Monday, November 18, 2013

DBA Rules of Thumb - Part 2 (Automate)

Why should you do it by hand if you can automate DBA processes? Anything you can do probably can be done better by the computer – if it is programmed to do it properly. And once it is automated you save yourself valuable time. And that time can be spent tackling other problems, learning about new features and functionality, or training others.

Furthermore, don’t reinvent the wheel. Someone, somewhere, at some time many have already solved the problem you currently are attempting to solve. Look to the web for sites that allow you to download and share scripts. Or if you have budget money look to purchase DBA tools from ISVs. There are a lot of good tools out there, available from multiple vendors, that can greatly simplify the task of database administration. Automating performance management, change management, backup and recovery, and other tasks can help to reduce the amount of time, effort, and human error involved in managing database systems.

Of course, you can take the automation idea too far. There has been a lot of talk and vendor hype lately about self-managing database systems. For years now, pundits and polls have been asking when automation will make the DBA job obsolete. The correct answer is "never" - or, at least, not any time soon.

There are many reasons why DBAs are not on the fast path to extinction. Self-managing databases systems are indeed a laudable goal, but we are very far away from a “lights-out” DBMS environment. Yes, little-by-little and step-by-step, database maintenance and performance management is being improved, simplified, and automated. But you know what? DBAs will not be automated out of existence in my lifetime – and probably not in your children’s lifetime either.
Many of the self-managing features require using the built-in tools from the DBMS vendor, but many organizations prefer to use heterogeneous solutions that can administer databases from multiple vendors (Oracle, DB2, SQL Server, MySQL, etc.) all from a single console. Most of these tools have had self-managing features for years and yet they did not make the DBA obsolete.

And let’s face it, a lot of the DBMS vendors claims are more hyperbole than fact. Some self-managing features are announced years before they will become generally available in the DBMS. All vendors claims to the contrary, no database today is truly 100% self-contained. Every database needs some amount of DBA management – even when today’s best self-management features are being used.

What about the future? Well, things will get better – and probably more costly. You don’t think the DBMS vendors are building this self-management technology for free, do you? But let’s remove cost from the equation for a moment. What can a self-managing database actually manage?

Most performance management solutions allow you to set performance thresholds. A threshold allows you to set up a trigger that says something like “When x% of a table’s pages contain chained rows or fragmentation, schedule a reorganization.” But these thresholds are only as good as the variables you set and the actions you define to be taken upon tripping the threshold. Some software is bordering on intelligent; that is, it “knows” what to do and when to do it. Furthermore, it may be able to learn from past actions and results. The more intelligence that can be built into a self-managing system, the better the results typically will be. But who among us currently trusts software to work like a grizzled veteran DBA? The management software should be configurable such that it alerts the DBA as to what action it wants to take. The DBA can review the action and give a “thumbs up” or “thumbs down” before the corrective measure is applied. In this way, the software can earn the DBA’s respect and trust. When the DBA trusts the software, he can turn it on so that it self-manages “on the fly” without DBA intervention. But today, in most cases, a DBA is required to set up the thresholds, as well as to ensure their on-going viability.

Of course, not all DBA duties can be self-managed by software. Most self-management claims are made for performance management, but what about change management? The DBMS cannot somehow read the mind of its user and add a new column or index, or change a data type or length. This non-trivial activity requires a skilled DBA to analyze the database structures, develop the modifications, and deploy the proper scripts or tools to implement the change. Of course, software can help simplify the process, but software cannot replace the DBA.

Furthermore, database backup and recovery will need to be guided by the trained eye of a DBA. Perhaps the DBMS can become savvy enough to schedule a backup when a system process occurs that requires it. Maybe the DBMS of the future will automatically schedule a backup when enough data changes. But sometimes backups are made for other reasons: to propagate changes from one system to another, to build test beds, as part of program testing, and so on. A skilled professional is needed to build the proper backup scripts, run them appropriately, and test the backup files for accuracy. And what about recovery? How can a damaged database know it needs to be recovered? Because the database is damaged any self-managed recovery it might attempt is automatically rendered suspect. Here again, we need the wisdom and knowledge of the DBA.

And there are many other DBA duties that cannot be completely automated. Because each company is different, the DBMS must be customized using configuration parameters. Of course, you can opt to use the DBMS “as is,” right out-of-the-box. But a knowledgeable DBA can configure the DBMS so that it runs appropriately for their organization. Problem diagnosis is another tricky subject. Not every problem is readily solvable by developers using just the Messages and Codes manual and a help desk. What happens with particularly thorny problems if the DBA is not around to help?

Of course, the pure, heads-down systems DBA may (no, let's say should) become a thing of the past. Instead, the modern DBA will need to understand multiple DBMS products, not just one. DBAs furthermore must have knowledge of the business impact of each database under their care (more details here). And DBAs will need better knowledge of logical database design and data modeling – because it will advance their understanding of the meaning of the data in their databases.

Finally, keep in mind that we didn't automate people out of existence when we automated HR or finance. Finance and HR professionals are doing their jobs more efficiently and effectively, and they have the ability to deliver a better product in their field. That's the goal of automation. So, as we automate portions of the DBA’s job, we'll have more efficient data professionals managing data more proficiently.

This blog entry started out as a call to automate, but I guess it kinda veered off into an extended dialogue on what can, and cannot, be accomplished with automation. I guess the bottom line is this... Automation is key to successful, smooth-running databases and applications... but don't get too carried away by the concept.

I hope you found the ideas here to be useful... and feel free to add your own thoughts and comments below! 

No comments: