DB2 Locking, Part 3: Locks Versus Latches

So far in this series on DB2 locking we have offered up a broad overview of what locking is and then delved into the world of table and table space locks. In this short entry, before we tackles page and row locks, we are going to look at the difference between a lock and a latch.

A true lock is handled by DB2 using the IRLM. The IRLM, or internal resource lock manager, is both a separate subsystem and an integral component of DB2. As its name implies, it manages locks for DB2.

However, whenever doing so is practical, DB2 can lock resources without going to the IRLM. This type of lock is called a latch. True locks are always set in the IRLM. Latches, by contrast, are set internally by DB2, without going to the IRLM.

When a latch is taken instead of a lock, it is handled in the Buffer Manager by internal DB2 code; so the cross-memory service calls to the IRLM are eliminated. Latches are usually held only briefly—for a shorter duration than locks. Also, a latch requires about one-third the number of instructions as a lock. Therefore, latches are more efficient than locks because they avoid the overhead associated with calling an external address space. Latches are used when a resource serialization situation is required for a short time. Both latches and locks guarantee data integrity. 

In subsequent blog entries, any usage of the term lock generically, refers to both locks and latches.

DB2 Locking, Part 2: Table Space and Table Locks

Today's post is the second in our DB2 locking series and it covers the topic of table space and table locks.

Table Space Locks

A table space lock is acquired when a DB2 table or index is accessed. Note that I said accessed, not updated. The table space is locked even when simple read-only access is ­occurring.

Refer to to Table 1 below for a listing of the types of table space locks that can be acquired during the execution of an SQL statement. Every table space lock implies two types of access: the access acquired by the lock requester and the access allowed to other subsequent, concurrent processes.

Table 1. Table Space Locks

(click on image for larger view)


When an SQL statement is issued and first accesses data, it takes an intent lock on the table space. Later in the process, actual S-, U-, or X-locks are taken. The intent locks (IS, IX, and SIX) enable programs to wait for the required S-, U-, or X-lock that needs to be taken until other processes have released competing locks.

The type of table space lock used by DB2 during processing is contingent on several factors, including the table space LOCKSIZE specified in the DDL, the bind parameters chosen for the plan being run, and the type of processing requested. Table 2 provides a synopsis of the initial table space locks acquired under certain conditions.

Table 2. How Table Space Locks Are Acquired

(click on image for larger view)

A table space U-lock indicates intent to update, but an update has not occurred. This is caused by using a cursor with the FOR UPDATE OF clause. A U-lock is non-exclusive because it can be taken while tasks have S-locks on the same table space. More information on table space lock compatibility follows in Table 3.

An additional consideration is that table space locks are usually taken in combination with table and page locks, but they can be used on their own. When you specify the ­LOCKSIZE TABLESPACE DDL parameter, table space locks alone are used as the locking mechanism for the data in that table space. This way, concurrent access is limited and concurrent update processing is eliminated.

Similar in function to the LOCKSIZE DDL parameter is the LOCK TABLE statement. The LOCK TABLE statement requests an immediate lock on the specified table. The LOCK TABLE statement has two forms—one to request a share lock and one to request an exclusive lock.

     LOCK TABLE table_name IN SHARE MODE; 
     LOCK TABLE table_name IN EXCLUSIVE MODE;

You also can issue LOCK TABLE against a specific partition, for example:

     LOCK TABLE table-name 
          PARTITION integer IN SHARE MODE;

A locking scheme is not effective unless multiple processes can secure different types of locks on the same resource concurrently. With DB2 locking, some types of table space locks can be acquired concurrently by discrete processes. Two locks that can be acquired concurrently on the same resource are said to be compatible with one another.

Refer to Table 3 for a breakdown of DB2 table space lock compatibility. A Yes in the matrix indicates that the two locks are compatible and can be acquired by distinct processes on the same table space concurrently. A No indicates that the two locks are incompatible. In general, two locks cannot be taken concurrently if they allow concurrent processes to negatively affect the integrity of data in the table space.


Table 3. Table Space Lock Compatability

(click on image for larger view)

Table Locks

When segmented table spaces are involved, DB2 can use table locks . Table locks are always associated with a corresponding table space lock.

The same types of locks are used for table locks as are used for table space locks. S, U, X, IS, IX, and SIX table locks can be acquired by DB2 processes when data in segmented table spaces is accessed. Table 1 describes the options available to DB2 for table locking. The compatibility chart in Table 3 applies to table locks as well as table space locks.

For a table lock to be acquired, an IS-lock must first be acquired on the segmented table space in which the table exists. The type of table lock to be taken depends on the LOCKSIZE specified in the DDL, the bind parameters chosen for the plan being run, and the type of processing requested. Table 4 (below) is a modified version of Table 2 (shown earlier), depicting the initial types of table spaces and table locks acquired given a certain set of conditions. Table locks are not acquired when the LOCKSIZE TABLESPACE parameter is used.


Table 4. How Table Locks Are Acquired

(click on image for larger view)

Summary

And so concludes Part 2 of our multi-part series on DB2 locking. Be sure to keep watching this blog for additional postings in this series!

DB2 Locking, Part 1: An Overview

DB2 automatically guarantees the integrity of data by enforcing several locking strategies. These strategies permit multiple users from multiple environments to access and modify data concurrently. Basically, DB2 combines the following strategies to implement an overall locking strategy:

  • Table and table space locking

  • IRLM page and row locking

  • Internal page and row latching

  • Claims and drains to achieve partition independence

  • Checking commit log sequence numbers (CLSN) and PUNC bits to achieve lock avoidance

  • Global locking through the coupling facility in a data sharing environment

But what exactly is locking? How does DB2 utilize these strategies to lock pages and guarantee data integrity? Why does DB2 have to lock data before it can process it? What is the difference between a lock and a latch? How can DB2 provide data integrity while operating on separate partitions concurrently? Finally, how can DB2 avoid locks and still guarantee data integrity?

Today's blog post will offer a high level introduction to DB2 locking, why it is needed, as well as the benefits and drawbacks that it introduces to using DB2 databases.

Anyone accustomed to application programming when access to a database is required understands the potential for concurrency problems. When one application program tries to read data that is in the process of being changed by another, the DBMS must forbid access until the modification is complete to ensure data integrity. Most DBMS products, DB2 included, use a locking mechanism for all data items being changed. Therefore, when one task is updating data on a page, another task cannot access data (read or update) on that same page until the data modification is complete and committed.

When multiple users can access and update the same data at the same time, a locking mechanism is required. This mechanism must be capable of differentiating between stable data and uncertain data. Stable data has been successfully committed and is not involved in an update in a current unit of work. Uncertain data is currently involved in an operation that could modify its contents. 

Consider two DB2 application programs. If program #1 updates a piece of data on page 1, you must ensure that program #2 cannot access the data until program #1 commits the unit of work. Otherwise, a loss of integrity could result. Without a locking mechanism, the following sequence of events would be possible:

1. Program #1 retrieves a row from DSN81010.EMP for EMPNO ‘000010’.
2. Program #1 issues an update statement to change that employee’s salary to 55000.
3. Program #2 retrieves the DSN81010.EMP row for EMPNO ‘000010’. Because the change was not committed, the old value for the salary, 52750, is retrieved.
4. Program #1 commits the change, causing the salary to be 55000.
5. Program #2 changes a value in a different column and commits the change.
6. The value for salary is now back to 52750, negating the change made by program #1.

The DBMS avoids this situation by using a locking mechanism. DB2 supports locking at four levels, or granularities: table space-, table-, page-, and row-level locking. DB2 also provides LOB locking for large objects (BLOBs, CLOBs, and DBCLOBs).

Locks can be taken at any level in the locking hierarchy without taking a lock at the lower level. However, locks cannot be taken at the lower levels without a compatible higher-level lock also being taken. For example, you can take a table space lock without taking any other lock, but you cannot take a page lock without first securing a table space-level lock (and a table lock as well if the page is part of a table in a segmented table space).

