7 SQL*Loader Command-Line Reference

Alternative Ways to Specify Parameters

If the length of the command line exceeds the size of the maximum command line on your system, you can put certain command-line parameters in the control file by using the OPTIONS clause.

You can also group parameters together in a parameter file. You specify the name of this file on the command line using the PARFILE parameter when you invoke SQL*Loader.

These alternative ways of specifying parameters are useful when you often use the same parameters with the same values.

Parameter values specified on the command line override parameter values specified in either a parameter file or in the OPTIONS clause.

BAD (bad file)

Default: The name of the datafile, with an extension of .bad.

BAD specifies the name of the bad file created by SQL*Loader to store records that cause errors during insert or that are improperly formatted. If you do not specify a filename, the default is used. A bad file is not automatically created if there are no rejected records.

A bad file filename specified on the command line becomes the bad file associated with the first INFILE statement in the control file. If the bad file filename was also specified in the control file, the command-line value overrides it.

BINDSIZE (maximum size)

Default: To see the default value for this parameter, invoke SQL*Loader without any parameters, as described in Invoking SQL*Loader.

BINDSIZE specifies the maximum size (bytes) of the bind array. The size of the bind array given by BINDSIZE overrides the default size (which is system dependent) and any size determined by ROWS.

COLUMNARRAYROWS

Default: To see the default value for this parameter, invoke SQL*Loader without any parameters, as described in Invoking SQL*Loader.

Specifies the number of rows to allocate for direct path column arrays. The value for this parameter is not calculated by SQL*Loader. You must either specify it or accept the default.

CONTROL (control file)

Default: none

CONTROL specifies the name of the SQL*Loader control file that describes how to load the data. If a file extension or file type is not specified, it defaults to .ctl. If the filename is omitted, SQL*Loader prompts you for it.

If the name of your SQL*Loader control file contains special characters, your operating system may require that they be preceded by an escape character. Also, if your operating system uses backslashes in its file system paths, you may need to use multiple escape characters or to enclose the path in quotation marks. See your Oracle operating system-specific documentation for more information.

DATA (datafile)

Default: The name of the control file, with an extension of .dat.

DATA specifies the name of the datafile containing the data to be loaded. If you do not specify a file extension or file type, the default is .dat.

If you specify a datafile on the command line and also specify datafiles in the control file with INFILE, the data specified on the command line is processed first. The first datafile specified in the control file is ignored. All other datafiles specified in the control file are processed.

If you specify a file processing option when loading data from the control file, a warning message will be issued.

DATE_CACHE

Default: Enabled (for 1000 elements). To completely disable the date cache feature, set it to 0.

The date cache is used to store the results of conversions from text strings to internal date format. The cache is useful because the cost of looking up dates is much less than converting from text format to date format. If the same dates occur repeatedly in the data file, then using the date cache can improve the speed of a direct path load.

DATE_CACHE specifies the date cache size (in entries). For example, DATE_CACHE=5000 specifies that each date cache created can contain a maximum of 5000 unique date entries. Every table has its own date cache, if one is needed. A date cache is created only if at least one date or timestamp value is loaded that requires datatype conversion in order to be stored in the table.

The date cache feature is only available for direct path loads. It is enabled by default. The default date cache size is 1000 elements. If the default size is used and the number of unique input values loaded exceeds 1000, then the date cache feature is automatically disabled for that table. However, if you override the default and specify a nonzero date cache size and that size is exceeded, then the cache is not disabled.

You can use the date cache statistics (entries, hits, and misses) contained in the log file to tune the size of the cache for future similar loads.

DIRECT (data path)

Default: false

DIRECT specifies the data path, that is, the load method to use, either conventional path or direct path. A value of true specifies a direct path load. A value of false specifies a conventional path load.

DISCARD (filename)

Default: The name of the datafile, with an extension of .dsc.

DISCARD specifies a discard file (optional) to be created by SQL*Loader to store records that are neither inserted into a table nor rejected.

A discard file filename specified on the command line becomes the discard file associated with the first INFILE statement in the control file. If the discard file filename is specified also in the control file, the command-line value overrides it.

