Relational database management systems (RDBMSs) often persist mission-critical data which is updated by many applications and potentially thousands if not millions of end users. Furthermore, they implement important functionality in the form of database methods (stored procedures, stored functions, and/or triggers) and database objects (e.g. Java or C# instances). The best way to ensure the continuing quality of these assets, at least from a technical point of view, you should have a full regression test suite which you can run on a regular basis.
In this article I argue for a fully automated, continuous regression testing based approach to database testing. Just as agile software developers take this approach to their application code, we should also do the same for our databases.
Table of Contents
Why test an RDBMS?
What should we test?
When should we test?
How should we test?
Database sandboxes
Writing database tests
Setting up database tests
Database testing tools
Who should test?
Introducing database testing into your organization
Database testing and data inspection
Best practices
1. Why Test an RDBMS?
There are several reasons why you need to develop a comprehensive testing strategy for your RDBMS:
Data is an important corporate asset. Doesn't it make sense to invest the effort required to validate the quality of data via effective testing? My July 2006 survey into the current state of data management indicates that 95.7% of respondents believe that data is a corporate asset. Yet of them only 40.3% had a database test suite in place to validate the data and of those without a test suite only 31.6% had even discussed the concept.
Mission-critical business functionality is implemented in RDBMSs. In the survey, 63.7% of respondents indicated that their organizations did this, but of those only 46% had regression tests in place to validate the logic. Shouldn't we be doing better?
Current approaches aren't sufficient. The current state of the art in many organizations is for data professionals to control changes to the database schemas, for developers to visually inspect the database during construction, and to perform some form of formal testing during the test phase at the end of the lifecycle. Unfortunately, none of these approaches prove effective. Application developers will often go around their organization's data management group because they find them too difficult to work with, too slow in the way they work, or sometimes they don't even know they should be working together. The end result is that the teams don't follow the desired data quality procedures and as a result quality suffers. Although visual inspection of query results is a good start it is little more than a debugging technique in practice that will help you to find problems but not prevent them. Testing late in the lifecycle is better than nothing, but as Barry Boehm noted in the early 80s it's incredibly expensive to fix any defects you find at that point.
Testing provides the concrete feedback required to identify defects. How do you know how good the quality of your source data actually is without an effective test suite which you can run whenever you need to?
Support for evolutionary development. Many evolutionary development techniques, in particular database refactoring, are predicated upon the idea that it must be possible to determine if something in the database has been broken when a change has been made. The easiest way to do that is to simply run your regression test suite.
Uncomfortable Question:
Isn't it time that we stopped talking about data quality and actually started doing something about it?
Here's a few interesting questions to ask someone who isn't convinced that you need to test the DB:
If you're implementing code in the DB in the form of stored procedures, triggers, ... shouldn't you test that code to the same level that you test your app code?
Think of all the data quality problems you've run into over the years. Wouldn't it have been nice if someone had originally tested and discovered those problems before you did?
Wouldn't it be nice to have a test suite to run so that you could determine how (and if) the DB actually works?
I think that one of the reasons that we don't hear much about database testing is because it is a relatively new idea within the data community. Many traditional data professionals seem to think that testing is something that other people do, particularly test/quality assurance professionals, do. This reflects a penchant for over-specialization and a serial approach towards development by traditionalists, two ideas which have also been shown to be questionable organizational approaches at best.
2. What Should We Test?
Figure 1 indicates what you should consider testing when it comes to relational databases. The diagram is drawn from the point of view of a single database, the dashed lines indicate threat boundaries, indicating that you need to consider threats both within the database (clear box testing) and at the interface to the database (black box testing). Table 1 lists the issues which you should consider testing for both internally within the database and at the interface to it. For details, read the article What To Test in an RDBMS.
Figure 1. What to test.
Table 1. What to test in an RDBMS.
Black-Box Testing at the Interface White/Clear-Box Testing Internally Within the Database
O/R mappings (including the meta data)
Incoming data values
Outgoing data values (from queries, stored functions, views ...)