Additionally, a page lock does not have to be taken before a row lock is taken. Your locking strategy requires an “either/or” type of choice by table space: either row locking or page locking. An in-depth discussion on the merits of both are beyond the scope of today's blog post, but will be covered in future posts. 

Both page locks and row locks can escalate to a table level and then to a table space level for segmented tables or straight to a table space level for partitioned table spaces. A table or table space cannot have both page locks and row locks held against it at the same time.

Many modes of locking are supported by DB2, but they can be divided into two types:

  • Locks to enable the reading of data                                

  • Locks to enable the updating of data

But remember, we are still talking at a very high, and somewhat simplistic level. DB2 uses varieties of these two types of locks to indicate the type of locking required. 

Locks Versus Latches

A true lock is handled by DB2 using the IRLM. However, whenever doing so is practical, DB2 tries to lock resources without going to the IRLM. This type of lock is called a latch. Whereas true locks are always set in the IRLM, latches are set internally by DB2, without going to the IRLM.

When a latch is taken instead of a lock, it is handled in the Buffer Manager by internal DB2 code; so the cross-memory service calls to the IRLM are eliminated. Latches are usually held only briefly—for a shorter duration than locks. Also, a latch requires about one-third the number of instructions as a lock. Therefore, latches are more efficient than locks because they avoid the overhead associated with calling an external address space. Latches are used when a resource serialization situation is required for a short time. Both latches and locks guarantee data integrity. In subsequent sections, any usage of the term lock generically, refers to both locks and latches.

Lock Duration

Before you learn about the various types of locks that can be acquired by DB2, you should understand lock duration, which refers to the length of time that a lock is maintained. The duration of a lock is based on the BIND options chosen for the program requesting locks. Locks can be acquired either immediately when the program is requested to be run or iteratively as needed during the execution of the program. Locks can be released when the plan is terminated or when they are no longer required for a unit of work.

The BIND parameters affecting DB2 locking are summarized below:

The BIND parameters that impact table space locks are the ACQUIRE and RELEASE parameters. 

• ACQUIRE(ALLOCATE | USE): The ALLOCATE option specifies that locks will be acquired when the plan is allocated, which normally occurs when the first SQL statement is issued. The USE option indicates that locks will be acquired only as they are required, SQL statement by SQL statement. As of DB2 10, ACQUIRE(ALLOCATE) is no longer supported when binding or rebinding plans.
  
• RELEASE(DEALLOCATE | COMMIT): When you specify DEALLOCATE for a program, locks are not released until the plan is terminated. When you specify COMMIT, table space locks are released when a COMMIT is issued.

The BIND parameter affecting page and row locks is the ISOLATION parameter. There are four choices for isolation level:

• ISOLATION(CS), or Cursor Stability, acquires and releases page locks as pages are read and processed. CS provides the greatest level of concurrency at the expense of potentially different data being returned by the same cursor if it is processed twice during the same unit of work.
• ISOLATION(RR), or Repeatable Read, holds page and row locks until a COMMIT point; no other program can modify the data. If data is accessed twice during the unit of work, the same exact data will be returned.
• ISOLATION(RS), or Read Stability, holds page and row locks until a COMMIT point, but other programs can INSERT new data. If data is accessed twice during the unit of work, new rows may be returned, but old rows will not have changed.
• ISOLATION(UR), or Uncommitted Read, is also known as dirty read processing. UR avoids locking altogether, so data can be read that never actually exists in the ­database.

Regardless of the ISOLATION level chosen, all page locks are released when a COMMIT is encountered.

I think that is enough for this first installment on DB2 locking... come back soon and we will expand on table space locks in the next installment of this blog series on DB2 locking...

DB2 Table Editors

In today's blog post I want to briefly discuss one of the more useful, yet often ignored, DB2 tools available on the market -- Table Editors...

