Test-Driven Development (TDD) is a software development practise that has been around for a few years. Wikipedia describes it as:

Test-driven development (TDD) is a software development process that relies on the repetition of a very short development cycle: first the developer writes a failing automated test case that defines a desired improvement or new function, then produces code to pass that test and finally refactors the new code to acceptable standards. Kent Beck, who is credited with having developed or 'rediscovered' the technique, stated in 2003 that TDD encourages simple designs and inspires confidence.

http://en.wikipedia.org/wiki/Test-driven_development

Since 2003 TDD practises have seen refinements such as Behavior-Driven Development and Uncle Bob's Three Rules of TDD, all the while TDD has pretty much become an accepted way of developing quality software. Accepted that is everywhere outside of the database development arena and that is the arena in which I spend my working life. TDD simply has not, in my opinion, caught on with database developers like it has our appdev brethren and I was reminded of this yesterday when Atul Thakor asked on Twitter:

anyone done TDD for database development and would they recommend it?

https://twitter.com/#!/atulthakor/status/161886007929733120

To which my answer was an emphatic:

(1) yes & (2) absolutely, yes

https://twitter.com/#!/jamiet/status/161894215217987585

I'll use this blog post to expand on that outside of 140 characters.

In October 2010 I undertook a mini-project for the client I was working for at the time (a bank) where a colleague and I were tasked with building the database portion of a system that would support reconciliation of our ETL processes. It was a nice piece of work in that it was small, well-scoped, time-bound, greenfield, did not have any external dependancies and had a technically savvy product owner. We sat down at the start and decided that this was an ideal opportunity to trial TDD as a method of developing a database; I would write the failing tests and my colleague would make the tests pass. We came up with some guiding principles and, although we didn't know it at the time, they were pretty close to Uncle Bob's three rules.

I used Visual Studio 2010's database unit testing framework¹ to write my tests and have them run as part of our Continuous Integration (CI) build (see Setting up database unit testing as part of a Continuous Integration build process). I would write the tests, check-in, the CI build would fail and my colleague would "get latest" in order to see what code he had to write to stop the build from failing. To cut a long story short the use of TDD was considered to be a great success; we shipped a working system on time/on budget and moreover, even though I didn't write a scrap of code that went into production I have never had more confidence that a system I was involved in building worked as intended. That's quite a statement. My confidence stemmed from the fact that as the test author I was ultimately responsible for ensuring that the system did what it was supposed to; I could qualify my confidence by pointing at our CI build and highlighting the number of tests that were passing and how that number had steadily increased as the project progressed.

By the time the project had finished the database consisted of (if memory serves me correctly) 6 tables and about 10 stored procedures or functions (so yes, very small). To test that we had roughly 70 tests that were getting run up to 20 times a day. The project had taken about two months from start-up to final delivery - you can make your own opinions as to whether you consider that prompt or tardy but our product owner was happy and that's pretty much all that counted as far as I was concerned.

Since that project I have moved onto other clients and at each one I have always extolled the use of database unit testing; we haven't always practised TDD but at each one we have been writing database unit tests and in the future I suspect that a client's willingness (or lack thereof) to use database unit testing will be a major factor in influencing whether we end up working together or not.

Are you a database developer doing database unit testing or perhaps even TDD? Let me know in the comments, I'd love to hear about others' experiences.

@jamiet

¹Yes, that linked-to article from 7 years ago is the best one I could find to describe what Visual Studio's Database Unit testing Framework actually is - sort it out Microsoft!

UPDATE: I have just remembered that Jamie Laflen has written an excellent whitepaper entitled Apply Test-Driven Development to your Database Projects that goes into much more detail about how to achieve database TDD using Visual Studio than I have here. Well worth a read.

My ex-colleague Paul Mcmillan pointed me at a thread on Stack Overflow that demonstrated a neat T-SQL trick to get the maximum value from a collection of columns in a row. Paul had never seen it before and neither had I so I figure one or two of you out there might learn something from it too.

In short you can use the VALUES clause to effectively union the values into a dataset and get the MAX from that dataset. Better demonstrated with code:

DECLARE @t TABLE(a INT,b INT,c INT);
INSERT @t VALUES(1,2,3),(9,8,7),(4,6,5);
SELECT *
,      (   SELECT  MAX(val)
           FROM    (VALUES (a)
                       ,   (b)
                       ,   (c)
                   ) AS value(val)
       ) AS MaxVal
FROM @t;

I'm sure many of you knew this already but if you didn't, well, you too have learnt something today. See more uses for the VALUES clause at Interesting enhancements to the VALUES Clause in SQL Server 2008

@jamiet

P.S. Oh, this only works in SQL Server 2008 and beyond.

Over the past eighteen months I have worked on four separate projects for customers that wanted to make use of Visual Studio 2010 Database projects to manage their database schema.* All the while I have been trying to take lots of notes in order that I can write the blog post that you are reading right now – a compendium of experiences and tips from those eighteen months. I should note that this blog post should not necessarily be taken as a recommendation to actually use database projects in Visual Studio 2010 – it is intended to be useful for those of you that have already made the decision to use them; having said that, I do make recommendations as to actions I think you should take if you have made that decision.

First let’s be clear what we’re talking about here. Visual Studio Database Projects have been known by a few different names down the years, some of which you may be familiar with. If you have ever heard the terms datadude, DBPro, teamdata, TSData or Visual Studio Team System for Database Professionals then just know that all of these terms refer to the same thing, namely the project type highlighted below when starting a new project in Visual Studio:

From here onwards I am going to refer to Visual Studio Database projects and all the features therein simply as datadude because that’s a popular colloquial name (and is also much quicker to type). Know also that at the time of writing the features that I am talking about here are currently undergoing some changes ahead of the next release of SQL Server (i.e. SQL Server 2012) in which these features are mooted to be delivered under a new moniker - SQL Server Developer Tools (SSDT).

OK, with all those preliminaries out of the way let’s dig in.

Continuous Integration

Continuous Integration (CI) is a development practise that has existed for many years but in my experience has not been wholly embraced by the database community. The idea behind CI for databases is that every time a developer checks-in a piece of code be it a stored procedure, a table definition or whatever, the entire database project is built and then deployed to a database instance. Microsoft provide a useful article An Overview of Database Build and Deployment that goes some way to explaining how to setup your CI deployment.

CI is one of the fundamental tenets that underpins a lot of the things I talk about later in this blog post and hence gives rise to my first recommendation when using datadude:

Recommendation #1: Use Source Control and implement a Continuous Integration deployment

In an earlier draft of this blog post I outlined in detail the CI configuration from one of the aforementioned projects. Its not suitable for inclusion at this point in the current draft but I still think there is some useful information to be gleaned so I have included it below in “Appendix – An example CI configuration”.

Composite Projects

Each of the four aforementioned projects were brownfield projects meaning that each already encompassed some established, deployed, databases and they wanted to bring those databases under the control of datadude. Each project had thousands of objects across multiple databases and in this situation it is very likely that some of the stored procedures, views or functions will refer to objects in one of the other databases. The way to resolve those references is to use database references however once you have applied all of your database references it is still very possible that you will run into a situation where code in database A refers to an object in database B while at the same time database B refers to an object in database A. This is depicted in the following figure:

Here we have a view [DB1]..[View2] that selects data from [DB2]..[Table1] and a view [DB2]..[View1] that selects data from [DB1]..[Table1]. Datadude does not allow a database reference from [DB2] to [DB1] if there is already a database reference from [DB2] to [DB1] and hence will return an error akin to:

• SQL03006: View: [dbo].[View1] has an unresolved reference to object [DB1].[dbo].[Table1].

We have the proverbial chicken-and-egg problem, [DB1] can’t be created before [DB2] and vice versa. This problem is solved by using Composite Projects (not to be confused with Partial Projects) which allow you to split objects that are intended to be in the same database over multiple datadude projects. I could go over how you set one of these things up but there’s really no need because there is a rather excellent walkthrough on MSDN at Walkthrough: Partition a Database Project by Using Composite Projects; the reason for me mentioning it here is to make you aware that composite projects exist and of the problem that they solve. If you are introducing datadude into a brownfield project then it is highly likely that you are going to require composite projects so learn them and learn them good.

One important last note about composite projects is to answer the question “How does the tool know that the multiple projects refer to the same database?” The answer is given at the walkthrough that I linked to above; namely, it says:

“Do not specify server variables and values or database variables and values when you define references in a composite project. Because no variables are defined, the referenced project is assumed to share the target server and database of the current project.”

So now you know! To put it another way, if you reference one project from another and don’t tell datadude that the two projects refer to different databases then it assumes they refer to the same database.

Code Analysis

Datadude provides the ability to analyse your code projects for code in stored procedures and functions that it considers to be inferior and highlight it – this feature is called Code Analysis. Note that Code Analysis will not highlight code that is syntactically incorrect (datadude does that already, which may well be considered its core feature), it highlights code that is syntactically correct but may be considered defective when executed. Specifically Code Analysis will highlight the following perceived code defects (click through on the links for explanations of why these are considered code defects):

• Use of SELECT *
• Use of @@IDENTITY rather than SCOPE_IDENTITY()
• NCHAR & NVARCHAR fields whose maximum length is 1 or 2
• Deprecated join syntax
• An output parameter that might not be assigned to
• Casts that could result in data loss (in my experience this is the one that shows up most often – particularly in inherited code)
• Use of special characters in object names
• Use of reserved words
• Use of sp_ prefix for stored procedures
• Use of unindexed columns in IN predicates
• Use of patterns beginning with a wildcard in a LIKE predicate
• Code that would not use an existing index that might actually be beneficial
• Use of ISNULL() on nullable columns
• Deterministic function calls in a WHERE predicate

In my opinion the best aspect of Code Analysis is that it can be run as part of your Continuous Integration deployment meaning that if anyone checks in some deficient code, BOOM, your CI deployment fails and the developer is left red-faced. Nothing else has increased the code quality on my projects quite like running Code Analysis as part of a CI deployment.

Hopefully I have convinced you that turning on Code Analysis is a good idea. If you agree then head to the project properties and check the box labelled Enable Code Analysis on Build. I also recommend checking the Treat warnings as errors boxes otherwise you’ll find that undisciplined developers will simply ignore the warnings.

N.B. Incidentally if you have time I highly recommend that you go and read the blog post I linked to there – Discipline Makes Strong Developers by Jeff Atwood. I’ve read many thousands of blog posts in my time and that is the one that has influenced me more than any other.

Turning on Code Analysis on a greenfield project is a no-brainer. On a brownfield project its not quite so easy – on a recent engagement I moved a single database into datadude and turned on Code Analysis which immediately found over two thousand perceived code defects. I generally abhor the use of that famous maxim if it aint broke, don’t fix it in our industry but on occasions like this you may be well advised to heed that advice and leave well alone for fear of breaking code that does what it is supposed to (no matter how inefficiently it does it). Instead you do have the option to suppress Code Analysis warnings/errors:

I advise using Code Analysis suppression sparingly. Recently I discovered that one of the developers on my team had decided it was OK to simply suppress every error that was thrown by Code Analysis without first investigating the cause. I was not amused!

Recommendation #2: Turn on Code Analysis

Realising the value of idempotency

An operation is considered idempotent if it produces the same result no matter how many times that operation is applied; for example, multiplication by a factor of one is an idempotent operation – no matter how many times you multiple a number by one the result will always be the same.

Idempotency is a vital facet of database deployment using datadude. Datadude tries to ensure that no matter how many times you deploy the same project the state of your database should be the same after each deployment. The implication here is that during a deployment datadude will examine the target database to see what changes (if any) need to be made rather than simply attempting to create lots of objects; if all the objects already exist nothing will be done. In my opinion this is the single biggest benefit of using datadude – you don’t have to determine what needs to be done to change your database schema to the desired state, datadude does it for you.

If I have convinced you about the value of idempotency within datadude then you should also realise that the same rigour should be applied to data as well. Datadude provides Post-Deployment scripts that allow you to deploy data to your schema however there is no inbuilt magic here – datadude will simply go and run those scripts as-is, it will not try and comprehend the contents of those scripts. What this means is that you, the developer, are responsible for making your Post-Deployment scripts idempotent and the easiest way to do that is to employ the T-SQL MERGE statement.

T-SQL’s INSERT is not sufficient as it will work once and thereafter fail as it will be attempting to insert already inserted data; this gives rise to my third recommendation:

Recommendation #3: When running your deployment in a test environment, run it more than once.

No-brainer Recommendations

I consider Code Analysis and Idempotency to be so important that I called them out as dedicated headlines. In this section I’ll outline some additional simple measures that you can undertake and which will, if employed correctly, have a profound effect on the success of your datadude projects.

Putting a build number into the DB

I find it is very useful to maintain a log of deployments that have been made to a database and my chosen method is to use a Post-Deployment script to insert a value into some table. Here’s the definition of the table I use for this:

CREATE TABLE [dbo].[DeployLog]
(
   [BuildId]           NVARCHAR(50)
,  [DeployDatetime]    SMALLDATETIME
,  CONSTRAINT  PK_dboDeployLog PRIMARY KEY ([DeployDatetime])
)

In my Post-Deployment script I will use:

INSERT [dbo].[DeployLog]([BuildId],[DeployDatetime])
VALUES ('$(BuildId)',GETDATE());

to insert a row into that table during every deployment. $(BuildId) is a variable defined in the .sqlcmdvars file of my project:

Here is what we see inside that file:

The $(BuildId) variable has been defined with a default value of UNKNOWN and hence subsequent deployments from Visual Studio will result in the following:

On first glance that might not seem particularly useful however it comes into its own if you are doing CI deployments (see recommendation #1) because each build in a CI environment will result in a new build identifier. The following command-line call to vsdbcmd.exe is how deployments are generally done using datadude, note the presence of the /p:BuildId switch:

..\Tools\VSDBCMD\vsdbcmd.exe /Action:Deploy /ConnectionString:"Data Source=.;Integrated Security=True;Pooling=False" /p:BuildId="some-value" /DeployToDatabase:+ /ManifestFile:.\FinanceDB\sql\release\FinanceDB.deploymanifest

Your CI tool should be able to replace “some-value” with an identifier for the current build (that’s outside the scope of this blog post but any CI tool worth its salt will be able to do this) – when the deployment executes that value will then make its way into your [dbo].[DeployLog] table and you will have a self-maintaining history of all the deployments (datetime & build identifier) that have been made to your database.

Recommendation #4: Maintain an automated history of your deployments

Use Schema View

It is natural to navigate through all of the objects in your database project using Solution Explorer however datadude provides a better mechanism for doing just that – the Schema View window.

