Voice of the DBA

Tag: Backup/Recovery

Always Check on the Basics

I’ve been working with SQL Server for a long time, and one of the things I’ve learned is to not assume others view the platform and its administration needs in the same way that I do. I have usually started examining new instances with the same skepticism I’d use if my Mom told me she’d installed the software. I’m sure she could do it, and likely use some wizard and Google to get some backup scheme implemented, but I don’t know that it would be the schema I’d want to use.

This week I noticed a piece from Lori Brown, of SQLRx, which talked about a few basic settings that I’d always want running on my systems. One of these is the CHECKSUM setting. It’s a checkbox in the SSMS dialog, and an option in T-SQL. Most third party tools, like SQL Backup Pro, include similar settings. To me, this ought not to be a setting, but rather a default that always runs. NO_CHECKSUM is the default, which is silly in 2019.

In any case, I’ve seen more than a few presentations on the backup process in SQL Server. They always seem to be beginner sessions, always have more people than I expect, and remind me that this process, which is solid and stable, still has a lot that people don’t think about. There are certainly nuances to performing backups, and restores, in a manner that doesn’t generate any RGEs.

I don’t usually use the VERIFYONLY option, as to me the file isn’t really tested until it’s restore. This is one reason I recommend having a process to regularly restore your backup files on a test system. Not for use, though you can certainly use them, but more just to ensure your file system, your storage network, all the hardware involved hasn’t caused any issues with the backup file. If you build a server for this process, make sure you add enough RAM, as someone recently learned.

My feeling is that backup and restore is the most critical aspect of managing your SQL Server instances. This is the first thing I get working, and the number one ongoing concern I have to ensuring data is available. Second would be security, and everything else follows from there, but having a solid backup and restore process is the foundation of all other system administration.

There are lots of ways you can learn more. We have articles, a free ebook, and more at SQLServerCentral. The best way, however, is what Lori has done. Do some testing. Run through some scenarios, check how long things take in your environment, and ensure that your backups are capable of meeting the RTO and RPO needs of your organization.

Steve Jones

August 31, 2019
Production Scripts

One of the most useful parts of SQL Server is the SQL Agent scheduler. Over the years I’ve used this subsystem to automate work and ease the administrative burden of running a SQL Server instance. At times I’ve even used my SQL Agent to fire off business reports and alerts to nontechnical people that might need to take some action based on the data in our system.

One common task that I see SQL Agent used for is to backup a database using some method (Ola’s scripts, SQL Backup, etc.) and then delete the oldest backup. This is a common way of ensuring that you keep xx amount of backups around for your business. Often if the first step (the backup) fails, then the second step (delete backup files) doesn’t run. There are times where the execution choices between the steps aren’t configured or get changed, and you run into the situation that Paul Randall describes in issue #170 of the SQLskills newsletter. The backup step starts to fail, but the deletes keep running until you don’t have any more backups stored.

How does this happen? It’s surprisingly easy because most people don’t really treat their SQL Agent scripts and code like other code. This often isn’t tested well, and rarely tested across time and with unusual conditions. After all, who wants to muck with the time on a production server to be sure that your Agent job works across days? Who wants to force a backup to fail to see how the job handles an issue? Who wants to double check their code when BACKUP is fairly simple syntax and a few quick tests of the delete code works with text files renamed with .bak extensions?

Do you treat your SQL Agent jobs like the production code that they contain? You should. In fact, moving to a more reliable, repeatable, DevOps style environment means that any code in an Agent job needs to be version controlled, it needs to be tested, and it should be a part of some (hopefully, automated) deployment process that ensures that changes to the code are recorded and you are confident of which version of code is on your system.

SQL Agent is a powerful tool, but it’s also one that should be treated like a production system. Downtime and simple errors from careless scripting shouldn’t be tolerated. We should, and can, do better.

Steve Jones

The Voice of the DBA Podcast

Listen to the MP3 Audio ( 3.7MB) podcast or subscribe to the feed at iTunes and Libsyn.

February 12, 2018
Restoring a Copy Only Backup–#SQLNewBlogger
Another post for me that is simple and hopefully serves as an example for people trying to get blogging as #SQLNewBloggers.

There was a question posted recently at SQLServerCentral about whether a copy only backup could be restore with a transaction log backup from a database. I was positive this could, but decided I needed to repro and test for someone as there wasn’t a good BOL reference.

The Tests

Here’s what I did. First, I created a table in a database. I often do this and drop in messages to allow me to track the progress of backups and restores. This post follows my progress.

The Backups

Here’s my basic script:
```
CREATE TABLE logger(msg VARCHAR(200), msgdate DATETIME DEFAULT GETDATE())

INSERT logger (msg) SELECT 'pre full backup'
```
Next, I made a backup and added a message.
```
INSERT logger (msg) SELECT 'pre-log backup 1'
BACKUP LOG nba TO disk = 'nba_1.trn'
INSERT logger (msg) SELECT 'log backup 1 complete'
```
Once this is done, I’m in a state that I expect. A normal full backup, a normal log backup, and some data to help me track where I am.

