Tag Archives: normalization

Dealing with a Bad Design – #1

I am sure that there will be many posts about dealing with bad design. So, I am going to number them as we go so I can see just how many different bad designs need a solution!

Duplicate Fields

Like many shops, the Court never had a DBA.  If you look at the Court’s PDB (Platypus Database) it is obvious that developers with no idea on how to create a relational database had “designed” it.  For over 20 years, developers had just added tables as they needed for their specific task.  To make matters worse, the platform is an IBM iSeries (formerly known as the AS400) where, depending on how a program is written, a change to a table design requires the recompilation of all programs that use the table.  And since there was no source control process it was impossible to tell WHICH programs would need to be recompiled.

The solution of course was to copy an existing table that had most of what the new process needed, rename it, and just add a few new fields…viola!!
That was the database design process.

This causes havoc when trying to write applications that need to interface with the database. The first issue, how do you keep all the data synchronized? If you update data in one table, how does it get updated in the other? The second issue, which fields need to be included in a query? The ones from the first table or the second, almost identical, table? Unfortunately in cases like this the only way to know is to ask someone and learn the structure. Eventually you will figure out when to use which table.

Fake Dates

Another example, back in the day when storage was a main consideration and before there were specific date fields, the Court decided to use a six digit INTEGER data type as the ‘Date’ field, YYMMDD. So looking at the date 111 – what date would you figure that to be? Did you guess January 11, 2000? No? huh, I’m surprised that you didn’t see that 😉

Then there was the Y2K issue. A perfect time to fix that bad decision because by then the AS400 DID have a Date datatype. Nope…let’s just add an EIGHT long INTEGER data type to every table that duplicates the six digit date and store the dates as YYYYMMDD.

And again we have to rely on someone making sure that all the dates are updated correctly. But of course there are exceptions. There’s at least one table that the eight digit date did not get added. It has to be joined into another table using the six digit date…but sometimes that data isn’t updated so the query returns no data.

Committing Spreadsheet

But I think one of the more prevalent bad design decisions is not creating the proper many to many structure – committing spreadsheet. Remember in the last post we talked about normalization and the many-to-many relationship of Jurors to Trials. The WRONG way to store this information is to do something like:

TrialID DefendantName Juror1 Juror2 Juror3 Juror4 Juror5 Juror6 ....Juror 25

Here’s why this is wrong. It is rigid – what if someone wants 26 jurors? Now you have to change the table design (and maybe the application) just to add one more person to a Jury Panel. What if they only want 20 Jurors? now there are NULLs in the data for Juror21 – Juror25. (While NULL has its place in databases this is not one of them – NULL means ‘unknown’ – these Jurors are not unknown – they don’t exist, that is a different state.)

Other Platypus Designs

I would love to hear how your PDB is designed! Send me an email or comment on your design issues and I will see if I have a solution to post for you!

Developer Golden Egg

When you have to work with a database where a non-PDB Hunter has committed spreadsheet, you can use SQL to create what SHOULD have been done. Using a UNION we can create the many-to-many table like so:

SELECT 1 as ID, TrialID, Juror1 As JurorID FROM BadTable Where Juror1 is not null
UNION
SELECT 2, TrialID, Juror2 FROM BadTable Where Juror2 is not null
...
SELECT 25, TrialID, Juror25 FROM BadTable Where Juror25 is not null

Now you have a result set that is normalized and can be used as part of another query, to join into, display results from, the same things you could do with it if it HAD been designed correctly.

Normalizing – How do you do it??

Hopefully everyone has read FundamentalsOfDBDesign by Paul Litwin that I mentioned in the last post.  (If not, perhaps take a few minutes to stop and read it and then come back and join us, understanding some of those basics will be key to this installment.  It’s ok, this post will still be here waiting when you finish.)

There are many pages out there that talk about normalizing data.  I feel that many of them only go part way.  Some of them are a little too “canned”.  Even Litwin’s examples are some what simplified.  They all have a membership database, or a customer database or something simple and a list of fields to be categorized.  Well that’s all fine and dandy, but how do you get to that point?  Where did that list of fields come from?  How do you work with data in the real world?  I could never find any websites or blog posts on HOW to get that.  Here’s how I do it.

I was lucky enough at the Court to be able to design and build from scratch and then re-do, a couple of times for version upgrades, a core Court application – the Jury Management System (JMS).  When I started at the Court the “Jury System” on the AS400 was basically an upload of a text file of summoned Jurors on which one or two functions or processes could be performed.  All the rest of the work was performed manually – mail merging letters, tracking time worked, tracking panels served on, tracking trials for which a panelist was selected, and making payments, just to name a few.  it took months to get a Jury term processed and delivered to Accounting to start the payment process. There was one report that was manually compiled every two weeks. I was assigned the task of creating a new system that would allow the staff to automate all these processes and more.

During the initial design phase I took the time to create a proper relational database.  Over the past 10 years and 3 major revisions that database structure is still valid and the only required modifications have been to include new data that wasn’t part of the original application or to accommodate major changes in functionality.

When you are lucky enough to get to design from scratch here are some things that you should look for to help determine what fields you need to have in your database:

What are your inputs?

In my case it was the text file that I was being sent that contained the names of people who had been summoned for jury duty.  Did I need everything from the file? No, but I could look at the file and determine I should probably get their name, address, date for which they were summoned, other information that identified the people that have been asked to complete a form and return it to the Jury Division. I didn’t need the name of the Authorizing Clerk of the Court or the Court’s address. Those pieces of information may have been important to the Agency creating and sending the Court the file, but not to our application.

