Wednesday, February 17, 2016

The Most Misunderstood Features of DB2 – Part 4: Base Table Views

Welcome to my continuing series focusing on The Most Misunderstood Features of DB2. Today’s topic is an oldie but a goodie: base table views.

What is a Base Table View?
I am going to assume that regular readers of this blog know what a view is (if not, go here). So what is a base table view? It is a view that contains all of the columns and all of the rows of the base table and nothing else. In other words, it is like a SELECT * against the base table with no WHERE clauses.

Now why would somebody create such a beast? This gets to the heart of the misunderstanding here. There is a continuing notion “out there” that it is a good idea to create a base table view for every table and to give programmers access to the base table view only, not the base table itself.

The reason given for doing so is to insulate programs from change. But this is a weak argument at best. I admit that this particular misunderstanding is less prevalent than it was in the earlier years of DB2. But I do still find shops adhering to this archaic, and poorly founded, idea.

Reasons Given for Base Table Views
One reason given by base table view proponents is that when you add a column to a table you do not have to change the program. But this is the case, too, if you code your programs appropriately, avoiding SELECT * and coding only the specific columns you need to access. Most shops do this even when using base table views.

Other reasons given revolve around removing columns or splitting tables. If you are interested in all of the reasons and rebuttals I urge you to read my article (PDF) on this topic from 1991 titled One View Per Base Table? Don’tDo It! (Wow! This is a quarter of a Century old now!)

Another View Fallacy
I recall yet another misunderstanding about views from many years ago that I have not heard in a long time. Basically, the argument was that accessing a base table view would out-perform accessing the base table.

This was never proven to me – and I do not believe it was ever true. How could it be? Static SQL against a base table view would use view merge so it would be the same as going against the base table. And dynamic SQL would have the added step of having to resolve the view to the base table. Sooo…

Summary

Base table views are not worth the time and effort because the bring no added value of any type whatsoever. Do not use them.

Wednesday, February 10, 2016

The Most Misunderstood Features of DB2 – Part 3: Nulls

Welcome to Part 3 in my on-going blog series on The Most Misunderstood Features of DB2. You can find the first two parts here: Part 1 (on locking) and Part 2 (OPTIMIZE FOR v. FETCH FIRST). Today’s topic is one that confuses many SQL developers, Nulls.

What is a Null?
A null represents missing or unknown information at the column level. When a column is set as null, it can mean one of two things: the attribute is not applicable for certain occurrences of the entity, or the attribute applies to all entity occurrences, but the information may not always be known. Of course, it could be a combination of these two situations, too.

A null is not the same as 0 or blank. Null means no information has been specified for the column and it implies that the value is either unknown or not applicable.

Because DB2 supports null you can distinguish between a deliberate entry of 0 (for numerical columns) or a blank (for character columns) and an unknown or inapplicable entry (NULL for both numerical and character columns). Null indicates that the user did not explicitly make an entry or has explicitly entered NULL for the column. For example, a null “value” in the Price column of the ITEM table in some database does not mean that the item is being given away for free; instead it means that the price is not known or has not yet been set.

Nulls sometimes are inappropriately referred to as “null values.” Using the term value to describe a null is inaccurate because a null implies the lack of a value. Therefore, it is a good practice to simply use the term null or nulls (without appending the term “value” or “values” to it).

A Few Examples 
When are nulls useful? Well, defining a column as NULL provides a placeholder for data you might not yet know. For example, when a new employee is hired and is inserted into the EMP table, what should the employee termination date column be set to? I don’t know about you, but I wouldn’t want any valid date to be set in that column for my employee record. Instead, null can be used to specify that the termination date is currently unknown.

Let’s consider another example. Suppose that we also capture employee’s hair color when they are hired. Consider three potential entity occurrences: a man with black hair, a woman with unknown hair color, and a bald man. The woman with the unknown hair color and the bald man both could be assigned as null, but for different reasons. The woman’s hair color would be null meaning presently unknown; the bald man’s hair color could be null too, in this case meaning not applicable. But there are probably better methods of dealing with the bald man’s hair color.