Typically, the only method of updating DB2 data (indeed, any data stored in a relational database) is with  SQL data manipulation language statements DELETE, INSERT, and UPDATE (or with a database load). Because these SQL statements operate on data a set at a time, multiple rows -- or even all of the rows -- can be affected by a single SQL statement. Coding SQL statements for every data modification required during the application development and testing phase can be time-consuming and error-prone.

A table editing tool can reduce the time needed to make simple data modifications by providing full-screen edit capability for database tables. The user specifies the table to edit and is placed into an edit session. The data is presented to the user as a series of rows, with the columns separated by spaces. A header line indicates the column names. The data can be scrolled up and down as well as left and right. To change data, the user simply types over the current data.

This type of tool is ideal for supporting the application development process. A programmer can make quick changes without coding SQL. Also, if properly implemented, a table editor can reduce the number of erroneous data modifications made by beginning SQL users.

When a table editor is used, all columns are available for update. Thus, if a table editor is used to change production data, a simple mis-keying can cause unwanted updates. Native SQL should be used if you must ensure that only certain columns are updated.

One final note: Tested SQL statements and application programs are characterized by their planned nature. These modification requests are well thought out and tested. This is not true for changes implemented through a table editor, so always exercise caution when using a table editor.

Examples of DB2 Table Editors include:

• BMC Catalog Manager (which includes an integrated table editor)
• CA RC/Update (which includes an integrated table editor)
• IBM DB2 Table Editor 
• RazorSQL DB2 Table Editor

The Importance of Database Design

It is impossible to over-estimate the importance of database design on the effectiveness and efficiency of application systems. The first step, of course, is to create a logical data model of the business information that must be stored in, and accessed through, the database. This is a non-trivial task, but it is not the focus of today’s blog post, which is implementing optimal and well-designed database schemata.

Oftentimes the database design does not get the attention it deserves. This can occur for numerous reasons such as:

• Insufficient specifications and/or poor logical data modeling
• Not enough time in the development schedule
• Too  many changes occurring throughout the development cycle
• Database design assigned to, or performed by novices

And even when the database design is being performed by experienced professionals with sufficient time and a solid logical model, it is very easy for design flaws to creep into the database. This is especially the case for larger and more complex databases required to support mission critical applications.

Of course, the first step in constructing a physical database should be transforming the logical design using best practices. The transformation consists of the following:

• Transforming entities into tables
• Transforming attributes into columns
• Transforming domains into data types and constraints
• Transforming relationships into primary and foreign keys

But a simple transformation will not result in a complete and correct physical database design – it is merely the first step.  And design flaws can be introduced even during such a transformation.

The process of normalizing your data should be conducted during the logical design phase, but sometimes mistakes are made during the logical modeling process. In simple terms, normalization is the process of identifying the one best place where each fact belongs. A normalized design minimizes data redundancy and optimizes data structures by systematically and properly placing data elements into the appropriate groupings.

• First normal form (1NF) eliminates repeating groups and non-atomic data from an entity.
• Second normal form (2NF) ensures that all the attributes of each entity are dependent on the primary key.
• Third normal form (3NF) ensures that no relationships between attributes exist within an entity.

Although normalization is a logical process and does not necessarily dictate physical database design it is good practice to implement normalized physical databases – especially with today’s powerful hardware and database systems. A normalized data model will ensure that each entity is well formed and that each attribute is assigned to the proper entity. And denormalized data structures should always be documented.

A database that is not normalized can be difficult to modify (because facts may be strewn about within the database) and prone to data quality issues (again because one fact may be in more than one place). Failing to achieve appropriate normalization is a significant design flaw.

Design flaws can make it difficult to access a database in an efficient manner. Something as simple as creating columns with inconsistent data type and length across tables can introduce inefficiency. For example, perhaps JOB_CODE is defined as CHAR(10) in one table and CHAR(15) in another. If these columns need to be compared or joined, additional work is required by the DBMS to make the columns comparable.  It is easy to see how a database design issue can make applications inefficient no matter how adept the developers are. And this is just one type of design flaw.

