data modeling

Events – Before Database Generate (More or less)

And we’re back.  Sorry for the delay.  Between day job, the worlds worst sinus infection, and becoming a pee-wee soccer coach (which required actually learning the rules of soccer), it’s been a busy end of summer.

Part n of our m part series on events continues.  Today we’re going to talk about the Before Database Generate event.  I could take the easy road by pointing out that I’ve already covered these in my post on adding seed data to a table but we’ll take it a little further.

First, let’s talk about before database generate events (and their cousin after database generate events) in general.  What can we use them for?  Well, here’s the list I came up with.

• To set the properties of an object that need to be modified at generation time such as parameters that are dependent on the target.  Many of these things can be done in pre-generation transforms, and this is probably the right thing to do in most cases.  Putting them in a a before database generate event has the advantage that the changes remain in the model and don’t just exist in the generated DDL.  So if it’s important to keep a model for posterity, or perhaps audit purposes, using a pre-generation event might be better.
• To integrate data or information from other sources which may also be changing (e.g. seed data)
• To automate versioning activities such as performing a model check-in when DDL is generated
• Setting object properties that are dependent on the current date

Now, just because you can do these things doesn’t necessarily mean you should.  This event is at the MODEL level.  This should give you an idea that Sybase really intended this as something you’d use for versioning, external tool interface, etc. rather than for manipulating individual objects (that’s what transforms are for) prior to generation.  You can still write code to go through your objects and take whatever action you want, but think long and hard about it.

I’ve covered external integration with my seed data demo and Sybase has some good material provided on interacting with the repository, so for our example I’m going to modify objects based on the date.

I’m going to configure my model so that for any date range partitioned table, I only have to specify the starting partition, and it will built out partitions until the current date.  I’ll be honest, this isn’t how I would choose to do this, in part because my sample is too simplistic to really be useful, in part because most databases now have interval partitions where you only have to tell it how long the date range for each partition is and it takes care of the rest, and in part because if this is something I need to automate in a warehousing environment, it’s usually something I prefer to build into the ETL or operations routine anyway.  But, it’s still a fun example, and it gave me an excuse to play with the GetPhysicalOptionValue method.

The idea here is we’ll define a single partition at the beginning of your date range.  If I wanted to make this more flexible, I’d establish what the interval was for each partition (daily, weekly, monthly, quarterly, yearly) as an extended attribute and the routine would take care of the rest using the options on the initial partition as a template.  If you take it this far, let me know.

We’ll start with a custom method on the table object.  Since partitions are fairly database specific, I should probably put this in the XDB file but I’m going to leave it in an XEM simply to make it easier to distribute.  This particular method is based on Oracle 11g.

Here’s the code for my method.  I’ll walk through it below.

sub %method%(obj)
'BuildPartitions method

dim part
 dim part_col_code
 dim part_col
 dim col
 dim part_value
 dim start_dt
 dim part_cnt

dim myregexp
 dim mymatches
 dim mymatch

if not obj.getextendedattributetext("tablepropertiestablepartitioningclausesrangeorcompositepartitioningclausepresence") then
 output "table " & obj.code & " is not range partitioned"
 exit sub
 end if

' now assuming we have range partitions, is it a date? get the name of the column
 part_col_code = obj.getphysicaloptionvalue("<table_partitioning_clauses>/partition by range/<column_list>/<column>:1")

'now find that column object

for each col in obj.columns
 if col.code = part_col_code then
 set part_col = col
 end if
 next

' and see if it is a date.
 if part_col.datatype = "DATE" then

'now we retrieve the partition data for the single partition we've defined
 part_value = obj.getphysicaloptionvalue _
 ("<table_partitioning_clauses>/partition by range/<partition_list>/partition:1/values less than/<value_list>")

'now lets get the current date and build out partitions to support it

set myregexp = new regexp
 myregexp.ignorecase = true
 myregexp.global = true
 myregexp.pattern = "\d+/\d+"

set mymatches = myregexp.execute(part_value)

for each mymatch in mymatches
 start_dt = cdate(mymatch.value)
 next

output "Building partitions from " & start_dt & " to " & date & " on " & obj.Code

else
 output "table " & obj.code & " is not range partitioned by date"
 exit sub
 end if

' now we know we have a table that is range partitioned by a date column
 ' let's add monthly partitions from start date to current date + 1 (remember
 ' the partition is defined by the upper bound

