SQL Database Point-In-Time Restore of Transaction Log Backup

A Teaching Assistant (TA) deleted all the grades in an assignment in our online learning management system, eLearning (powered by Blackboard Learning System CE). What happened was a series of actions that leaded to this. Here is what was described to me (I'm yet to be able to successfully simulate it):

The TA used the grade book tool to enter the grades of all students in a particular assignment (instead of using the assignment tool). He then changed the due date of that assignment to a later date. The system gave him a warning message, but he simply went ahead to proceed. He subsequently found that all the grades he entered for that assignment were gone.

Now I had to find a way to retrieve the grades from our old backups. But the challenge was he did not remember when (approximate date and time) he entered the grades, and when he modified the assignment properties which caused all the grades to be cleared.

I parsed our web server logs (using Cygwin's gawk) looking for all his POST requests with grade book and other strings in the URL. Yes! I saw the log entires dated April 14 around midnight, with the last update at 0:56am.

Therefore, I restored our elearning database backups (onto a test instance) to the April 14 0:57am state. Here is the SQL script.

restore database webctdatabase from DISK='D:\backupfolder\webctdatabase_backup_200904120005.bak' WITH NORECOVERY, REPLACE

restore database webctdatabase from DISK='D:\backupfolder\webctdatabase_backup_200904140005.dif' WITH NORECOVERY

restore log webctdatabase from DISK='D:\backupfolder\webctdatabase_backup_200904140250.trn' WITH RECOVERY, STOPAT = 'Apr 14, 2009 00:57:00 AM'

restore database webctdatabase$PF from DISK='D:\backupfolder\webctdatabase$PF_backup_200904120005.bak' WITH NORECOVERY, REPLACE

restore database webctdatabase$PF from DISK='D:\backupfolder\webctdatabase$PF_backup_200904140005.dif' WITH NORECOVERY

restore log webctdatabase$PF from DISK='D:\backupfolder\webctdatabase$PF_backup_200904140250.trn' WITH RECOVERY, STOPAT = 'Apr 14, 2009 00:57:00 AM'

A little background info: We ran eLearning full database backup once a week (Sunday), a differential backup nightly (Monday to Saturday), and transaction log backups throughout the day. Therefore, when restoring from backup, I needed to restore the full database backup (as of April 12 Sunday), the last differential backup (as of April 14 Tuesday), and the transaction log backup to the specific point-in-time.

After the restore was completed, I also had to execute the following SQL commands to set the webct and webct$PF users' SID on the test instance properly.

Use webctdatabase

go

sp_change_users_login 'report'

Use webctdatabase

go

sp_change_users_login 'update_one', 'webct', 'webct'

Use webctdatabase

go

sp_change_users_login 'update_one', 'webct$PF', 'webct$PF'

Use webctdatabase$PF

go

sp_change_users_login 'report'

Use webctdatabase$PF

go

sp_change_users_login 'update_one', 'webct', 'webct'

Use webctdatabase$PF

go

sp_change_users_login 'update_one', 'webct$PF', 'webct$PF'

Then I restarted the webct service on the frontend node of the test instance to bring up the application. I logoned to it and could see all the student grades in that section/assignment. We exported them for the TA.

Saved by the transaction log backups.

APC UPS Monitoring using SNMP

I just finished configuring IPSentry to monitor our APC UPS (via SNMP).

Currently, IPSentry will send email alerts to us when the UPS batteries need replacing, etc. But if the main power is out and the battery run time remaining is under certain threshold, it will send us alerts to our cell phones instead. We also tested using it to call a script which automate the shutdown of our servers and it works. We may decide to implement it eventually, but I feel we need more observations to verify the reliability of the monitoring before doing so (since that script is so powerful and any false alarm can cause our entire shop to be shutdown).

Anyway, here are the OIDs we are now monitoring:

OID	Description	Date Type
.1.3.6.1.4.1.318.1.1.1.2.1.2	The elapsed time since the UPS has switched to battery power.	timeticks n mins = n*6000 timeticks
.1.3.6.1.4.1.318.1.1.1.2.2.1	The remaining battery capacity expressed in percent of full capacity.	gauge
.1.3.6.1.4.1.318.1.1.1.2.2.2	The current internal UPS temperature expressed in Celsius.	gauge
.1.3.6.1.4.1.318.1.1.1.2.2.3	The UPS battery run time remaining before battery exhaustion.	timeticks n mins = n*6000 timeticks
.1.3.6.1.4.1.318.1.1.1.2.2.4	Indicates whether the UPS batteries need replacing.	integer noBatteryNeedsReplacing (1) batteryNeedsReplacing (2)

More eLearning User Penetration Numbers

In December 2007, I wrote a script to determine the user penetration rate of eLearning, our online Learning Management System (powered by Blackboard Learning System CE6). We have subsequently upgraded the system to CE8 with all the latest service packs. Fortunately, the script continues to work.

Here are the numbers for the main Pullman campus:

Semester	Undergraduate Students			Graduate Students			Undergraduate + Graduate Students
	Total	LMS	%	Total	LMS	%	Total	LMS	%
2007 Fall	15441	12862	83.30	1981	749	37.81	17422	13611	78.13
2008 Spring	14383	12079	83.98	1874	747	39.86	16257	12826	78.90
2008 Fall	15571	14521	93.26	1998	1165	58.31	17569	15686	89.28
2009 Spring	14650	12968	88.52	1939	936	48.27	16589	13904	83.81

Using rsync on Windows

I posted an article on an issue of SQL Memory Paged Out During Large File Copy last month, which documented my effort in finding a way to solve the issue. Yes, rsync (from Cygwin) is still our final choice.

