             finally:
 
                 new_dom = None
 
                 # NOTE(dkang): because previous managedSave is not called
 
                 # for LXC, _create_domain must not be called.
 
                 if CONF.libvirt.virt_type != 'lxc' and not live_snapshot:
 
                     if state == power_state.RUNNING:
 
                         new_dom = self._create_domain(domain=virt_dom)   ##恢复做快照之前虚拟机的状态
 
                     elif state == power_state.PAUSED:
 
                         new_dom = self._create_domain(domain=virt_dom,
 
                                 launch_flags=libvirt.VIR_DOMAIN_START_PAUSED)
 
                     if new_dom is not None:
 
                         self._attach_pci_devices(new_dom,
 
                             pci_manager.get_instance_pci_devs(instance))
 
                         self._attach_sriov_ports(context, instance, new_dom)
 
                 LOG.info(_LI("Snapshot extracted, beginning image upload"),
 
                          instance=instance)
 
             # Upload that image to the image service
 
             update_task_state(task_state=task_states.IMAGE_UPLOADING,
 
                      expected_state=task_states.IMAGE_PENDING_UPLOAD)
 
             with libvirt_utils.file_open(out_path) as image_file:       ###将生成的快照上传到glance
 
                 self._image_api.update(context,
 
                                        image_id,
 
                                        metadata,
 
                                        image_file)
 
                 LOG.info(_LI("Snapshot image upload complete"),
 
                          instance=instance)