' set the partition count - we have one defined by default, so we'll start with 2
 part_cnt = 2

do while start_dt <= dateadd("m",1,date)

obj.SetPhysicalOptionValue "<table_partitioning_clauses>/partition by range/<partition_list>/partition:" & _
 part_cnt & "/<partition_name>", "part" & year(start_dt) & Right("0" & month(start_dt), 2)
 obj.SetPhysicalOptionValue "<table_partitioning_clauses>/partition by range/<partition_list>/partition:" & _
 part_cnt & "/values less than/<value_list>", "to_date('" & year(start_dt) & "/" & Right("0" & month(start_dt), 2) & "','YYYY/MM')"

start_dt = dateadd("m",1,start_dt)
 part_cnt = part_cnt + 1
 loop

output part_cnt & " partitions built"

end sub
 

Let’s break it down. Our first task is to determine if the table is range partitioned, and if it’s a date.  To do this, we’ll check and extended  attribute on the table, then we’re going to inspect the physical options on the table to get the partitioning column we’ve defined and determine if it’s a date.

Lines 15-19 check the partitioning and terminate the subroutine if it’s not range partitioned.  We follow that immediately by using the getPhysicalOptionValue method to retrieve the name of the partitioning column.  I don’t find the help file on this method terribly informative.  So I’ll give you my Cliff’s Notes version of common errors.

1. There are two similar methods getPhysicalOptionValue and getPhysicalOptionValues.  The first retrieves the value you’ve set, the second retrieves the list of possible values, which for the partitioning column would retrieve a list of all columns in the table.  A misplaced S will cause a lot of confusion.
2. When you enter the option path, enter it exactly as displayed in the tree.  If it’s surrounded by “<” and “>” include them, if it doesn’t, do not.  You won’t get an error if you enter it wrong, you just won’t get a value.

Lines 26-33 find the column object referred to in the physical options by looping through the columns collection.  If the column is not a date, the subroutine will exit.

Lines 36 and 37 retrieve the “values less than clause”, again using the getPhysicalOption method.

I’m counting on my method having the value expressing in some sort of standard date format.  VBScript is pretty good at automatic date conversions, but use your best judgement here.  Lines 41-50 use regular expressions to extract the date portion from the value.  I’m extracting all numbers and the “/” character.  If you want a good tutorial on regular expressions, you are NOT on the right blog.  I know enough to be dangerous and how to use Google.  Here are two sites I find handy:

Regular Expression Library

Regular Expression Tester

Now that we’ve completed all of that work, we’ll loop through from our starting date to the current date (plus one month since we’re specifying an upper bound) in lines 66 to 75 and add partitions using the setPhysicalOptionValue method.  The same warnings apply as for getPhysicalOptionValue(s).  Notice that we need to change our index to build new partitions and that we start with 2 so we don’t overwrite the starter partition.

So, a lot of work and still no BeforeDatabaseGenerate event (I guess maybe I should’ve titled this one “Getting and Setting Physical Options”).  Let’s add it now.

I’m going to assume you’re familiar with adding classes and events to an XEM file, we’ve covered that before. Remember it’s got to be at the MODEL level.  Here’s the code:

Function %BeforeDatabaseGenerate%(model, fct)

Dim myObj
 Dim mySel

Set mySel = model.CreateSelection
 mySel.AddObjects ActiveModel, cls_table, false, true

For each myObj in mySel.Objects
 With myObj
 Output .Name
 .ExecuteCustomMethod("Events2.BuildPartitions")
 end with
 Next

%BeforeDatabaseGenerate% = True

End Function

Nothing surprising here at this point (I hope).  As usual, we’re using a selection to find all our tables, and then executing our method for each.  You’ll notice that the input parameter for the method isn’t Obj for this event, since it only applies to models, it’s been replaced with the more specific Model on lines 1 and 6.

So there you have it.  A few words of wisdom (or maybe just words) on what you can and should do with BeforeDatabaseGenerate events, and a brief sample.  Remember, this event is probably best reserved for integration with external files or sources  and version control rather than actual model manipulation.  Use transforms for that.  However, in a pinch, here’s another tool for the box.

Best of luck.