DISCARDMAX (integer)

Default: ALL

DISCARDMAX specifies the number of discard records to allow before data loading is terminated. To stop on the first discarded record, specify one (1).

ERRORS (errors to allow)

Default: To see the default value for this parameter, invoke SQL*Loader without any parameters, as described in Invoking SQL*Loader.

ERRORS specifies the maximum number of insert errors to allow. If the number of errors exceeds the value specified for ERRORS, then SQL*Loader terminates the load. To permit no errors at all, set ERRORS=0. To specify that all errors be allowed, use a very high number.

On a single-table load, SQL*Loader terminates the load when errors exceed this error limit. Any data inserted up that point, however, is committed.

SQL*Loader maintains the consistency of records across all tables. Therefore, multitable loads do not terminate immediately if errors exceed the error limit. When SQL*Loader encounters the maximum number of errors for a multitable load, it continues to load rows to ensure that valid rows previously loaded into tables are loaded into all tables and rejected rows are filtered out of all tables.

In all cases, SQL*Loader writes erroneous records to the bad file.

EXTERNAL_TABLE

Default: NOT_USED

EXTERNAL_TABLE instructs SQL*Loader whether or not to load data using the external tables option. There are three possible values:

• NOT_USED - the default value. It means the load is performed using either conventional or direct path mode.

• GENERATE_ONLY - places all the SQL statements needed to do the load using external tables, as described in the control file, in the SQL*Loader log file. These SQL statements can be edited and customized. The actual load can be done later without the use of SQL*Loader by executing these statements in SQL*Plus.

• EXECUTE - attempts to execute the SQL statements that are needed to do the load using external tables. However, if any of the SQL statements returns an error, then the attempt to load stops. Statements are placed in the log file as they are executed. This means that if a SQL statement returns an error, then the remaining SQL statements required for the load will not be placed in the log file.

  If you use EXTERNAL_TABLE=EXECUTE and also use the SEQUENCE parameter in your SQL*Loader control file, then SQL*Loader creates a database sequence, loads the table using that sequence, and then deletes the sequence. The results of doing the load this way will be different than if the load were done with conventional or direct path. (For more information about creating sequences, see CREATE SEQUENCE in Oracle Database SQL Language Reference.)

Note that the external tables option uses directory objects in the database to indicate where all datafiles are stored and to indicate where output files, such as bad files and discard files, are created. You must have READ access to the directory objects containing the datafiles, and you must have WRITE access to the directory objects where the output files are created. If there are no existing directory objects for the location of a datafile or output file, SQL*Loader will generate the SQL statement to create one. Therefore, when the EXECUTE option is specified, you must have the CREATE ANY DIRECTORY privilege. If you want the directory object to be deleted at the end of the load, you must also have the DELETE ANY DIRECTORY privilege.

When using a multitable load, SQL*Loader does the following:

1. Creates a table in the database that describes all fields in the datafile that will be loaded into any table.

2. Creates an INSERT statement to load this table from an external table description of the data.

3. Executes one INSERT statement for every table in the control file.

To see an example of this, run case study 5, but add the EXTERNAL_TABLE=GENERATE_ONLY parameter. To guarantee unique names in the external table, SQL*Loader uses generated names for all fields. This is because the field names may not be unique across the different tables in the control file.

FILE (tablespace file to load into)

Default: none

FILE specifies the database file to allocate extents from. It is used only for direct path parallel loads. By varying the value of the FILE parameter for different SQL*Loader processes, data can be loaded onto a system with minimal disk contention.

LOAD (number of records to load)

Default: All records are loaded.

LOAD specifies the maximum number of logical records to load (after skipping the specified number of records). No error occurs if fewer than the maximum number of records are found.

LOG (log file)

Default: The name of the control file, with an extension of .log.

LOG specifies the log file that SQL*Loader will create to store logging information about the loading process.

MULTITHREADING

Default: true on multiple-CPU systems, false on single-CPU systems

This parameter is available only for direct path loads.

By default, the multithreading option is always enabled (set to true) on multiple-CPU systems. In this case, the definition of a multiple-CPU system is a single system that has more than one CPU.