How could you handle bald without using nulls? You could create special values for the HairColor column that mean “bald” and “unknown.” This is possible for a CHAR column like HairColor, but what about a DB2 DATE column? All occurrences of a column assigned as a DATE data type are valid dates. It might not be possible to use a special date value to mean “unknown.” This is where using a null is most practical.
DB2 does not differentiate between nulls that signify unknown data and those that signify inapplicable data. This distinction must be made by the program logic of each application.

Keep in mind, though, that using null to indicate “not applicable” can be an indication of improper database design. By properly modeling and normalizing your data structures you can usually eliminate the need to use nulls to indicate that a column is inapplicable for a specific row.

DB2 and Indicator VariablesDB2 represents null in a special “hidden” column known as an indicator variable. An indicator variable is defined to DB2 for each column that can accept nulls. The indicator variable is transparent to the end user, but must be managed when programming in a host language (such as Java, COBOL or PL/I).

The null indicator is used by DB2 to track whether its associated column is null or not. A positive value or a value of 0 means the column is not null and any actual value stored in the column is valid. If a CHAR column is truncated on retrieval because the host variable is not large enough, the indicator value will contain the original length of the truncated column. A negative value indicates that the column is set to null. If the value is -2 then the column was set to null as the result of a data conversion error.

Let’s take a moment to clear up one common misunderstanding right here: nulls NEVER save storage space in DB2 for z/OS. Every nullable column requires one additional byte of storage for the null indicator. So, a CHAR(10) column that is nullable will require 11 bytes of storage per row – 10 for the data and 1 for the null indicator. This is the case regardless of whether the column is set to null or not.

DB2 for Linux, Unix, and Windows has a compression option that allows columns set to null to save space. Using this option causes DB2 to eliminate the unused space from a row where columns are set to null. This option is not available on the mainframe, though.

Every column defined to a DB2 table must be designated as either allowing or disallowing nulls. A column is defined as nullable – meaning it can be set to NULL – in the table creation DDL. Null is the default if nothing is specified after the column name. To prohibit the column from being set to NULL you must explicitly specify NOT NULL after the column name. In the following sample table, COL1 and COL3 can be set to null, but not COL2, COL4, or COL5:

CREATE TABLE SAMPLE1
 (COL1 INTEGER,
  COL2 CHAR(10) NOT NULL,
  COL3 CHAR(5),
  COL4 DATE NOT NULL WITH DEFAULT,
  COL5 TIME NOT NULL);

In SELECT statements, testing for null is accomplished differently than testing for other “values.” You cannot specify WHERE COL = NULL, because this does not make any sense. Remember, null is a lack of a value so the column does not equal anything. Instead, you would have to code WHERE COL IS [NOT] NULL. In INSERT statements NULL can be specified in the VALUES clause to indicate that a column is to be set to NULL; but in UPDATE statements you can use the equality predicate (=) to assign a column to NULL. When inserting data, if the user fails to make an entry in a column that allows nulls, DB2 supplies the NULL as the default (unless another default value exists). If an attempt to insert NULL is made against a column defined as NOT NULL, the statement will fail.

More Confusion
OK, now that we have discussed what nulls are and the basics of using nulls in DB2, let’s look at how the presence of nulls can confuse SQL developers.

The AVG, COUNT DISTINCT, SUM, MAX, and MIN functions omit column occurrences set to null. The COUNT(*) function, however, does not omit columns set to null because it operates on rows. Thus, AVG is not equal to SUM/COUNT(*) when the average is being computed for a column that can contain nulls. You must be aware of which columns can be null as well as which functions operate on columns versus rows to make sure you code correctly!

It is important to understand, too, that for comparison in a SELECT statement, two null columns are not considered to be equal. When a nullable column participates in a predicate in the WHERE or HAVING clause, the nulls that are encountered cause the comparison to evaluate to UNKNOWN. But ORDER BY, GROUP BY, DISTINCT, and unique indexes consider nulls to be equal and handle them accordingly.