Scaffolding code (e.g. triggers or updateable views) which support refactorings
Typical unit tests for your stored procedures, functions, and triggers
Existence tests for database schema elements (tables, procedures, ...)
View definitions
Referential integrity (RI) rules
Default values for a column
Data invariants for a single column
Data invariants involving several columns
3. When Should We Test?
Agile software developers take a test-first approach to development where they write a test before you write just enough production code to fulfill that test. The steps of test first development (TFD) are overviewed in the UML activity diagram of Figure 2. The first step is to quickly add a test, basically just enough code to fail. Next you run your tests, often the complete test suite although for sake of speed you may decide to run only a subset, to ensure that the new test does in fact fail. You then update your functional code to make it pass the new tests. The fourth step is to run your tests again. If they fail you need to update your functional code and retest. Once the tests pass the next step is to start over.
Figure 2. The Process of Test First Development (TFD).
Test-driven development (TDD) is an evolutionary approach to development which combines test-first development and refactoring. When an agile software developer goes to implement a new feature, the first question they ask themselves is "Is this the best design possible which enables me to add this feature?" If the answer is yes, then they do the work to add the feature. If the answer is no then they refactor the design to make it the best possible then they continue with a TFD approach. This strategy is applicable to developing both your application code and your database schema, two things that you would work on in parallel.
When you first start following a TDD approach to development you quickly discover that to make it successful you need to automate as much of the process as possible? Do you really want to manually run the same build script(s) and the same testing script(s) over and over again? Of course not. So, agile developers have created OSS tools such as ANT, Maven, and Cruise Control (to name a few) which enable them to automate these tasks. More importantly, it enables them to automate their database testing script into the build procedure itself.
Agile developers realize that testing is so important to their success that it is something they do every day, not just at the end of the lifecycle. They test as often and early as possible, and better yet they test first. As you can see with the agile system development lifecycle (SDLC) of Figure 3 testing is in fact something that occurs during the development and release cycles, not just during release. Furthermore, many agile software developers realize that you can test more than just your code, you can in fact validate every work product created on a software development project if you choose to. This philosophy is exemplified by the Full Lifecycle Object-Oriented Testing (FLOOT) Methodology.
Figure 3. The Agile Lifecycle.
4. How to Test
Although you want to keep your database testing efforts as simple as possible, at first you will discover that you have a fair bit of both learning and set up to do. In this section I discuss the need for various database sandboxes in which people will test: in short, if you want to do database testing then you're going to need test databases (sandboxes) to work in. I then overview how to write a database test and more importantly describe setup strategies for database tests. Finally, I overview several database testing tools which you may want to consider.
4.1 Database Sandboxes
A common best practice on agile teams is to ensure that developers have their own "sandboxes" to work in. A sandbox is basically a technical environment whose scope is well defined and respected. Figure 4 depicts the various types of sandboxes which your team may choose to work in. In each sandbox you'll have a copy of the database. In the development sandbox you'll experiment, implement new functionality, and refactor existing functionality, validate your changes through testing, and then eventually you'll promote your work once you're happy with it to the project integration sandbox. In this sandbox you will rebuild your system and then run all the tests to ensure you haven't broken anything (if so, then back to the development sandbox). Occasionally, at least once an iteration/cycle, you'll deploy your work to the level (demo and pre-production testing), and rerun your test suite (including database tests) each time that you do so to ensure that your changes integrate with the changes made by other developers. Every so often (perhaps once every six to twelve months) into production. The primary advantage of sandboxes are that they help to reduce the risk of technical errors adversely affecting a larger group of people than is absolutely necessary at the time.
Figure 4. Sandboxes.
4.2 Writing Database Tests
There's no magic when it comes to writing a database test, you write them just like you would any other type of test. Database tests are typically a three-step process:
Setup the test. You need to put your database into a known state before running tests against it. There are several strategies for doing so.
Run the test. Using a database regression testing tool, run your database tests just like you would run your application tests.
