当遇到备份或者还原操作占用较长时间时，很多人会问：
备份/还原是不是僵死了？要不要kill掉，再重来？ 到底是哪一个部分的操作占用较长时间？ 到底现在进行到什么阶段了？在SQL 2016 之前，要回答这些问题会比较困难一些，或者借助某些不受支持的方式。SQL 2016开始引入了新扩展事件 backup_restore_progress_trace 来跟踪备份和还原操作。我们可以使用它们来观察备份和还原的更详细的信息。
先创建一个XE Session观察备份：
CREATE EVENT SESSION [xe_backup] ON SERVER ADD EVENT sqlserver.backup_restore_progress_trace( ACTION( sqlos.task_time,sqlserver.database_id,sqlserver.database_name, sqlserver.nt_username,sqlserver.session_id,sqlserver.session_nt_username, sqlserver.sql_text,sqlserver.username ) ) ADD TARGET package0.event_file(SET filename=N'C:\Joe\xe\xeBackup.xel') WITH (STARTUP_STATE=ON) GO ALTER EVENT SESSION [xe_backup] ON SERVER STATE=START; GO完成备份后，再来分析我们捕获的信息：
BACKUP DATABASE [AdventureWorks2016CTP3] TO DISK = N'C:\SQL2016\MSSQL13.MSSQLSERVER\MSSQL\Backup\aw.bak' WITH NOFORMAT, INIT, NAME = N'AdventureWorks2016CTP3-Full Database Backup', SKIP, NOREWIND, NOUNLOAD, COMPRESSION, STATS = 10 GO ALTER EVENT SESSION [xe_backup] ON SERVER STATE=STOP; GO ;WITH xevent AS ( SELECT timestamp,operation_type,database_name,trace_level,trace_message,event_sequence FROM ( SELECT timestamp = xevent.value(N'(event/@timestamp)[1]', N'datetime2'), operation_type = xevent.value(N'(event/data[@name="operation_type"]/text)[1]', N'nvarchar(32)'), database_name = xevent.value(N'(event/data[@name="database_name"])[1]', N'nvarchar(128)'), trace_message = xevent.value(N'(event/data[@name="trace_message"])[1]', N'nvarchar(max)'), trace_level = xevent.value(N'(event/data[@name="trace_level"])[1]', N'nvarchar(max)'), event_sequence = xevent.value(N'(event/action[@name="event_sequence"])[1]', N'int') FROM ( SELECT xevent = CONVERT(XML, event_data) FROM sys.fn_xe_file_target_read_file(N'c:\joe\xe\xeBackup_*.xel', NULL, NULL, NULL) ) AS y ) AS xevent ) SELECT database_name,timestamp,trace_level,trace_message, Duration = COALESCE( DATEDIFF(MILLISECOND, xevent.timestamp,LEAD(xevent.timestamp, 1) OVER(ORDER BY event_sequence)),0) FROM xevent ORDER BY event_sequence;我这里的查询结果总共56行，highlight部分主要操作的信息。Duration列表示此操作所有时间。
由结果可以看到备份的trace_level分为Information of major steps in the operation和Verbose I/O related information，前者表示备份操作的中的主要步骤，后者表示某个步骤IO详细情况：
由这些信息，我们能够知道此备份操作主要的步骤有哪些，哪些步骤最耗时间。
还原我用的是上一个备份生成的文件,并且使用了REPLACE。
CREATE EVENT SESSION [xe_restore] ON SERVER ADD EVENT sqlserver.backup_restore_progress_trace( ACTION(package0.event_sequence,sqlos.task_time,sqlserver.database_id,sqlserver.database_name, sqlserver.nt_username,sqlserver.server_instance_name,sqlserver.session_id,sqlserver.sql_text)) ADD TARGET package0.event_file(SET filename=N'C:\Joe\xe\xeRestore.xel') WITH (STARTUP_STATE=ON) GO ALTER EVENT SESSION [xe_restore] ON SERVER STATE=START GO USE [master] RESTORE DATABASE [AdventureWorks2016CTP3] FROM DISK = N'C:\SQL2016\MSSQL13.MSSQLSERVER\MSSQL\Backup\aw.bak' WITH FILE = 1, NOUNLOAD, REPLACE, STATS = 5 GO观察XE的数据：
ALTER EVENT SESSION [xe_restore] ON SERVER STATE=STOP GO ;WITH xevent AS ( SELECT timestamp,operation_type,database_name,trace_level,trace_message,event_sequence FROM ( SELECT timestamp = xevent.value(N'(event/@timestamp)[1]', N'datetime2'), operation_type = xevent.value(N'(event/data[@name="operation_type"]/text)[1]', N'nvarchar(32)'), database_name = xevent.value(N'(event/data[@name="database_name"])[1]', N'nvarchar(128)'), trace_message = xevent.value(N'(event/data[@name="trace_message"])[1]', N'nvarchar(max)'), trace_level = xevent.value(N'(event/data[@name="trace_level"])[1]', N'nvarchar(max)'), event_sequence = xevent.value(N'(event/action[@name="event_sequence"])[1]', N'int') FROM ( SELECT xevent = CONVERT(XML, event_data) FROM sys.fn_xe_file_target_read_file(N'c:\joe\xe\xeRestore_*.xel', NULL, NULL, NULL) ) AS y ) AS xevent ) SELECT database_name,timestamp,trace_level,trace_message, Duration = COALESCE( DATEDIFF(MILLISECOND, xevent.timestamp,LEAD(xevent.timestamp, 1) OVER(ORDER BY event_sequence)),0) FROM xevent ORDER BY event_sequence;
通过以上信息，我们能够知道还原数据库时的主要操有哪些，哪些步骤比较耗时。
现实情况中数据库还原，roll-forward之后应该还有一个undo(roll-back)操作来撤消未提交事务的修改。