Schema View provides a logical view of all the objects defined in your database project regardless of which file they may be defined in. That is very useful for many reasons, not least because it makes it easy to locate whichever object you are after – that’s advantageous if multiple objects are defined in the same file. Moreover if some files have property BuildAction=”Not In Build” (see later) they won’t show up in Schema View (this is a good thing by the way). Schema View is also the place that operations such as refactoring and dependency analysis are launched from.

Some people think that it is important that the name of each file in a datadude project should accurately reflect the object defined within. I disagree; object renames mean that maintaining the filenames becomes laborious and having the Schema View means you never have to use the filenames to navigate your project anyway.

One final reason to use Schema View is the External Elements button:

Toggling this button on means that objects in referenced projects show up in the project that they are referenced from (this is particularly useful if you are using Composite Projects). Note in the following screenshot how the object [dbo].[t1] in project Database2 appears in the [dbo] schema of Database3 – that’s because Database3 has a reference to Database2.

For those reasons my fifth recommendation is:

Recommendation #5: Use Schema View in preference to Solution Explorer

You will still need Solution Explorer to navigate files that do not contain database objects (e.g. Post-Deployment scripts) but ordinarily you should spend most of your time interacting with Schema View.

Make liberal use of PRINT statements in Pre/Post-Deployment Scripts

When you deploy a datadude project datadude will take care of telling you what it is up to. For example, the following screenshot shows the output from deploying the already discussed [dbo].[DeplogLog]:

Of course it only does this for objects that it knows about and that doesn’t include anything in your Pre or Post \deployment scripts so you need to take responsibility for outputting pertinent information from those scripts. Here I have amended the script that inserts into [dbo].[DeployLog]:

SET NOCOUNT ON;
INSERT [dbo].[DeployLog]([BuildId],[DeployDatetime])
VALUES ('$(BuildId)',GETDATE());
PRINT CAST(@@ROWCOUNT as NVARCHAR(5)) + N'rows inserted into [dbo].[DeployLog], BuildId=$(BuildId)';

This gives us much more useful output:

Adding PRINT statements to your Pre & Post Deployment scripts is so easy it really is a no-brainer and hence my next recommendation is:

Recommendation #6: Any action in a Pre or Post-Deployment Script should use PRINT to state what has been done

Output variable values in your Pre-Deployment script

This is in the same vein as the previous bullet-point – output as much information as is possible. In this case we’re talking about outputting the values of all variables that are stored in the .sqlcmdvars file; first, a reminder of what’s in that file:

Here is the contents of my amended Pre-Deployment Script:

PRINT 'DefaultDataPath=$(DefaultDataPath)';
PRINT 'DatabaseName=$(DatabaseName)';
PRINT 'DefaultLogPath=$(DefaultLogPath)';
PRINT 'BuildId=$(BuildId)';

And the resultant output:

This is the sort of simple amendment that will pay off in spades later in your project (especially if you are supplying many values from the command-line) and again, its so easy to do it there really is no reason not to. Just remember to update your Pre-Deployment script whenever you add new variables to .sqlcmdvars.

Recommendation #7: Output the value of all variables in your Pre-Deployment script

One Object Per File

Datadude doesn’t restrict what can go in a file, for example the following file, “t.table.sql”, defines three objects; a table, a primary key and a view:

Even though they’re all defined in the same file they show up in Schema View separately (one of the aforementioned benefits of using Schema View):

That said, just because you can doesn’t mean that you should. I prefer to go for one object per file for the simple reason that its easier to track the history of an object via Source Control. Moreover, if an object is no longer required then it is a simple change to just remove the file containing that object from the build (see “Don’t delete anything from your project” later) as opposed to editing a file to remove all traces of an object.

Recommendation #8: Each database object should be defined in a dedicated file

Time your Pre and Post Deployment Scripts

Its always useful to know where time is spent when doing deployments, in my experience the majority of time spent is in the Post-Deployment script (your mileage may vary of course). An easy win is to output the time taken to run your Pre and Post Deployment scripts. Adapt your Pre-Deployment script so that it looks something like this:

DECLARE @vPreDeploymentStartTime DATETIME = GETDATE();
PRINT '****************Begin Pre-Deployment script at ' +CONVERT(VARCHAR(30),GETDATE(),120) + '***********************';

/*Call other scripts from here using SQLCMD's :r syntax
Example:      :r .\myfile.sql                          
*/

PRINT 'Pre-Deployment duration = ' + CONVERT(VARCHAR(5),DATEDIFF(ss,@vPreDeploymentStartTime,GETDATE())) + ' seconds';
PRINT '****************End Pre-Deployment script at ' +CONVERT(VARCHAR(30),GETDATE(),120) + '***********************';

then do similar for your Post-Deployment script. When you deploy your output will include the following:

Note the lines:

    ****************Begin Pre-Deployment script at 2011-12-31 20:00:34***********************
    Pre-Deployment duration = 0 seconds
    ****************End Pre-Deployment script at 2011-12-31 20:00:34***********************
   

    ****************Begin Post-Deployment script at 2011-12-31 20:00:34***********************
    Post-Deployment duration = 0 seconds
    ****************End Post-Deployment script at 2011-12-31 20:00:34***********************

In this particular case its not all that useful to know that the deployment took 0 seconds but if and when your deployments snowball to many minutes it will be useful to know how long your scripts are taking at which point you can investigate further by timing each individual step in your Pre and Post Deployment scripts.

