You are probably wondering why am I writing a blog post on concurrent backups! Well this is a topic that brings up a few questions during my customer visits. Apart from that the biggest reason, I want to talk about concurrent backups is because of backup softwares that you have active in your environments.
A lot of the backup softwares available in the market have the option of scheduling backups of the entire system. This involves backing up of SQL Server databases using the interfaces exposed through sqlvdi.dll. Ain’t your software really smart? However, sometimes the right hand doesn’t really talk to the left hand! It is quite possible that your server admins have scheduled backups of your systems around the same time you have scheduled a maintenance plan to take a backup of the SQL Server instance.
If you initiate two differential or full database backups concurrently, then you will experience a blocking condition as shown in the table below. SPID 53 was taking a VDI full-backup using the SQL Server Backup Simulator tool where as SPID 55 was also taking a native full-backup of the same database. In such a scenario, you will find that the backup which was running first will complete and only then will the second backup will continue. The second concurrent backup will be blocked during the entire duration of the first backup.
|55||53||2386||LCK_M_U||DB: 8 [BULKOP_BACKUP_DB]||BACKUP DATABASE||8|
This issue doesn’t manifest itself when you are you taking a volume snapshot backup of the database. The wait resource will always be DB: <dbid> [BULKOP_BACKUP_DB] with a wait type of LCK_M_U (update lock) for the second session that is running the concurrent backup operation. Concurrent log backups are allowed though starting from SQL Server 2005.
One of the situations where this can pose to be problematic is when the application that installed the database creates a maintenance plan during the application setup process. This can cause problems if you have concurrent full or differential backups scheduled from another software during the same time when your backup maintenance plan is running. Another big pitfall of concurrent backups running is that sometimes, the same job is responsible for both full and log backups of the database. If your log backup job doesn’t kick-off because your database backup is still being blocked, then you could run into a bloated transaction log issue for a database in full or bulk-logged recovery model. My colleague, Arvind, has also blogged about the same issue a few years back.
The following T-SQL script can help you determine if you have overlapping backups being taken for a SQL Server 2012 function. This script can also work for releases lower than SQL Server 2012 if you remove the LAG function as that T-SQL enhancement was introduced in SQL Server 2012.
select backup_start_date, backup_finish_date, LAG(backup_start_date, 1) OVER (PARTITION BY database_name ORDER BY backup_start_date) as previous_start_date, CASE datediff(mi,(LAG(backup_start_date, 1) OVER (PARTITION BY database_name ORDER BY backup_start_date)),backup_start_date) WHEN 0 THEN 'OVERLAPPING' ELSE '-' END as [Status] , datediff(mi,backup_start_date,backup_finish_date) as backup_time_mins, type from msdb..backupset where database_name = '<database name>' -- Replace with appropriate database name and type <> 'L' order by backup_start_date