What are you expected to output?

Like most applications, what I was building was replacing something, some manual task that some person has to spend a lot of time compiling and working. Well, what did that person produce? Is it a report? What is on the report? If you are being asked to report the number of Trials that a person served on you need to make sure that somewhere in the database you capture some trial information and what jurors served on that trial. If you don’t capture it, you can’t report on it. A lot of developers skip this step and it is CRUCIAL!

What is your application expected to do? What data will it need in order to perform that task?

One function the Jury application was expected to perform was to calculate the number of hours a Juror worked in order to pay them. Well, in order to do that you have to know all the days a person was in attendance, what time they arrive and what time they left. But, there are special rules for Public Employees. They only get paid when they are at the Court outside their regular working hours because Agencies are required to pay employees while on Jury Duty. So now we need to know which of the people who showed up for Jury Duty are public employees and what their regular start and end times are. But then there are special cases like Substitute Teachers, who are Public Employees, but get paid for all hours worked; or, teachers on Summer Break. Jurors also get paid for mileage to and from their primary residence to the Court round-trip for each day they come. Now you need to know the Round Trip Mileage for each person.
All of these rules determine what data is required to be stored in order for the application to do the appropriate calculation to pay the Juror correctly.

Now you have a list of all the fields you think that your application will need to do the work it is supposed to do.  Now what?  Well, you have to decide what tables to create.  How do you do that??

Well, by looking at the functions your application is required to perform, inputs and outputs you should have a general idea of some of your major tables.  In the case of the JMS, and given just the information I shared above, I would determine a need for the following tables: JUROR, TRIAL and ATTENDANCE.  Plus, we also need some where to keep all the information about PUBLIC EMPLOYEES – we need to know who they are and when they work and if there are any special conditions that affect their pay.  We need to know where the JURORs live so an ADDRESS table might be a good candidate.

So you go through your list of potential fields and determine what table they belong to.  It doesn’t matter if you don’t get it right the first time, you just need to make a first pass and see what you KNOW goes certain places and then you’ll be left with the fields you’re not quite sure about.  It may turn out that something you thought was a FOR SURE you end up changing.

Each table must have a PRIMARY KEY.  I prefer auto-incrementing keys.  They are easy to create and maintain.  But regardless, perhaps there is a natural key that you prefer using, every table has a KEY.  So each table starts off having an ID field.  This ID refers to ONE row in the set.

The rest of the fiels in each table are added by asking yourself, when deciding if a field belongs in a table, “Does this field pertain to ONLY what is in this table?”.  Some things are no brainers.  In our list we have the fields from our input file: name, address, SSN, DOB, the date they are to show up for jury duty.  Those might at first be all things that you would select for the JURORs table.  The TRIALS that a JUROR served on?  That really isn’t information about the JUROR.  So you wouldn’t want to put TRIAL information in the JUROR table.

So where would you put TRIAL information that is about the JURORS?  We have a JUROR table that lists everything we know about the JUROR. We have a TRIAL table that lists everything about the TRIAL.  Where DO you store which JURORS served on which TRIALS?  In the JUROR_TRIAL table!   We will have a table that contains an ID for each row in this set and an ID for the JUROR and an ID for the TRIAL they served on.  (This structure is how a many-to-many relationship is represented: many jurors can serve on many trials.  We will get into relationships another day.)

Hopefully this has helped you to understand the actual steps taken to get a database design into a normalized state.

Feel free to contact me if you need any assistance with your design or have any questions about normalization!

theplatypusdatabase@gmail.com

Welcome all Platypus Hunters

Welcome to the Platypus Database – a blog dedicated to the databases all over the world that were not designed, but seemingly put together with very little thought – like the Platypus

220px-Platypus-sketch

Not to malign the poor Platypus, but it does look like no one really thought about how it would look with all those different parts stuck together like that – and it lays eggs, which for a mammal is kind of weird.

About ten years ago I posted this discussion at www.tek-tips.com asking if anyone else had to deal the same issues in their databases that I was finding – no normalization, shortcuts to avoid having to do things the right way, dates, validation (mostly the lack of),  and other situations.  (The thread quickly deteriorated into a treatise on Chimera Database – which almost became the name of this blog but I couldn’t get it to work as well.)

So why am I writing this blog?

I developed many applications against the Platypus Database over the past ten years and learned many things that are useful in certain situations when dealing with bad data and bad design.  You need to learn monster SQL skills and really understand the power of OUTER JOINS to overcome some of the issues you will find.  I would like to share those tips with other developers to help them learn how to use the power of the query to only get the data you really need into your application and minimize database calls.

As a way to continually improve and learn myself, I am working towards my MS SQL Server 2012 certification.  I want to blog about things I find as I learn about MS SQL Server and the studying and testing that I find exciting and want to share.

At this same time, my friend Jenny is trying to find a way to get into the IT world and I have been looking and looking for some web pages that can explain some basics about databases.  Every page that I found was either someone asking for help to design their database or did not mention normalization at all or if it did mention it, didn’t apply it correctly.  If you are going to work with databases in any way, you should understand the basic concepts and how to create a real database that follows the rules.

So, I am finally spurred into action to write about these topics so my friend can benefit and learn things the right way and we can all move towards certification as Platypus Hunters.

Thanks for dropping by, I hope to see you again soon!

Leslie Andrews, Platypus Hunter Extraordinaire