Recommendation #9: Time your deployments and output the timings

Use sqlcmdvars and the command-line as much as possible

Hardcoding any value into a piece of code is a fraught practise; you should assume that values previously thought to be constant may not be so in the future. You can protect yourself from future changes by storing all literal values as variables in your .sqlcmdvars file. Sure, you can supply default values for those variables but you have the added advantage that they can be overridden from the command-line when deploying using vsdbcmd.exe. Moreover, if you have values that are hardcoded in multiple places in your code then specifying those values in .sqlcmdvars ensures that your code adheres to the principle of DRY. Lastly, if values are stored in the .sqlcmdvars file then you can output them at deploy time (see recommendation #7).

Recommendation #10: All literal values should be stored in your .sqlcmdvars file

Every developer gets their own development database

In most SQL Server development shops that I’ve been on all developers work against a single centralised development database. To me this is an antiquated way of working because its possible that work one person is doing can conflict with that of someone else, I find it much better for every developer to work in isolation and then use the CI deployment to check that one’s code is not in conflict with anyone else’s. Datadude supports (nay encourages) this way of working with the notion of an Isolated Development Environment:

Every developer should configure their isolated development environment which, typically, would be their local instance. And so to my next recommendation:

Recommendation #11: Every developer should use the Isolated Dev Environment settings in order to author their code

Incidentally, if every developer has their own development database and you are following my earlier recommendation to use a [DeployLog] table then you can track how often a developer is bothering to deploy and test their code. On a recent project we used this evidence in a (ahem) discussion with a developer who tried to convince us that he was testing his code sufficiently even though he was repeatedly causing the CI deployment to fail.

Don’t delete anything from your project

When projects are no longer required in your database then intuitively it makes sense to remove the file containing that object from the datadude project, I would however like to suggest a different approach. Rather than removing a file just change the Build property to “Not in Build”:

This has the advantage that your project maintains some semblance of history of what objects have been removed from your database – that can be useful to anyone inheriting your code in the future.

Recommendation #12: Use “Not in Build” to remove an object from your database

Build and Deploy your datadude projects outside of Visual Studio

Building and deploying your datadude projects within Visual Studio can become a real time hog; in my experience its not unusual for deployments to take many minutes and your Visual Studio environment will be unavailable for further development work during that time. For that reason I recommend investing some time in writing some msbuild scripts that will build and deploy your project(s) from the command-line. Here are some examples that you can adapt for your own use, firstly a script to build a solution:

and secondly a script that will deploy a datadude project:

Writing these scripts may appear to be laborious but they’ll save you heaps of time in the long run.

Recommendation #13: Build and deploy to your development sandbox using scripts

UPDATE: Upon reading this blog post Mordechai Danielov wrote a follow-up in which he published a useful script that builds a series of projects using Powershell. Its at building your database solutions outside of Visual Studio.

Useful links

Over the years I’ve collected some links to MSDN articles that have proved invaluable:

• Build and Deploy Databases to an Isolated Development Environment
• An Overview of Database Build and Deployment
• Apply Test-Driven Development to your Database Projects
• Walkthrough: Partition a Database Project by Using Composite Projects
• Right sizing the master.dbschema file for better design time performance
• Referencing for Shared Server-Level Objects

Datadude bugs

Like any substantial piece of software datadude is not without bugs. Many of the issues I have found are concerned with the datadude interpreter not correctly parsing T-SQL code, here’s a list of some bugs that I have found down the years:

• Under certain circumstances a Refactor->Rename operation can crash Visual Studio (followed up here)
• Valid T-SQL MERGE syntax can cause false warnings
• Refactor->Rename doesn’t work properly for database unit tests
• Removal of triggers from your datadude project may, under certain circumstances, cause a failed deployment
• Rare naming circumstance can cause false warnings
• Valid function definition containing a CTE can cause false warnings
• A default constraint is attempted to be deployed before a function referenced by that constraint – hence an error occurs (confirmed as fixed in VS2010 SP1)
• Datadude doesn’t catch a type mismatch between a table-valued-parameter and a parameter passed to it
• DateTime2 Data generator doesn't generate unique values when the column is part of a composite key
• Unit test class files cannot contain periods
• Valid CREATE VIEW definition throws errors
• MERGE syntax in a stored procedure creates false warnings

Some of these bugs were reported a long time ago and may well have been fixed in later service packs.