There are numerous other types of design flaws that can negatively impact the usability and efficiency of a database implementation (and the applications that use it). Improperly defined constraints (referential, check, uniqueness) can cause data quality problems. Improper indexing (to support constraints and other physical structures) can cause a database to operate inefficiently or even cease operating altogether. In DB2 for z/OS, failing to explicitly name a clustering index causes DB2 to default cluster sequence to the oldest index. Changing index structures, which sometimes require dropping and recreating the index, can cause the data to be ordered in a completely different physical sequence, thereby impacting performance.

Some database design flaws are more subtle. What happens if you create two incompatible check constraints? For example, consider the following:

     CHECK (empno < 100)

     CHECK (empno >= 101)

No data can be added to the table because no number is less than 100 and greater than or equal to 101. Of course, this is an extreme example to highlight the problem. Another situation can occur if the default value is not one of the values permitted by the check constraint, for example:

    emp_type  CHAR(8)  DEFAULT 'new'

       CHECK (emp_type IN ('temp',
                           'fulltime',
                           'contract')), ...

Cycles are another potential problem that can be created in a physical database schema. A cycle is a referential path that connects a table to itself. This can occur when multiple tables are related back to each other and it looks like a loop when diagrammed. DB2 forbids a table from being delete-connected to itself in a cycle. A table is delete-connected to another table if it is a dependent of a table specified with a CASCADE delete rule.

Nested triggers can also create problems. A trigger is fired by an INSERT, UPDATE, or DELETE operation. However, a trigger can also contain INSERT, UPDATE, or DELETE logic within itself. Therefore, a trigger is fired by a data modification, but can also cause another data modification, thereby firing yet another trigger. When a trigger contains INSERT, UPDATE, and/or DELETE logic (including MERGE statements), the trigger is said to be a nested trigger. If referential integrity is combined with triggers, additional cascading updates and/or deletes can occur.

DB2 restricts the number of nested triggers that can be executed within a single firing event. If this were not done, it could be quite possible to have triggers firing triggers ad infinitum until all of the data was removed from an entire database. DB2 limits this cascading effect to 16 levels to prevent endless looping. If more than 16 levels of nesting occur, the transaction is aborted. Such nesting may restrict certain types of data modification from happening at all because the number of nested calls will always be exceeded. So nesting levels need to be controlled and managed in the database schema to avoid problems.

And this is by no means a comprehensive list of database design flaws that can cause real problems for application developers and end users. Getting the database design correct is imperative if you wish to have any hope of assuring application performance.

So how can you go about examining the correctness of your database structures? The best approach is an automated one such as that provided by Database Examiner, a product offered by DBE Software that performs a comprehensive validation of a database schema. The in-depth technology incorporated into Database Examiner allows you to review your database schema, identify problems, and remediate the issues to ensure a quality database implementation offering peak performance.

Simply provide Database Examiner with the data model, DDL, or a link to the active database and it will perform a series of comprehensive diagnostics. Database Examiner applies the rules of relational technology to detect flaws, inconsistencies and lack of integrity. The product applies more than 50 diagnostics that can be organized by category or severity levels. And you can customize the diagnostics by selecting those to be executed and assigning each one a severity level.

But Database Examiner doesn’t just highlight the flaws, it also helps to fix them be providing recommendations and corrections for each issue it detects, including the generation of SQL DDL scripts to remedy the flaws. Even better, Database Examiner provides multi-platform DBMS support, including DB2 for z/OS, DB2 LUW, Oracle, Microsoft SQL Server and Sybase.

And DBE Software is currently offering folks the ability to download and use a scaled down “Performance Edition” of Database Examiner, free ofcharge for a limited time. Be sure to enter the promo code MULL on the download request to get the best offer available.

Hopefully this blog entry has convinced you that database design is important… important enough to take some time to evaluate the quality of your existing database structures. And to take a look at automating the process using Database Examiner (using the promo code MULL ).