最近在Gerrit中看到一个change:https://review.openstack.org/#/c/94295/ , 它主要是对当前在Ceph中创建虚拟机的流程的改进。如果glance的backend是ceph, 则nova创建虚拟机到RBD的流程是这样的:
通过glance从ceph中下载image --> 本地 --> 复制image到rbd
这个change的目的就是:不需要下载到本地,直接在rbd中复制image,以提高虚拟机创建的速度。
以前只知道nova从glance下载image,作为虚拟机的base,却没有仔细的了解过这个过程,正好借这个机会,看一看nova创建虚拟机到rbd过程中关于image的部分。
经过nova-scheduler选择节点后,创建VM的请求到达了nova-compute,即nova/compute/manager.py:ComputeManager._run_instance():
def _run_instance(self, context, request_spec, filter_properties, requested_networks, injected_files, admin_password, is_first_time, node, instance, legacy_bdm_in_spec):
关于镜像的元数据就保存在request_spec,
image_meta = request_spec['image']
取得元数据后就开始build_instance()了,但是下面的过程与image没太大关系,所以从简带过。
def _build_instance(self, context, request_spec, filter_properties, requested_networks, injected_files, admin_password, is_first_time, node, instance, image_meta, legacy_bdm_in_spec) ----> def _spawn(self, context, instance, image_meta, network_info, block_device_info, injected_files, admin_password, set_access_ip=False) ----> self.driver.spawn(context, instance, image_meta, injected_files, admin_password, network_info, block_device_info)
这里的driver就是你用的Hypervioser, 我用的是KVM,所以这个driver.spawn=nova/virt/libvirt/driver.py:LibvirtDriver.spawn():
def spawn(self, context, instance, image_meta, injected_files, admin_password,network_info=None, block_device_info=None): disk_info = blockinfo.get_disk_info(CONF.libvirt.virt_type, instance, block_device_info, image_meta) self._create_image(context, instance, disk_info['mapping'], network_info=network_info, block_device_info=block_device_info, files=injected_files, admin_pass=admin_password)
那么这个disk_info[‘mapping‘]是什么呢?这里有一个方法,我们可以从test_libvirt_blockinfo.py里找到答案,所以结合测试用例来看代码真的很有用。在Nova/tests/virt/libvirt/test_libvirt_blockinfo.py:
LibvirtBlockInfoTest.test_get_disk_mapping_simple_swap()里可以看到:
expect = { 'disk': {'bus': 'virtio', 'dev': 'vda', 'type': 'disk', 'boot_index': '1'}, 'disk.local': {'bus': 'virtio', 'dev': 'vdb', 'type': 'disk'}, 'root': {'bus': 'virtio', 'dev': 'vda', 'type': 'disk', 'boot_index': '1'} }
Expect就是期望disk_info[‘mapping‘]的样子。
下面就开始创建VM的image了:
def _create_image(self, context, instance, disk_mapping, suffix='', disk_images=None, network_info=None, block_device_info=None, files=None, admin_pass=None, inject_files=True): #下面这个函数就是返回VM image格式的相关类,位于 #libvirt/imagebackend.py中,这里image是rbd, 返回的就是Rbd #类。 def image(fname, image_type=CONF.libvirt.images_type): return self.image_backend.image(instance, fname + suffix, image_type) ...... if not booted_from_volume: root_fname = imagecache.get_cache_fname(disk_images, 'image_id')#以image id作为文件名 size = instance['root_gb'] * units.Gi if size == 0 or suffix == '.rescue': size = None #这里有点复杂,用到了回调函数fetch_image,这里的cache是Rbd的父类#Image类的cache(),主要功能是从模板,也就是glance的image, 为VM创建一个image. image('disk').cache(fetch_func=libvirt_utils.fetch_image, context=context, filename=root_fname, size=size, image_id=disk_images['image_id'], user_id=instance['user_id'], project_id=instance['project_id']) #可以在cache()中看到: if not self.check_image_exists() or not os.path.exists(base): self.create_image(fetch_func_sync, base, size, *args, **kwargs)
那么现在到class Rbd下可以找到create_image():
def create_image(self, prepare_template, base, size, *args, **kwargs): if self.rbd is None: raise RuntimeError(_('rbd python libraries not found')) if not os.path.exists(base): prepare_template(target=base, max_size=size, *args, **kwargs)##这里的prepare_temple()就是 libvirt_utils.fetch_image啦。
libvirt_utils.fetch_image=nova/virt/libvirt/utils.fetch_image():
def fetch_image(context, target, image_id, user_id, project_id, max_size=0): """Grab image.""" images.fetch_to_raw(context, image_id, target, user_id, project_id, max_size=max_size) ---> def fetch_to_raw(context, image_href, path, user_id, project_id, max_size=0): path_tmp = "%s.part" % path fetch(context, image_href, path_tmp, user_id, project_id, max_size=max_size) ----> def fetch(context, image_href, path, _user_id, _project_id, max_size=0): (image_service, image_id) = glance.get_remote_image_service( context, image_href)#从glance获取image with fileutils.remove_path_on_error(path): #这里就是把image_id的数据download到path了。Download()位于 #nova/image/glance.py。 image_service.download(context, image_id, dst_path=path)
回到class Rbd的create_image()中,
libvirt_utils.import_rbd_image(*args)把path的image数据写入rbd,至此,整个流程就到这里结束了。
现在回到文章开始中提到的那个change, 看看它是怎么实现的。
首先它在fetch_to_raw中加入了一个判断。
def fetch_to_raw(context, image_href, path, user_id, project_id, max_size=0): #判断backend是否具有‘direct_fetch’的属性,如果有,则直接返回#direct_fetch() if backend and hasattr(backend, 'direct_fetch'): try: return backend.direct_fetch(context, image_href) except exception.ImageUnacceptable: LOG.debug(_('could not fetch directly, falling back to download'))
给Rbd类添加了一个属性:
def direct_fetch(self, context, image_href): #判断driver是否支持layering(分层 #http://ceph.com/docs/firefly/dev/rbd-layering/ ,指的是块设备 #的cow克隆,支持快速创建image) if not self.driver.supports_layering(): reason = _('installed version of librbd does not support cloning') raise exception.ImageUnacceptable(image_id=image_href, reason=reason) image_meta, locations = images.get_meta(context, image_href) LOG.debug(_('Image locations are: %(locs)s') % {'locs': locations}) if image_meta.get('disk_format') not in ['raw', 'iso']: reason = _('Image is not raw format') raise exception.ImageUnacceptable(image_id=image_href, reason=reason) #克隆镜像(http://ceph.com/docs/master/rbd/librbdpy/) for location in locations: if self.driver.is_cloneable(location, image_meta): return self.driver.clone(location, self.rbd_name) reason = _('No image locations are accessible') raise exception.ImageUnacceptable(image_id=image_href, reason=reason)
这样就不需要想1中的那样先把image下载到local, 在写到rbd中,直接在rbd中克隆,从而提高了虚拟机的创建速度。
借这个机会既熟悉了创建VM时的image流程,又熟悉了ceph的用法,同时学习了高手们是怎么实现一个功能的,看来review的益处大大的呀。:)
在Ceph中创建虚拟机流程改进之分析,布布扣,bubuko.com
原文:http://blog.csdn.net/epugv/article/details/27970625