Previous datadude blog posts

I have blogged on datadude quite a bit in the past:

• Workaround for datadude deployment bug – NullReferenceException
• Building a database installer with WiX, datadude and Visual Studio 2010
• Generate multiple SqlCmdVars files in your database projects
• Querying Visual Studio project files using T-SQL and Powershell
• Maintaining version history in your database using Visual Studio 2010
• Enforcing naming conventions using database unit testing
• Experiences from writing sp_CascadingDataViewer – DB unit testing and code distribution
• Some thoughts on Visual Studio database references and how they should be used for SQL Server BI
• Microsoft publish Visual Studio 2010 Database Project Guidance
• Setting up database unit testing as part of a Continuous Integration build process [VS2010 DB Tools - Datadude]
• A strategy for managing security for different environments using the Database Development Tools in Visual Studio 2010

Summing up

This has been an inordinately large blog post so if you’ve read this far – well done. For easy reference, here are all the recommendations that I have made:

1. Use Source Control and implement a Continuous Integration deployment
2. Turn on Code Analysis
3. When running your deployment in a test environment, run it more than once
4. Maintain an automated history of your deployments
5. Use Schema View in preference to Solution Explorer
6. Any action in a Pre or Post-Deployment Script should use PRINT to state what has been done
7. Output the value of all variables in your Pre-Deployment script
8. Each database object should be defined in a dedicated file
9. Time your deployments and output the timings
10. All literal values should be stored in your .sqlcmdvars file
11. Every developer should use the Isolated Dev Environment settings in order to author their code
12. Use “Not in Build” to remove an object from your database
13. Build and deploy to your development sandbox using scripts

I really hope this proves useful because its taken a good long while to get it published If you have any feedback then please let me know in the comments.

Thanks for reading!

@jamiet

* When I started writing this blog post the first sentence was “Over the past six months I have worked on two separate projects for customers that wanted to make use of Visual Studio 2010 Database projects to manage their database schema.” as opposed to what it is now: “Over the past eighteen months I have worked on four separate projects…” Yes, that’s how long its taken to write it!

Appendix – An example CI configuration

As stated above, an earlier draft of this blog post included full details of the CI configuration from one of the projects that I have worked on. Although it may repeat some of what has already been said I have included that text below.

==============================================================

Introduction

This project has invested heavily in using a Continuous Integration (CI) approach to development. What that means, succinctly, is that whenever someone checks-in some code to our source control system an automated build process is kicked off that constructs our entire system from scratch on a dedicated server. CI is not a new concept but it is fairly rare that anyone applies the same rigour to their database objects as they do to so-called “application code” (e.g. the stuff written in .Net code) and on this project we have made a conscious decision to properly build our databases as part of the CI build.

Datadude employs a declarative approach to database development. In other words you define what you want database schema to look like and datadude will work out what it needs to do to your target in order to turn it into what you have defined. What this means in practice is that you only ever write CREATE … DDL statements rather than IF <object-exists> THEN ALTER …ELSE CREATE … statements which is what you may have done in the past.

Here’s our CI environment setup:

• SubVersion (SVN) is being used for source control
• Hudson, an open source CI server, is being used to orchestrate our whole CI environment. It basically monitors our SVN repository and when it spots a checked-in file, kicks off the CI build
• Our CI scripts (the stuff that actually does the work) are written using msbuild
• We have 2 msbuild scripts:
  • Build.proj which is responsible for:
    • Compiling all our .Net website code
    • Building/Compiling our datadude projects  (every datadude project file is a msbuild-compliant script)
  • Deploy.proj which is responsible for:
    • Restoring latest database backups from our production environment into our CI environment
    • Deploying our built datadude projects on top of those restored backups
    • Build a folder structure to hold all the artefacts that get deployed
    • Creating folder shares
    • Moving SSIS packages into folder structure
    • Deploying SSRS reports to our SSRS server
    • Deploy our Analysis Services cube definitions to our Analysis Server
• Both Build.proj and Deploy.proj get executed by our CI build

Building datadude projects

Datadude makes it very easy to build datadude projects in a CI environment because they are already msbuild-compliant; its simply a call to the MSBuild task, passing in the location of the solution file as an argument. We use the Release configuration (although there is no particular reason for you to do the same – purely your choice):

That’s it! The output from a datadude build includes a number of files but the most important one is a .dbschema file which is an XML representation of all the objects in your database.

Deploying the output from a built datadude project

This is a little more difficult. We *could* simply use the MSBuild task to call our deployment script as we do for build script (see above) but the problem with that is that there are many pre-requisites (including datadude itself) and we don’t want to install Visual Studio and all the assorted paraphernalia onto our various environments. Instead we chose to make use of a command-line tool called VSDBCMD.exe to deploy datadude projects. VSDBCMD does basically the same job as what happens if you were to right-click on a datadude project in Visual Studio and select “Deploy” i.e. It compares the output of a build (A) to the target database (B) and works out what it needs to do to make B look like A. It then produces a .sql script that will actually make those requisite changes, then goes and executes it.

The difficulty comes in VSDBCMD.exe having its own list of file dependencies that are listed at MSDN article How to: Prepare a Database for Deployment From a Command Prompt by Using VSDBCMD.EXE, thankfully a much smaller list then if we were using the MSBuild task.

