trove backup

2016-12-19

introduction

Code Reading

backup

api:create_backup —-> taskmanager:create_backup —-> self.guest.create_backup —-> rabbitmq —-> guestagent:mysql_manager:create_backup —-> innobackupex —-> swift put

taskmanager 中的 create_backup

trove/taskmanager/manager.py

class Manager(periodic_task.PeriodicTasks):
 def create_backup(self, context, backup_info, instance_id):
 instance_tasks = models.BuiltInstanceTasks.load(context, instance_id)
 instance_tasks.create_backup(backup_info)

trove/taskmanager/models.py

class BuiltInstanceTasks(BuiltInstance, NotifyMixin, ConfigurationMixin):
 def create_backup(self, backup_info):
 LOG.info(_(“Initiating backup for instance %s.”) % self.id)
 self.guest.create_backup(backup_info)

这里的 self.guest trove/instance/models.py

1
2
3

class BaseInstance(SimpleInstance):
 def get_guest(self):
 return create_guest_client(self.context, self.db_info.id)

trove/common/remote.py

create_guest_client = import_class(CONF.remote_guest_client)

def guest_client(context, id, manager=None):
 from trove.guestagent.api import API
 if manager:
 clazz = strategy.load_guestagent_strategy(manager).guest_client_class
 else:
 clazz = API
 return clazz(context, id)

所以 guest 对应的是 API 实例, cast()发送消息。guestagent 接受到消息

guestagent 中的 create_backup

trove/guestagent/datastore/mysql/manager.py

1
2
3

class Manager(periodic_task.PeriodicTasks):
 def create_backup(self, context, backup_info):
 backup.backup(context, backup_info)

trove/guestagent/backup/backupagent.py

# innobackupex 或 mysqldump
RUNNER = get_backup_strategy(STRATEGY, BACKUP_NAMESPACE)
# 额外的参数，如使用 innobackupex，可以有–user os_admin –socket /var/lib/mysql/mysql.sock
EXTRA_OPTS = CONF.backup_runner_options.get(STRATEGY, ‘’)

class BackupAgent(object):

 def stream_backup_to_storage(self, backup_info, runner, storage,
 parent_metadata={}, extra_opts=EXTRA_OPTS):
 backup_id = backup_info[‘id’]
 ctxt = trove_context.TroveContext(
 user=CONF.nova_proxy_admin_user,
 auth_token=CONF.nova_proxy_admin_pass)
 conductor = conductor_api.API(ctxt)

 # Store the size of the filesystem before the backup.
 mount_point = CONFIG_MANAGER.mount_point
 stats = get_filesystem_volume_stats(mount_point)
 backup_state = {
 ‘backup_id’: backup_id,
 ‘size’: stats.get(‘used’, 0.0),
 ‘state’: BackupState.BUILDING,
 }
 conductor.update_backup(CONF.guest_id,
 sent=timeutils.float_utcnow(),
 backup_state)
 LOG.debug(“Updated state for %s to %s.”, backup_id, backup_state)

 with runner(filename=backup_id, extra_opts=extra_opts,
 parent_metadata) as bkup:
 try:
 LOG.debug(“Starting backup %s.”, backup_id)
 success, note, checksum, location = storage.save(
 bkup.manifest,
 bkup)

 backup_state.update({
 ‘checksum’: checksum,
 ‘location’: location,
 ‘note’: note,
 ‘success’: success,
 ‘backup_type’: bkup.backup_type,
 })
 . . . 

 if not success:
 raise BackupError(note)

 meta = bkup.metadata()
 meta[‘datastore’] = backup_info[‘datastore’]
 meta[‘datastore_version’] = backup_info[
 ‘datastore_version’]
 storage.save_metadata(location, meta)

 backup_state.update({‘state’: BackupState.COMPLETED})

 return meta
 . . .

 def execute_backup(self, context, backup_info,
 runner=RUNNER, extra_opts=EXTRA_OPTS):

 LOG.debug(“Running backup %(id)s.”, backup_info)
 storage = get_storage_strategy(
 CONF.storage_strategy,
 CONF.storage_namespace)(context)

 #是否是 incremental backup
 . . . 
 self.stream_backup_to_storage(backup_info, runner, storage,
 parent_metadata, extra_opts)

RUNNER 对应的是 trove.guestagent.strategies.backup.mysql_impl:InnoBackupEx/InnoBackupExIncremental , 使用 stream backup

class BackupRunner(Strategy):

 def enter(self):
 “””Start up the process.”””
 self._run_pre_backup()
 self.run()
 return self

storage 对应的是 trove.guestagent.strategies.storage.swift:SwiftStorage , 从 stream read 出来使用 swift put 上传.

radosgw’s swift api

rgw 集成 keystone 部分配置

1
2
3

rgw keystone url = http://192.168.6.125:5000
# 若不配置该项，需要配置 rgw_keystone_admin_user rgw_keystone_admin_password rgw_keystone_admin_tenant 三项，去请求 url 得到 admin token (rgw_swift.cc:244)
rgw keystone admin token = 85cc6b3914c042fe8be37032ff03fa49

rgw 接受到 swift 请求之后处理

1	ret = handler->authorize();

请求携带 token，对 token 进行验证

1	bool RGWSwift::verify_swift_token(RGWRados store, req_state s)

validate_token 返回详细信息和其 tenant、user、role 等信息, 对应请求的 keystone API /v2.0/tokens/{tokenId}

1	validate_keystone_token(store, s->os_auth_token, &info, s->user);

1	parse_keystone_token_response(token, bl, info, t)

parse 之后的 info

1	{status = 200, auth_groups = “”, user = “a400d920a63642d4913e8f2f3afda408”, display_name = “demo”, ttl = 0}

1	update_user_info(store, info, rgw_user);

对应关系：

1 2	info user –> keystone tenant_id –> rgw uid info display_name –> keystone tenant_name –> rgw display_name

那使用 rgw swift 来作 trove backup, 但是很遗憾。trove 配置项 backup_swift_container = database_backups , 创建 backup 时为 tenant 创建一个 container 用来存放 backup，但是 rgw 的 bucket_name 是全局的，swift 中 container_name 是 tenant 下的。当我使用第二个 tenant 去创建 backups 时，会先创建database<sub>backups</sub>的 bucket，而这个 bucket 已被占用，BuceketAlreadyExist

或许可以使用 database_backups_$(tenant_id) 的 bucket 来存储，这种方式在 juno 版是可以(trove juno 使 swift DLO 方式上传)trove master 使用了 swift SLO 的方式，目前 rgw swift 不支持。so, openstack or implement s3 in trove.

LYang blog