46
DBMS_PROFILER

Oracle8i provides a Profiler API to profile existing PL/SQL applications and to identify performance bottlenecks. You can use the collected profiler (performance) data for performance improvement or for determining code coverage for PL/SQL applications. Application developers can use code coverage data to focus their incremental testing efforts.

The profiler API is implemented as a PL/SQL package, DBMS_PROFILER, that provides services for collecting and persistently storing PL/SQL profiler data.

This chapter discusses the following topics:

• Using DBMS_PROFILER
• Requirements
• Security
• Exceptions
• Error Codes
• Summary of DBMS_PROFILER Subprograms

Using DBMS_PROFILER

Improving application performance is an iterative process. Each iteration involves the following steps:

1. Running the application with one or more benchmark tests with profiler data collection enabled.
2. Analyzing the profiler data and identifying performance problems.
3. Fixing the problems.

The PL/SQL profiler supports this process using the concept of a "run". A run involves running the application through benchmark tests with profiler data collection enabled. You can control the beginning and the ending of a run by calling the START_PROFILER and STOP_PROFILER functions.

A typical run involves:

• Starting profiler data collection in the run.
• Executing PL/SQL code for which profiler and code coverage data is required.
• Stopping profiler data collection, which writes the collected data for the run into database tables

  Note: The collected profiler data is not automatically stored when the user disconnects. You must issue an explicit call to the FLUSH_DATA or the STOP_PROFILER function to store the data at the end of the session. Stopping data collection stores the collected data.

As the application executes, profiler data is collected in memory data structures that last for the duration of the run. You can call the FLUSH_DATA function at intermediate points during the run to get incremental data and to free memory for allocated profiler data structures.

Flushing the collected data involves storing collected data in database tables. The tables should already exist in the profiler user's schema. The PROFTAB.SQL script creates the tables and other data structures required for persistently storing the profiler data.

Note that running PROFTAB.SQL drops the current tables. The PROFTAB.SQL script is in the RDBMS/ADMIN directory. Some PL/SQL operations, such as the first execution of a PL/SQL unit, may involve I/O to catalog tables to load the byte code for the PL/SQL unit being executed. Also, it may take some time executing package initialization code the first time a package procedure or function is called.

To avoid timing this overhead, "warm up" the database before collecting profile data. To do this, run the application once without gathering profiler data.

System-Wide Profiling

You can allow profiling across all users of a system, for example, to profile all users of a package, independent of who is using it. In such cases, the SYSADMIN should use a modified PROFLOAD.SQL script which:

• Creates the profiler tables and sequence
• Grants SELECT/INSERT/UPDATE on those tables and sequence to all users
• Defines public synonyms for the tables and sequence

  Note: Do not alter the actual fields of the tables.

  See Also: "FLUSH_DATA Function".

Requirements

DBMS_PROFILER must be installed as SYS.

Use the PROFLOAD.SQL script to load the PL/SQL Profiler packages.

Collected Data

With the Probe Profiler API, you can generate profiling information for all named library units that are executed in a session. The profiler gathers information at the PL/SQL virtual machine level. This information includes the total number of times each line has been executed, the total amount of time that has been spent executing that line, and the minimum and maximum times that have been spent on a particular execution of that line.

The profiling information is stored in database tables. This enables querying on the data: you can build customizable reports (summary reports, hottest lines, code coverage data, and so on. And you can analyze the data.

PROFTAB.SQL

The PROFTAB.SQL script creates tables with the columns, datatypes, and definitions as shown in Table 46-1, Table 46-2, and Table 46-3.

Table 46-1 Columns in Table PLSQL_PROFILER_RUNS

Column	Datatype	Definition
runid	number primary key	Unique run identifier from plsql_profiler_runnumber
related_run	number	Runid of related run (for client/server correlation)
run_owner	varchar2(32),	User who started run
run_date	date	Start time of run
run_comment	varchar2(2047)	User provided comment for this run
run_total_time	number	Elapsed time for this run in nanoseconds
run_system_info	varchar2(2047)	Currently unused
run_comment1	varchar2(2047)	Additional comment
spare1	varchar2(256)	Unused

Table 46-2 Columns in Table PLSQL_PROFILER_UNITS

Column	Datatype	Definition
runid	number	Primary key, references plsql_profiler_runs,
unit_number	number	Primary key, internally generated library unit #
unit_type	varchar2(32)	Library unit type
unit_owner	varchar2(32)	Library unit owner name
unit_name	varchar2(32)	Library unit name timestamp on library unit
unit_timestamp	date	In the future will be used to detect changes to unit between runs
total_time	number	Total time spent in this unit in nanoseconds. The profiler does not set this field, but it is provided for the convenience of analysis tools.
spare1	number	Unused
spare2	number	Unused

Table 46-3 Columns in Table PLSQL_PROFILER_DATA

Column	Datatype	Definition
runid	number	Primary key, unique (generated) run identifier
unit_number	number	Primary key, internally generated library unit number
line#	number	Primary key, not null, line number in unit
total_occur	number	Number of times line was executed
total_time	number	Total time spent executing line in nanoseconds
min_time	number	Minimum execution time for this line in nanoseconds
max_time	number	Maximum execution time for this line in nanoseconds
spare1	number	Unused
spare2	number	Unused
spare3	number	Unused
spare4	number	Unused

With Oracle8, a sample textual report writer(profrep.sql) is provided with the PL/SQL demo scripts.

Security

The profiler only gathers data for units for which a user has CREATE privilege; you cannot use the package to profile units for which EXECUTE ONLY access has been granted. In general, if a user can debug a unit, the same user can profile it. However, a unit can be profiled whether or not it has been compiled DEBUG. Oracle advises that modules that are being profiled should be compiled DEBUG, since this provides additional information about the unit in the database

Two Methods of Exception Generation

Each routine in this package has two versions that allow you to determine how errors are reported.

• A function that returns success/failure as a status value and will never raise an exception
• A procedure that returns normally if it succeeds and raises an exception if it fails

In each case, the parameters of the function and procedure are identical. Only the method by which errors are reported differs. If there is an error, there is a correspondence between the error codes that the functions return, and the exceptions that the procedures raise.

To avoid redundancy, the following section only provides details about the functional form.

Exceptions

Table 46-4

Exception	Description
version_mismatch	Corresponds to error_version.
profiler_error	Corresponds to either "error_param" or "error_io".

DBMS_PROFILER Exceptions

Error Codes

A 0 return value from any function denotes successful completion; a nonzero return value denotes an error condition. The possible errors are as follows:

• 'A subprogram was called with an incorrect parameter.'

      error_param constant binary_integer := 1; 
   
  

• 'Data flush operation failed. Check whether the profiler tables have been created, are accessible, and that there is adequate space.'

      error_io    constant binary_integer := 2; 
   
  

• There is a mismatch between package and database implementation. Oracle returns this error if an incorrect version of the DBMS_PROFILER package is installed, and if the version of the profiler package cannot work with this database version. The only recovery is to install the correct version of the package.

      error_version constant binary_integer := -1;