Some of those files, namely:

• Sqlceer35en.dll
• Sqlceme35.dll
• Sqlceqp35.dll
• Sqlcese35.dll
• System.Data.SqlServerCe.dll

get installed with SQL Server CE. We bundle along the x86 & x64 installers for SQL Server CE along with all the rest of our deployment artefacts and then, as part of Deploy.proj, install them like so:

That takes care of some of the dependencies but we still have to take care of:

• DatabaseSchemaProviders.Extensions.xml
• Microsoft.Data.Schema.dll
• Microsoft.Data.Schema.ScriptDom.dll
• Microsoft.Data.Schema.ScriptDom.Sql.dll
• Microsoft.Data.Schema.Sql.dll
• Microsoft.SqlServer.BatchParser.dll

as well as the actual VSDBCMD.exe file itself. Quite simply we keep those files in SVN and then bundle them along with all our deployment artefacts (I won’t show you how we do that because its out of the scope of this post and besides if you’re at all proficient with msbuild then you’ll know how to do that and if you’re not, well, why are you reading this?)

==============================================================

Five days ago I published a fairly scathing attack on the folks that built ExcelMashup.com due to their poor support for folks that wanted to provide feedback. Well, credit where credit is due, they have been pretty quick to respond as I have been informed that the site has been updated with a link to the dedicated forum which is hosted on Microsoft Answers. In an offline email exchange I promised them that if they did the right thing then I would update my blog accordingly, hence the post that you are reading right now.

Note that I do not know whether or not they have started responding to the email alias mentioned at http://www.excelmashup.com/Content/Community.html.

@jamiet

Earlier today I posted Data Explorer walkthrough – Parsing a Twitter list in which I explained that I have been using Data Explorer for a few weeks now. In that time I have been compiling some feedback for the Data Explorer product team which I was going to email privately but I see no harm in putting these thoughts into the public domain, hence why I am writing this blog post – a hodge podge of thoughts/suggestions/gripes from the past few weeks of using Data Explorer.

• I’m very impressed that if I change the name of a resources any references to that resource are updated accordingly. After years of working with the SSIS dev tools I’m automatically wired to expect that not to work.
• I love the composability of tasks and resources (e.g. Define a custom function as a resource and use it in another resource). That idea of composability and reuse is something that is sadly lacking in SSIS (and the SSIS team know that because I’ve moaned about it often enough ).
• If the name of the resource is too large for the box then there is no way to see the full name without editing it:

• It would be nice to be able to reorder the list of resources
• We *really* need the ability to widen the columns when viewing the results of a task.
• Data Explorer seems a bit lacking in its abilities to parse HTML & XML right now. For example, I'm surprised that I can't use XPath to extract the contents of an HTML/XML document. In fact, the lack of such features is the most disappointing aspect of Data Explorer thus far and means that the product is still a long way off being a Kapow competitor, remembering that this is still an early beta of course.
• We need better ways of visualising the output from a mashup (i.e. something that doesn’t require the user to have Excel installed). How about giving SSRS the ability to consume an OData feed – we could then host such a report on SQL Azure Reporting, perhaps even sell that report via the Azure Marketplace.
• If I make a mashup publicly available I would like to be able to know how many times it is called. I believe that the ability to INSERT data (e.g. into SQL Azure) using Data Explorer will be coming soon (its kind of already there with the Snapshot feature) so I'm wondering if there is a way to "trigger" an insertion when someone consumes one of my exposed resources? I suspect the answer is "not right now" so consider that a feature request. I guess you could crystalise that request as "Provide an eventing model within a mashup so that resources can be triggered when some event (e.g. Someone consumes a resource) occurs."
• We need a scheduler that enables us to run Snapshots at some pre-defined time.
• I have a mashup published at https://ws41451459.dataexplorer.sqlazurelabs.com/Published/TwitterListDemo. The link for the feed is https://ws41451459.dataexplorer.sqlazurelabs.com/Published/TwitterListDemo/Feed but that doesn’t actually show me any data, it shows me a list of resources that produce the data however in this case there is only one resource (https://ws41451459.dataexplorer.sqlazurelabs.com/Published/TwitterListDemo/Feed/jamiet-dataexplorer-text_user_screen_name). Would it not make sense to link directly to the feed with data in it if there is only actually one external resource in my mashup? It would save a mouse-click at least.
• I haven’t yet figured out if its possible to do recursion. Still working on that one.
• In my Twitter List example the values “jamiet” & “data-explorer” for the list owner and list slug respectively were hardcoded in the mashup; yes, they were parameters to a function but they were still hardcoded. It would have been much better if I were able to give the consumer of the mashup the ability to define/override those values when they consume it. In other words, a mashup needs to be parameterizable.
• Overall I am very impressed and the excitement that I displayed in my initial post Thoughts on Data Explorer is justified. Data Explorer coalesces nicely with existing interests of mine such as SQL Server, ETL & web-enabled data and I have high hopes that I will be using this extensively for clients in the years to come.

Ever since I started writing SSIS packages back in 2004 I have opted to store configurations in .dtsConfig (.i.e. XML) files rather than in a SQL Server table (aka SQL Server Configurations) however recently I inherited some packages that used SQL Server Configurations and thus had to immerse myself in their murky little world. To all the people that have ever gone onto the SSIS forum and asked questions about ambiguous behaviour of SQL Server Configurations I now say this... I feel your pain!