Now let’s make a copy only backup.
```
INSERT logger (msg) SELECT 'pre copy-only backup '
BACKUP DATABASE nba TO DISK  = 'nba_copy.bak' WITH COPY_ONLY
INSERT logger (msg) SELECT 'copy-only backup complete'
```
This now means I have an open log sequence in the first log backup (post full backup) and a few log records since then. Some of these are inside the copy only backup.

Now let’s add more data and make a new, regular, log backup.
```
INSERT logger (msg) SELECT 'pre-log backup 2'
BACKUP LOG nba TO disk = 'nba_2.trn'
INSERT logger (msg) SELECT 'log backup 2 complete'
```
It’s at this point that I have this sequence:
- Full backup
- Log backup
- Copy-Only Full backup
- Log backup
The Restore

What I want to test is can I restore the Copy-Only backup and a log backup? I think I can, so let’s do that. First, restore from the copy-only backup, using the MOVE option.
```
USE [master]
RESTORE DATABASE [NBA2] 
FROM  DISK = N'D:\SQLServerBackup\MSSQL13.SQL2016\MSSQL\Backup\nba_copy.bak' 
WITH  FILE = 1,  
      MOVE N'NBA' TO N'E:\SQLServer\MSSQL13.SQL2016\MSSQL\Data\NBA2.mdf',  
      MOVE N'NBA_log' TO N'E:\SQLServer\MSSQL13.SQL2016\MSSQL\Data\NBA2_log.ldf',  
      MOVE N'nba_mo_file1' TO N'E:\SQLServer\MSSQL13.SQL2016\MSSQL\Data\NBA2_mo',  
      MOVE N'nba_mo_file2' TO N'E:\SQLServer\MSSQL13.SQL2016\MSSQL\Data\NBA2_mo2'
,  NOUNLOAD,  STATS = 5
, NORECOVERY
```
Tip: Always use NORECOVERY

Now let’s try to restore the log.
```
RESTORE LOG NBA2 FROM DISK = 'nba_2.trn' WITH NORECOVERY

RESTORE DATABASE nba2 WITH RECOVERY
```
This works:

That should prove things. Let’s check the logger table.

That’s what we expect. The final message after log backup2 wasn’t captured in our backup files.

Copy Only Backups

What is a copy only backup? If we check the Copy-Only Backups page, we find that this is a regular backup in and of-itself, but it has the restriction that it cannot be used with differential backups. This also doesn’t change the differential bitmap, so that any differentials that are made ignore this backup and go back to include data changed since the last “normal” full backup.

SQLNewBlogger

Understanding backup and recovery is critical for a data professional. I’d say this is the most important skill, and it’s always worth writing about. Spend a few minutes reviewing scenarios and creating some posts like this to show you understand how the system works.
December 27, 2017
A Backup Change

Backups are a fundamental skill for most DBAs, and hopefully, for most technology professionals. For developers, I’d hope that most of you use some sort of version control, and that you back up your VCS database. I actually had someone ask why we needed to back up the VCS if we had the code on our machines. Certainly the local code on your system, or in your database, provides you some level of redundancy, but all the branches, all the code from other developers, you really want a real backup. For git, this is a simple file level backup.

I recently got a letter from Crashplan, who I’ve been using for a few years as a backup provider. Apparently they are exiting the home backup market, choosing to focus on businesses. I chose them since it was an economical provider, with good ratings, that let me back up multiple machines. I’ve been happy with them, tested a restores of a few files, but never needed the service. Now, I need a new solution. I keep two copies at home, but what about a fire or disaster? I want an offsite backup.

One of the things I’ve wanted with a backup solution is a hands off process. While I’ve managed to use cloud sync software and VCS reports to move most work stuff from one machine to others, there are pictures and other data that I don’t want to lose. I’ve also got computers for my wife and kids that I’d like to have backed up. The Crashplan subscription for 5 computers worked great for me.

It doesn’t seem there are a lot of providers out there for families. Most focus on businesses or the individual, which is fine. Backblaze seems like the next best choice, and at $50/yr/computer, perhaps that’s a fair price. I’ve considered using Amazon Glacier and CloudBerry software, but that feels like I’m giving myself another management job to track. Though, maybe with PoSh available cross platform, I could just build a set of scripts to let each computer notify me if there are issues. I’m still trying to decide what makes sense.

Backup is important, and it’s becoming a more cumbersome job over time. As my family generates more pictures and video, I get more concerned about backup. Especially the cost. The same problems and challenges I face as a DBA, though often with a slightly bigger budget. However, the challenge of balancing a budget with the requirements to meet some RPO is the same.

Steve Jones

September 13, 2017