Check the results. You'll need to be able to do "table dumps" to obtain the current values in the database so that you can compare them against the results which you expected.
The article What To Test in an RDBMS goes into greater detail.
4.3 Setting up Database Tests
To successfully your database you must first know the exact state of the database, and the best way to do that is to simply put the database in a known state before running your test suite. There are two common strategies for doing this:
Fresh start. A common practice is to rebuild the database, including both creation of the schema as well as loading of initial test data, for every major test run (e.g. testing that you do in your project integration or pre-production test sandboxes).
Data reinitialization. For testing in developer sandboxes, something that you should do every time you rebuild the system, you may want to forgo dropping and rebuilding the database in favor of simply reinitializing the source data. You can do this either by erasing all existing data and then inserting the initial data vales back into the database, or you can simple run updates to reset the data values. The first approach is less risky and may even be faster for large amounts of data.
An important part of writing database tests is the creation of test data. You have several strategies for doing so:
Have source test data. You can maintain an external definition of the test data, perhaps in flat files, XML files, or a secondary set of tables. This data would be loaded in from the external source as needed.
Test data creation scripts. You develop and maintain scripts, perhaps using data manipulation language (DML) SQL code or simply application source code (e.g. Java or C#), which does the necessary deletions, insertions, and/or updates required to create the test data.
Self-contained test cases. Each individual test case puts the database into a known state required for the test.
These approaches to creating test data can be used alone or in combination. A significant advantage of writing creation scripts and self-contained test cases is that it is much more likely that the developers of that code will place it under configuration management (CM) control. Although it is possible to put test data itself under CM control, worst case you generate an export file that you check in, this isn’t a common practice and therefore may not occur as frequently as required. Choose an approach that reflects the culture of your organization.
Where does test data come from? For unit testing, I prefer to create sample data with known values. This way I can predict the actual results for the tests that I do write and I know I have the appropriate data values for those tests. For other forms of testing -- particularly load/stress, system integration, and function testing, I will use live data so as to better simulate real-world conditions.
Beware Coupling:
One danger with database regression testing, and with regression testing in general, is coupling between tests. If you put the database into a known state, then run several tests against that known state before resetting it, then those tests are potentially coupled to one another. Coupling between tests occurs when one test counts on another one to successfully run so as to put the database into a known state for it. Self-contained test cases do not suffer from this problem, although may be potentially slower as a result due to the need for additional initialization steps.
4.4 What Testing Tools Are Available?
I believe that there are several critical features which you need to successfully test RDBMSs. First, as Figure 1 implies you need two categories of database testing tools, one for interface tests and one for internal database tests. Second, these testing tools should support the language that you're developing in. For example, for internal database testing if you're a Microsoft SQL Server developer, your T-SQL procedures should likely be tested using some form of T-SQL framework. Similarly, Oracle DBAs should have a PL-SQL-based unit testing framework. Third, you need tools which help you to put your database into a known state, which implies the need not only for test data generation but also for managing that data (like other critical development assets, test data should be under configuration management control).
To make a long story short, although we're starting to see a glimmer of hope when it comes to database testing tools, as you can see in Table 2, but we still have a long way to go. Luckily there are some good tools being developed by the open source software (OSS) community and there are some commercial tools available as well. Having said that, IMHO there is still significant opportunity for tool vendors to improve their database testing offerings.
Table 2. Some database testing tools.
Category Description Examples
Unit testing tools Tools which enable you to regression test your database. DBFit
DBUnit
NDbUnit
OUnit for Oracle (being replaced soon by Qute)
SQLUnit
TSQLUnit (for testing T-SQL in MS SQL Server)
Visual Studio Team Edition for Database Professionals includes testing capabilities
XTUnit
Testing tools for load testing Tools simulate high usage loads on your database, enabling you to determine whether your system's architecture will stand up to your true production needs. Empirix
Mercury Interactive
RadView
Rational Suite Test Studio
Web Performance
Test Data Generator Developers need test data against which to validate their systems. Test data generators can be particularly useful when you need large amounts of data, perhaps for stress and load testing. Data Factory
Datatect
DTM Data Generator
Turbo Data
5. Who Should Test?
During development cycles, the primary people responsible for doing database testing are application developers and agile DBAs. They will typically pair together, and because they are hopefully taking a TDD-approach to development the implication is that they'll be doing database unit testing on a continuous basis. During the release cycle your testers, if you have any, will be responsible for the final system testing efforts and therefore they will also be doing database testing.