Events – Can Create

So now we come to the speculative fiction part of the events series.  CanCreate events are not something I’ve used so we’re making this up as we go.  I’ve been playing around with them and doing some thinking on where and how they might be applicable.  canCreate events are a bit different than the others.  They prevent you from creating an object or assigning a stereotype instead of automating some task.  As I read through the examples, it seemed that these might be a bit more applicable in model types other than data models.

I’m generally not one to modify a tool to restrict someone from using an entire metaclass.  Most of the metaclasses in an RDBMS are there for a reason, and to take away a potentially useful tool seems like a bad idea to me.  Perhaps your company or DBAs have decided that “thou shalt not use views” or “thou shalt not use triggers” and if so, then restricting them with a canCreate trigger might be acceptable.  But not recommended.

So, thinking through the way we use models, I came up with a couple of cases for using the canCreate event.

1) Restricting the use of a metaclass (“thou shalt not use triggers”)

2) Ensuring that pre-conditions or model configuration is completed before allowing the modeler to create objects

3) Restricting the use of a stereotype

I’m going to do one example the shows demonstrates the first two cases, then we’ll handle stereotypes (poorly) separately.

Let me stress again that removing a metaclass entirely is NOT something I would typically recommend.  However, at least on one of my projects, I can actually think of a place where I might use it.  We currently have a team of modelers working on a Teradata implementation.  In this environment we’re building a set of base tables which will store data for a variety of user groups in one database.  Access to these tables will be managed through a set of view layers.  Each group will have their own view database.  Any physical structures such as indexes must be in the shared table database.  Each database has its own model, so by flagging the model as either a view or table database, I can use a canCreate trigger to ensure that the view databases don’t contain any tables.  We’ll create an extended attribute at the model level and make sure that the modeler has set the database type before they create any tables.  I think that this targeted, conditional use of canCreate triggers, which require you to make a few “declarations” at the model level when you start are probably the right way to go and could potentially be useful.

So, let’s walk through an example.

First, I’m going to create an extended attribute on the model metaclass.  I’ll call it DatabaseType and give it three potential values: “view”, “table” and “not set” (which will be the default).  Using a default makes sure someone actually goes in and sets the value.  Then, we’ll put our canCreate event on the table metaclass.

Notice that for the canCreate event the function has a “parent” parameter rather than the obj property that points to a specific object.  For objects such as tables or views, this will be the model.  For columns, indexes, etc. this will be the table.

Here’s the code:

Function %CanCreate%(parent)

   if parent.GetExtendedAttributeText("DatabaseType") = "View" then
      %CanCreate% = False
   elseif parent.GetExtendedAttributeText("DatabaseType") = "Not Set" then
      %CanCreate% = False
   else
      %CanCreate% = True
   end if

End Function

This will prevent anyone from creating a table without setting the DatabaseType property at the model level to “table”.  I have several extended model files that rely on setting a set of model level attributes that are used by initialization events and other scripts to automate common tasks.  Using canCreate triggers, I could ensure that model level variables are set before we begin.

Next case.. restricting use of a stereotype.  I thought long and hard on a good example for this and honestly I could not come up with one.  I’m sure there’s one out there, please send ’em if you’ve got ’em, but I just couldn’t come up with it.  I’m going to use a bad example, just so we can illustrate the point.

So, for the sake of argument, here goes.  I’ve decided to classify my tables as associative, attributive or independent.  In addition, I’m going to assign all of my columns political parties.  Naturally, an independent table can’t contain Republican or Democratic columns (independent or green columns are fine).  Did I mention this is a bad example?  For this example, I’m going to use a stereotype to indicates an independent table.  There are some rules I could use to determine this programmatically but that’s probably another post.

So, here’s what we end up with.

As you can see, we’ve got two stereotypes, each with a canCreate event.  The code is the same in both triggers.

Function %CanCreate%(parent)
   ' Implement your creation validation rule on <parent> here
   ' and return True in case of success

   if parent.stereotype = "Independent" then
      %CanCreate% = False
   else
      %CanCreate% = True
   end if

End Function

Here the parent property points to the enclosing table.  We evaluate the stereotype and return true or false, where false indicates you cannot utilize the stereotype and true indicates you can.

Notice that our table marked as Independent only allows columns to be assigned to the Green or Independent party.

So, another event down.  If you have a good use for the canCreate event in a data model, please do comment here or e-mail me.  I’d really love a better example.