Additionally, when a column set to null participates in a calculation, the result is null. This is true even if the calculation is NULL/0 (which as every math savvy person knows should be an error). But because there is a null in the calculation, the result will be null!

Let’s look at another statement that confuses a lot of people:

SELECT SUM(SALARY)
FROM   EMP
WHERE  JOBCODE > 999;

Assume that SALARY is defined as NOT NULL. Furthermore, assume that the largest JOBCODE is 500 (that is, less than the 999 we are checking for). What is the results of this SQL?

No rows satisfy the predicate. As such, many people say the results should be 0. But it is not. The result is null! Yes, you can get a null result when performing functions on a non-nullable column. The predicate resolves to the empty set and the sum of the salaries in the empty set is null because there are no employees, and hence no salaries. Therefore, you better make sure that you code a null-indicator in your program when you write code like this.

The Bottom Line
Nulls are confusing, but cannot be completely avoided in DB2. Take the time to learn how nulls work and how they impact your coding and development practices… or you will certainly make mistakes.



----------------------------------------------------------------------------------------------------------
Additional resources for understanding nulls and their usage:

Tuesday, February 02, 2016

The Most Misunderstood Features of DB2 – Part 2: Optimize vs. Limited Fetch

Welcome to Part 2 in my on-going blog series on The Most Misunderstood Features of DB2. In Part 1 of the series we tackled the topic of locking, which IMHO is easily the most misunderstood feature of DB2 (probably of most DBMSes).

Today's topic is a brief one, but one that I've found folks to be confused about. Namely, the difference between the OPTIMIZE FOR and FETCH x ROWS ONLY clauses.

The OPTIMIZE FOR n ROWS and FETCH FIRST n ROWS ONLY are similar when it comes to influencing the optimizer. Both will provide help to the optimizer about the scale of the query (that is, the number of rows to expect when optimizing the access path to the data).

But, the two clauses have different purposes. If all you want to do is influence the optimizer then you should use OPTIMIZE FOR n ROWS as its usage does not limit the number of rows that can be fetched when the statement is executed. For example, you might specify OPTIMIZE FOR 1 ROW on a SELECT statement. But if 500 rows qualify then you can FETCH all 500 rows at run time. 

(Note: the specific numbers used in the previous paragraph are not important; it is important only that you know the number of rows specified in the OPTIMIZE clause does not inhibit the number of rows that can later be accessed by the query/program.)

The FETCH FIRST n ROWS ONLY clause is different. It actually limits the number of rows that can be fetched when the query is run. If you specify the clause FETCH FIRST 5 ROWS ONLY, then only 5 rows can be fetched - even if 100 or 1000 or more actually qualify. And yes, the number of rows you specify for FETCH FIRST will be provided to the optimizer for access path formulation, but it also will change the way the query behaves when it runs. 

So, when you write a SELECT statement that qualifies a large number of rows, but you only need a few rows at run time, it is a good idea to use FETCH FIRST n ROWS ONLY. On the other hand, if you are simply trying to tweak query performance by getting the optimizer to consider a different number of qualifying rows, then you should consider using OPTIMIZER FOR n ROWS.

The two statements are quite different and it is important that developers not confuse the two!

Tuesday, January 26, 2016

The Most Misunderstood Features of DB2 – Part 1: Locking


Today I am introducing a new series of blog posts here on misunderstood DB2 features and functions. But before I start this blog post I want to emphasize that this is just my opinion. I’m sure many of you have your own ideas of the DB2 features that are most misunderstood. But please, take a moment to consider my thoughts here… and then share your own in the comments section below!

Locking!

One of the most misunderstood features of DB2 is how locking works. This is probably to be expected because database locking is a rather complex topic and it is constantly morphing with each and every new version of DB2.