Today I'm going to post some tips on using rsync on Windows.

To install rsync, run the Cygwin setup.exe installer and install rsync package. After that, I would recommend adding Cygwin's bin folder into your environment path. This way, you can call rsync directly from the command line.

• Right click on My Computer. Choose Properties.
• Click on Advanced tab.
  
• Click on Environment Variables button. (The Environment Variables window will be opened)
  
• Under System variables, highlight Path. Click Edit.
  
• At the end of the Variable value, append ;c:\cygwin\bin (or wherever your installation path is. Don't forget the semi-colon in front).
  
• Click OK | OK | OK to finish.
  

Now you can use rsync to copy file in the command window, like the followings:

rsync.exe \\fromserver\share\filename d:\backupfolder\filename

But there is another issue. The destination file (copied via rsync) is owned by the user account running rsync. One may prefer to have the file permissions inherited from the parent folder instead. To fix that, I use the setacl utility from SourceForge.

The following commands first change the file ownership to local Administrators group of the backup server, clear the discretionary and security access control lists (DACL, SACL) of the file, and finally set the ACLs to be inherited from the parent folder.

setacl.exe -on filename -ot file -actn setowner -ownr "n:backupserver\Administrators;s:n"
setacl.exe -on filename -ot file -actn clear -clr dacl,sacl

setacl.exe -on filename -ot file -actn setprot -op "dacl:np;sacl:np"

Put them all into a batch file, and you have a simple script to perform nightly backup using rsync.

Web Slides on Harvesting Grade Book

Diigo has a feature where you can create a "List" of bookmarks. Inside the list, you can order the bookmarks in any sequence you want, and create a slide show from it.

I just created a series called The Evolution of Harvesting Grade Book. It tells the story behind some of the transformative assessment projects our department has been working on - visualized through what is currently called the "Harvesting Grade Book".

SQL Memory Paged Out During Large File Copy

We currently use SQL server maintenance plan to generate full database backup files (nightly or weekly, depending on the application), differential and transaction log backup files throughout the day. We then run a script to robocopy (from Windows Server 2003 Resource Kit Tools) the generated backup files offsite.

Ever since we migrated our databases from SQL server 32-bit to 64-bit, we started encountering a problem where SQL server process memory was being paged out. As a result, our applications became momentarily unresponsive and unavailable to our customers. This happened consistently around the same time every night, which coincided with the time when the offsite copying of SQL database full backup files was running. Error message in the SQL server log was:

A significant part of sql server process memory has been paged out. This may result in a performance degradation. Duration: 0 seconds. Working set (KB): xxxxxx, committed (KB): xxxxxx, memory utilization: xx%.

According to this Microsoft Article: How to reduce paging of buffer pool memory in the 64-bit version of SQL Server 2005, only SQL Server 2005 Enterprise Edition can use the Lock pages in memory to reduce paging. We are running SQL Server 2005 Standard Edition. We may need to upgrade/ migrate to Enterprise Edition. This means additional costs of licensing and migration to consider.

In the mean time, our applications are failing momentarily every night. I need to come up with a better approach now.

Our requirements are pretty simple (listed in the order of importance):

• No paging of SQL server memory.
  
• Can be automated via GUI, or script.
• Still maintain a copy of database backup files locally. (This is preferable because in the event of a problem that requires a full database restore, the process would be much faster)
  
• Fast.

Is that too much to ask for?

Google helped me locate this article Technet's Ask the Performance Team Blog: Slow Large File Copy Issue. It has a pretty good explanation of what is going on (the use of buffered I/O) during the copy process which leads to the paging problem we encountered.

I followed the recommendation in the article and tested using eseutil to perform offsite copying of our database backup files. Unfortunately, it did not work as expected. SQL server process memory was still being paged out.

I continued to investigate options. Here is a list of what I have looked into:

• cat (from Cygwin)
  
• split to remote share and cat to combine (from Cygwin)
  
• split locally, copy to remote share, and cat to combine (from Cygwin)
  
• rsync (from Cygwin)
  
• Symantec Backup Exec backup agent
  
• Symantec Backup Exec SQL backup agent
  
• Windows NTBackup
  
• SQL backup to remote share
  
• SQL backup using MIRROR TO clause (Eliminated since it is available in SQL Server 2005 Enterpise Edition only)
  
• SQL database mirroring and backup from mirrored databases (Eliminated since backup option is not available in the read-only mirrored databases)
  
• SQL backup to a file server cluster volume. After database backup files are generated, failover to the other file server node for the offsite backup to run against (Not recommended due to additional complexity. We may better off migrate to SQL Server 2005 Enterprise Edition)

After some testings, the winner (or the best compromise) is rsync.

We continue to use SQL server maintenance plan to generate all database backup files. The files are kept locally on the SQL server. We then run a script to rsync the files offsite. It never pages out our SQL server process memory. It still takes a long time to run, but compare to other approaches, this is the fastest one.

Our largest database is the webctdatabase of our Blackboard LMS. It is now about 400GB, and it takes about 15 hours for rsync to copy it offsite.

SQL Injection Exploit... of a Mom

It's brilliant.