nova boot from volume在多主机zone下的坑

2025-01-04 技术教程

测试环境：3个计算节点，分别属于3个zone

[root@controller2 ~(keystone_admin)]# nova availability-zone-list

+-----------------------+----------------------------------------+

| Name | Status |

+-----------------------+----------------------------------------+

| internal | available |

| |- controller2 | |

| nova | available |

| |- controller3 | |

| ag1 | available |

| |- controller1 | |

| ag2 | available |

| |- controller2 | |

+-----------------------+----------------------------------------+

测试方式：启动虚拟机的时候选择Booting from p_w_picpath(creates a new volume)测试

产生的原因： cinder无法识别nova的多zone，cinder能获取的zone信息有：

1、cinder-volume所在的zone

2、cinder.conf配置文件中的两个参数storage_availability_zone = nova和default_availability_zone = nova

大致过下代码调用过程：(从nova那边调用cinder那部分开始)

1、nova/virt/block_device.py->class DriverImageBlockDevicedef attachvol = volume_api.create

2、nova/volume/cinder.py-> class APIdef createitem = client.volumes.create

3、cinder/api/v2/volumes.py->class VolumeControllerdef createnew_volume = self.volume_api.create

4、cinder/volume/api.py->class APIdef createflow_engine = create_volume.get_flow

在create函数中cinder获取能获得到的zone的信息：

raw_zones = self.list_availability_zones(enable_cache=True)

availability_zones = set([az['name'] for az in raw_zones])

if CONF.storage_availability_zone:

availability_zones.add(CONF.storage_availability_zone)

deflist_availability_zones services = objects.ServiceList.get_all_by_topic ->

cinder/objects/service.py（defget_all_by_topic）db.service_get_all_by_topic ->

cinder/db/api.py（defservice_get_all_by_topic）IMPL.service_get_all_by_topic ->

cinder/db/sqlalchemy/api.py（def service_get_all_by_topic）

@require_admin_contextdefservice_get_all_by_topic(context,topic,disabled=None):query=model_query(context,models.Service,read_deleted="no").\#models在这里cinder/db/sqlalchemy/models.pyfilter_by(topic=topic)#topic默认传过来的是cinder-volumeifdisabledisnotNone:query=query.filter_by(disabled=disabled)returnquery.all()

查询数据库找出cinder-volume所在的zone

5、cinder/volume/flows/api/create_volume.py ->def get_flow

#这里是taskflow，只需关注add方法里面的东东defget_flow(db_api,p_w_picpath_service_api,availability_zones,create_what,scheduler_rpcapi=None,volume_rpcapi=None):"""Constructsandreturnstheapientrypointflow.Thisflowwilldothefollowing:1.Injectkeys&valuesfordependenttasks.2.Extractsandvalidatestheinputkeys&values.3.Reservesthequota(revertsquotaonanyfailures).4.Createsthedatabaseentry.5.Commitsthequota.6.Caststovolumemanagerorschedulerforfurtherprocessing."""flow_name=ACTION.replace(":","_")+"_api"api_flow=linear_flow.Flow(flow_name)api_flow.add(ExtractVolumeRequestTask(p_w_picpath_service_api,availability_zones,rebind={'size':'raw_size','availability_zone':'raw_availability_zone','volume_type':'raw_volume_type'}))api_flow.add(QuotaReserveTask(),EntryCreateTask(db_api),QuotaCommitTask())ifscheduler_rpcapiandvolume_rpcapi:#Thiswillcastitouttoeithertheschedulerorvolumemanagervia#therpcapisprovided.api_flow.add(VolumeCastTask(scheduler_rpcapi,volume_rpcapi,db_api))#Nowload(butdonotrun)theflowusingtheprovidedinitialdata.returntaskflow.engines.load(api_flow,store=create_what)

我们关心的在class ExtractVolumeRequestTask里

先看下入口execute方法，为什么看execute方法，你翻下taskflow怎么用的，就懂了

defexecute(self,context,size,snapshot,p_w_picpath_id,source_volume,availability_zone,volume_type,metadata,key_manager,source_replica,consistencygroup,cgsnapshot):utils.check_exclusive_options(snapshot=snapshot,p_w_picpathRef=p_w_picpath_id,source_volume=source_volume)policy.enforce_action(context,ACTION)#TODO(harlowja):whatguaranteeistherethatthesnapshotorsource#volumewillremainavailableafterwedothisinitialverification??snapshot_id=self._extract_snapshot(snapshot)source_volid=self._extract_source_volume(source_volume)source_replicaid=self._extract_source_replica(source_replica)size=self._extract_size(size,source_volume,snapshot)consistencygroup_id=self._extract_consistencygroup(consistencygroup)cgsnapshot_id=self._extract_cgsnapshot(cgsnapshot)self._check_p_w_picpath_metadata(context,p_w_picpath_id,size)availability_zone=self._extract_availability_zone(availability_zone,#关心的在这里snapshot,source_volume)#_extract_availability_zone函数def_extract_availability_zone(self,availability_zone,snapshot,source_volume):"""Extractsandreturnsavalidatedavailabilityzone.Thisfunctionwillextracttheavailabilityzone(ifnotprovided)fromthesnapshotorsource_volumeandthenperformsasetofvalidationchecksontheprovidedorextractedavailabilityzoneandthenreturnsthevalidatedavailabilityzone."""#Trytoextracttheavailabilityzonefromthecorrespondingsnapshot#orsourcevolumeifeitherisvalidsothatwecanbeinthesame#availabilityzoneasthesource.ifavailability_zoneisNone:ifsnapshot:try:availability_zone=snapshot['volume']['availability_zone']except(TypeError,KeyError):passifsource_volumeandavailability_zoneisNone:try:availability_zone=source_volume['availability_zone']except(TypeError,KeyError):passifavailability_zoneisNone:ifCONF.default_availability_zone:#default_availability_zone判断availability_zone=CONF.default_availability_zoneelse:#Forbackwardscompatibilityusethestorage_availability_zoneavailability_zone=CONF.storage_availability_zoneifavailability_zonenotinself.availability_zones:#self.availability_zones就是上述cinder-volume的zone外加cinder.conf中两个配置参数ifCONF.allow_availability_zone_fallback:#allow_availability_zone_fallback这个很关键original_az=availability_zoneavailability_zone=(CONF.default_availability_zoneorCONF.storage_availability_zone)LOG.warning(_LW("Availabilityzone'%(s_az)s'""notfound,fallingbackto""'%(s_fallback_az)s'."),{'s_az':original_az,'s_fallback_az':availability_zone})else:msg=_("Availabilityzone'%(s_az)s'isinvalid.")msg=msg%{'s_az':availability_zone}raiseexception.InvalidInput(reason=msg)#没启动allow_availability_zone_fallback的话，就抛异常了