The role of your data management (DM) group, or IT management if your organization has no DM group, should be to support your database testing efforts. They should promote the concept that database testing is important, should help people get the requisite training that they require, and should help obtain database testing tools for your organization. As you have seen, database testing is something that is done continuously by the people on development teams, it isn't something that is done by another group (except of course for system testing efforts). In short, the DM group needs to support database testing efforts and then get out of the way of the people who are actually doing the work.
6. Introducing Database Regression Testing into Your Organization
Database testing is new to many people, and as a result you are likely to face several challenges:
Insufficient testing skills. This problem can be overcome through training, through pairing with someone with good testing skills (pairing a DBA without testing skills and a tester without DBA skills still works), or simply through trial and error. The important thing is that you recognize that you need to pick up these skills.
Insufficient unit tests for existing databases. Few organizations have yet to adopt the practice of database testing, so it is likely that you will not have a sufficient test suite for your existing database(s). Although this is unfortunate, there is no better time than the present to start writing your test suite.
Insufficient database testing tools. As I said earlier, we still have a way to go with respect to tools.
Reticent DM groups. My experience is that some data management (DM) groups may see the introduction of database regression testing, and agile techniques such as test-first development (TFD) and refactoring, as a threat. Or, as my July 2006 "state of data management" survey shows, a large percentage of organizations are not only not doing any database testing at all they haven't even discussed it. For many in the data management community the idea of doing database testing is rather new and it's simply going to take a while for them to think it through. I'm not so sure that you should wait to do such obvious process improvement.
In general, I highly suggest that you read my article Adopting Evolutionary/Agile Database Techniques and consider buying the book Fearless Change which describes a pattern language for successfully implementing change within organizations.
7. Database Testing and Data Inspection
A common quality technique s to use data inspection tools to examine existing data within a database. You might use something as simple as a SQL-based query tool such as DB Inspect to select a subset of the data within a database to visually inspect the results. For example, you may choose to view the unique values in a column to determine what values are stored in it, or compare the row count of a table with the count of the resulting rows from joining the table with another one. If the two counts are the same then you don't have an RI problem across the join.
As Richard Dallaway points out, the problem with data inspection is that it is often done manually and on an irregular basis. When you make changes later, sometimes months or years later, you need to redo your inspection efforts. This is costly, time consuming, and error prone.
Data inspection is more of a debugging technique than it is a testing technique. It is clearly an important technique, but it's not something that will greatly contribute to your efforts to ensure data quality within your organization.
8. Best Practices
I'd like to conclude this article by sharing a few database testing "best practices" with you:
Use an in-memory database for regression testing. You can dramatically speed up your database tests by running them, or at least portions of them, against an in-memory database such as HSQLDB. The challenge with this approach is that because database methods are implemented differently across database vendors that any method tests will still need to run against the actual database server.
Start fresh each major test run. To ensure a clean database, a common strategy is that at the beginning of each test run you drop the database, then rebuild it from scratch taking into account all database refactorings and transformations to that point, then reload the test data, and then run your tests. Of course, you wouldn't do this to your production database. ;-)
Take a continuous approach to regression testing. I can't say this enough, a TDD approach to development is an incredibly effective way to work.
Train people in testing. Many developers and DBAs have not been trained in testing skills, and they almost certainly haven't been trained in database testing skills. Invest in your people, and give them the training and education they need to do their jobs.
Pair with novices with people that have database testing experience. One of the easiest ways to gain database testing skills is to pair program with someone who already has them.
No comments:
Post a Comment