First of all you have to understand the lock options available and how DB2 locks at each level: table space, table, page and row. You need to understand share and exclusive locks and that the manner in which DB2 takes them depends upon the SQL statements you issue, the program bind options you choose, and even, in some cases, the DDL options you’ve chosen.

And more. What about claims and drains? Do you know the different between locks and latches? What about the difference between lock suspension, timeouts, and deadlocks?

Then there is the whole issue of lock avoidance and what that means. I visit a lot of shops each year and I still see a lot of ambiguous cursors. Simply adding FOR READ ONLY to all of the cursors in your program where you will not be modifying the data can make a world of performance difference because DB2 can avoid locking.

There is also the frequently-ignored need for a commit strategy. Every batch program should have a commit strategy implemented whereby modifications are committed to the database and locks are released. Failing to do this causes other programs to timeout waiting for locks. I wrote about this in the past and I call it Bachelor Programming Syndrome.

Another locking issue is the whole dirty read everywhere problem. Programmers learn that they can improve performance by adding WITH UR to the end of their SQL statements in their programs. UR, Uncommitted Read – aka dirty read, causes DB2 to not take locks. But that can cause problems because the program is reading uncommitted data – and that data it is reading may never actually exist in the database (it could be rolled back) and may not be consistent because it is not reading only committed units of work. I’ve visited shops that use WITH UR everywhere and they insist that they have not encountered any problems. To which I always add, “…yet.”

Then there are the new features added in the last few releases of DB2, including improved lock efficiencies when accessing the DB2 Catalog, the ability to access currently committed data, optimistic locking, skipping locked data, and more.

Summary

Yes, it is easy to misunderstand locking, and many do so. Taking the time to bolster your knowledge of how DB2 locks, the locking options available, and how they work can do you and your shop a world of good.


And be sure to check back here for future blogs about misunderstood DB2 features!

Monday, January 11, 2016

The Impact of Data Volume on Operational Databases

Operational databases are growing in size for many reasons, not the least of which is the growing importance of big data and analytics projects. There is the overarching trend of more and more data being generated every year. But also, there is the growing need to store more data for longer periods of time due to regulatory and compliance issues. Some organizations and business have encountered the need to store certain types of data for 100 years or more (as this video and this storage project point out).

But I doubt that I really need to convince you that your databases are growing in size. Most DBAs experience the reality of data growth every day.

As data volumes expand, it impacts operational databases in two ways:
  1. additional data stresses transaction processing and can cause performance slow downs, and;
  2. database administration tasks are negatively impacted.
In terms of performance, the more data in the operational database, the less efficient transactions running against that database tend to be. Table scans must reference more pages of data to return a result. Indexes grow in size to support larger data volumes, causing access by the index to degrade because there are more levels to traverse to return an answer. Such performance impacts are causing many companies to seek solutions that offload older data to either reference databases or to archive data stores.

The other impact, database administration complexity, causes longer processing time and outages to perform such functions as backups, unloads, reorganizations, recoveries, and disaster recoveries.  The larger the underlying data sets for your tables and table spaces, the longer it takes to run administrative utilities for them. In many cases the lengthened outages can become unacceptable, causing companies to again seek ways to lighten up the operational databases... or perhaps acquire next generation utility technology that understands the reality of large DB2 database objects.

But even though we want to keep all of that additional data, there is no reason that it necessarily has to be stored in operational databases that run the business. For many reasons, you probably want to separate active data from historical data. 

Some companies create purge jobs for all (or many) of their tables to remove data from the production databases as it ages. This can be an acceptable approach to reduce the size of your operational databases. But it also means that the data, which you might want to keep for analytical purposes, is lost. Another approach is to archive the data. Archiving data and purging data are two different processes. When data is purged, it is removed from the operational database and discarded. But archived data is removed from the operational database and maintained in an archive data store. The archive might be a flat file, another relational table or to HDFS using Hadoop.

The bottom line is that it makes sense for us, as DBAs, to keep any eye on the size of our operational databases and take action when production workload is impacted.