The biggest problem I have had was in dealing with the change to the order in which configurations get applied that came about in SSIS 2008. Those changes are detailed on MSDN at SSIS Package Configurations however the pertinent bits are:

As the utility loads and runs the package, events occur in the following order:

1. The dtexec utility loads the package.
2. The utility applies the configurations that were specified in the package at design time and in the order that is specified in the package. (The one exception to this is the Parent Package Variables configurations. The utility applies these configurations only once and later in the process.)
3. The utility then applies any options that you specified on the command line.
4. The utility then reloads the configurations that were specified in the package at design time and in the order specified in the package. (Again, the exception to this rule is the Parent Package Variables configurations). The utility uses any command-line options that were specified to reload the configurations. Therefore, different values might be reloaded from a different location.
5. The utility applies the Parent Package Variable configurations.
6. The utility runs the package.

The very nature of SQL Server Configurations means that the Connection String for the database holding the configuration values needs to be supplied from the command-line. Typically then the call to execute your package resembles this:

dtexec /FILE Package.dtsx /SET "\Package.Connections[SSISConfigurations].Properties[ConnectionString]";"\"Data Source=SomeServer;Initial Catalog=SomeDB;Integrated Security=SSPI;\"",

The problem then is that, as per the steps above, the package will (1) attempt to apply all configurations using the Connection String stored in the package for the "SSISConfigurations" Connection Manager before then (2) applying the Connection String from the command-line and then (3) apply the same configurations all over again. In the packages that I inherited that first attempt to apply the configurations would timeout (not unexpected); I had 8 SQL Server Configurations in the package and thus the package was waiting for 2 minutes until all the Configurations timed out (i.e. 15seconds per Configuration) - in a package that only executes for ~8seconds when it gets to do its actual work a delay of 2minutes was simply unacceptable.

We had three options in how to deal with this:

1. Get rid of the use of SQL Server configurations and use .dtsConfig files instead
2. Edit the packages when they get deployed
3. Change the timeout on the "SSISConfigurations" Connection Manager

#1 was my preferred choice but, for reasons I explain below*, wasn't an option in this particular instance. #2 was discounted out of hand because it negates the point of using Configurations in the first place. This left us with #3 - change the timeout on the Connection Manager. This is done by going into the properties of the Connection Manager, opening the "All" tab and changing the Connect Timeout property to some suitable value (in the screenshot below I chose 2 seconds).

This change meant that the attempts to apply the SQL Server configurations timed out in 16 seconds rather than two minutes; clearly this isn't an optimum solution but its certainly better than it was.

So there you have it - if you are having problems with SQL Server configuration timeouts within SSIS try changing the timeout of the Connection Manager. Better still - don't bother using SQL Server Configuration in the first place. Even better - install RC0 of SQL Server 2012 to start leveraging SSIS parameters and leave the nasty old world of configurations behind you.

@Jamiet

* Basically, we are leveraging a SSIS execution/logging framework in which the client had invested a lot of resources and SQL Server Configurations are an integral part of that.

Just recently I've inherited some SSIS packages that were in dire need of fixing however, as is often the case, most of my battles were with connection string configurations.

It always baffles me when I see packages that don't log information that would be useful for debugging purposes and when its me that has to debug those packages I tend to get a little irate. Do a favour to yourself and the poor soul that inherits your packages by placing a Script Task at the start of your package with the following code in it:

        public void Main()
        {
            bool failure = false;
            bool fireAgain = true;
            foreach (var ConnMgr in Dts.Connections)
            {
                Dts.Events.FireInformation(1, "", String.Format("ConnectionManager='{0}', ConnectionString='{1}'",
                    ConnMgr.Name, ConnMgr.ConnectionString), "", 0, ref fireAgain);
                try
                {
                    ConnMgr.AcquireConnection(null);
                    Dts.Events.FireInformation(1, "", String.Format("Connection acquired successfully on '{0}'", 
                        ConnMgr.Name), "", 0, ref fireAgain);
                }
                catch (Exception ex)
                {
                    Dts.Events.FireError(-1, "", String.Format("Failed to acquire connection to '{0}'. Error Message='{1}'", 
                        ConnMgr.Name, ex.Message),
                        "", 0);
                    failure = true;
                }
            }
            if (failure)
                Dts.TaskResult = (int)ScriptResults.Failure;
            else
                Dts.TaskResult = (int)ScriptResults.Success;
        }

SCR Output Connection Strings: ConnectionManager='DB', ConnectionString='Data Source=dev;Initial Catalog=AdventureWorks;Integrated Security=SSPI;'
SCR Output Connection Strings: Connection acquired successfully on 'DB'

@jamiet

P.S. Those of you that have been following my blog long enough may know that I posted this back in 2005 however I don't think there's any harm in putting it out there again, especially given that:

1. more people are now using SSIS
2. The previous code was VB.net
3. In the previous post the code was in a JPEG thus not copy/paste-able (god only knows why I did that)