On single-CPU systems, multithreading is set to false by default. To use multithreading between two single-CPU systems, you must enable multithreading; it will not be on by default. This will allow stream building on the client system to be done in parallel with stream loading on the server system.

Multithreading functionality is operating system-dependent. Not all operating systems support multithreading.

PARALLEL (parallel load)

Default: false

PARALLEL specifies whether direct loads can operate in multiple concurrent sessions to load data into the same table.

PARFILE (parameter file)

Default: none

PARFILE specifies the name of a file that contains commonly used command-line parameters. For example, the command line could read:

sqlldr PARFILE=example.par

The parameter file could have the following contents:

USERID=scott/tiger
CONTROL=example.ctl
ERRORS=9999
LOG=example.log

READSIZE (read buffer size)

Default: To see the default value for this parameter, invoke SQL*Loader without any parameters, as described in Invoking SQL*Loader.

The READSIZE parameter is used only when reading data from datafiles. When reading records from a control file, a value of 64 kilobytes (KB) is always used as the READSIZE.

The READSIZE parameter lets you specify (in bytes) the size of the read buffer, if you choose not to use the default. The maximum size allowed is platform dependent.

In the conventional path method, the bind array is limited by the size of the read buffer. Therefore, the advantage of a larger read buffer is that more data can be read before a commit operation is required.

For example:

sqlldr scott/tiger CONTROL=ulcas1.ctl READSIZE=1000000

This example enables SQL*Loader to perform reads from the external datafile in chunks of 1,000,000 bytes before a commit is required.

The READSIZE parameter has no effect on LOBs. The size of the LOB read buffer is fixed at 64 kilobytes (KB).

See BINDSIZE (maximum size).

RESUMABLE

Default: false

The RESUMABLE parameter is used to enable and disable resumable space allocation. Because this parameter is disabled by default, you must set RESUMABLE=true in order to use its associated parameters, RESUMABLE_NAME and RESUMABLE_TIMEOUT.

RESUMABLE_NAME

Default: 'User USERNAME (USERID), Session SESSIONID, Instance INSTANCEID'

The value for this parameter identifies the statement that is resumable. This value is a user-defined text string that is inserted in either the USER_RESUMABLE or DBA_RESUMABLE view to help you identify a specific resumable statement that has been suspended.

This parameter is ignored unless the RESUMABLE parameter is set to true to enable resumable space allocation.

RESUMABLE_TIMEOUT

Default: 7200 seconds (2 hours)

The value of the parameter specifies the time period during which an error must be fixed. If the error is not fixed within the timeout period, execution of the statement is terminated, without finishing.

This parameter is ignored unless the RESUMABLE parameter is set to true to enable resumable space allocation.

ROWS (rows per commit)

Default: To see the default value for this parameter, invoke SQL*Loader without any parameters, as described in Invoking SQL*Loader.

Keep in mind that if you specify a low value for ROWS and then attempt to compress data using table compression, your compression ratio will probably be degraded. Oracle recommends that you either specify a high value or accept the default value when compressing data.

Conventional path loads only: ROWS specifies the number of rows in the bind array. See Bind Arrays and Conventional Path Loads.

Direct path loads only: ROWS identifies the number of rows you want to read from the datafile before a data save. The default is to read all rows and save data once at the end of the load. See Using Data Saves to Protect Against Data Loss. The actual number of rows loaded into a table on a save is approximately the value of ROWS minus the number of discarded and rejected records since the last save.

SILENT (feedback mode)

When SQL*Loader begins, information about the SQL*Loader version being used appears on the screen and is placed in the log file. As SQL*Loader executes, you also see feedback messages on the screen, for example:

Commit point reached - logical record count 20

SQL*Loader may also display data error messages similar to the following:

Record 4: Rejected - Error on table EMP
ORA-00001: unique constraint <name> violated

You can suppress these messages by specifying SILENT with one or more values.

For example, you can suppress the header and feedback messages that normally appear on the screen with the following command-line argument:

SILENT=(HEADER, FEEDBACK)

Use the appropriate values to suppress one or more of the following:

• HEADER - Suppresses the SQL*Loader header messages that normally appear on the screen. Header messages still appear in the log file.

• FEEDBACK - Suppresses the "commit point reached" feedback messages that normally appear on the screen.

• ERRORS - Suppresses the data error messages in the log file that occur when a record generates an Oracle error that causes it to be written to the bad file. A count of rejected records still appears.

• DISCARDS - Suppresses the messages in the log file for each record written to the discard file.

• PARTITIONS - Disables writing the per-partition statistics to the log file during a direct load of a partitioned table.

• ALL - Implements all of the suppression values: HEADER, FEEDBACK, ERRORS, DISCARDS, and PARTITIONS.

SKIP (records to skip)

Default: No records are skipped.

SKIP specifies the number of logical records from the beginning of the file that should not be loaded.

This parameter continues loads that have been interrupted for some reason. It is used for all conventional loads, for single-table direct loads, and for multiple-table direct loads when the same number of records was loaded into each table. It is not used for multiple-table direct loads when a different number of records was loaded into each table.

If a WHEN clause is also present and the load involves secondary data, the secondary data is skipped only if the WHEN clause succeeds for the record in the primary data file.

SKIP_INDEX_MAINTENANCE

Default: false

The SKIP_INDEX_MAINTENANCE parameter stops index maintenance for direct path loads but does not apply to conventional path loads. It causes the index partitions that would have had index keys added to them to be marked Index Unusable instead, because the index segment is inconsistent with respect to the data it indexes. Index segments that are not affected by the load retain the Index Unusable state they had prior to the load.

The SKIP_INDEX_MAINTENANCE parameter:

• Applies to both local and global indexes

• Can be used (with the PARALLEL parameter) to do parallel loads on an object that has indexes

• Can be used (with the PARTITION parameter on the INTO TABLE clause) to do a single partition load to a table that has global indexes

• Puts a list (in the SQL*Loader log file) of the indexes and index partitions that the load set into Index Unusable state

SKIP_UNUSABLE_INDEXES

Default: The value of the Oracle database configuration parameter, SKIP_UNUSABLE_INDEXES, as specified in the initialization parameter file. The default database setting is TRUE.

Both SQL*Loader and the Oracle database provide a SKIP_UNUSABLE_INDEXES parameter. The SQL*Loader SKIP_UNUSABLE_INDEXES parameter is specified at the SQL*Loader command line. The Oracle database SKIP_UNUSABLE_INDEXES parameter is specified as a configuration parameter in the initialization parameter file. It is important to understand how they affect each other.

If you specify a value for SKIP_UNUSABLE_INDEXES at the SQL*Loader command line, it overrides the value of the SKIP_UNUSABLE_INDEXES configuration parameter in the initialization parameter file.

If you do not specify a value for SKIP_UNUSABLE_INDEXES at the SQL*Loader command line, then SQL*Loader uses the database setting for the SKIP_UNUSABLE_INDEXES configuration parameter, as specified in the initialization parameter file. If the initialization parameter file does not specify a database setting for SKIP_UNUSABLE_INDEXES, then the default database setting is TRUE.

A value of TRUE for SKIP_UNUSABLE_INDEXES means that if an index in an Index Unusable state is encountered, it is skipped and the load operation continues. This allows SQL*Loader to load a table with indexes that are in an Unusable state prior to the beginning of the load. Indexes that are not in an Unusable state at load time will be maintained by SQL*Loader. Indexes that are in an Unusable state at load time will not be maintained but will remain in an Unusable state at load completion.

The SKIP_UNUSABLE_INDEXES parameter applies to both conventional and direct path loads.

STREAMSIZE

Default: To see the default value for this parameter, invoke SQL*Loader without any parameters, as described in Invoking SQL*Loader.

Specifies the size, in bytes, for direct path streams.

USERID (username/password)

Default: none

USERID is used to provide your Oracle username/password. If it is omitted, you are prompted for it. If only a slash is used, USERID defaults to your operating system login.

If you connect as user SYS, you must also specify AS SYSDBA in the connect string. For example:

sqlldr \'SYS/password AS SYSDBA\' sample.ctl