How Transaction Processing Affects Record Locking (126870)

MORE INFORMATION

When you issue a BEGIN TRANSACTION command, the index of the currently open table is locked. Only the current user is able to update, add, or delete records in the table while the transaction is in progress.

Any command or function that normally places a lock will continue to do so. Automatic locks are released automatically when the transaction is completed. Either a ROLLBACK or END TRANSACTION command must be used to end the transaction. Manual locks - those placed with LOCK(), RLOCK() or FLOCK() - must be manually released after the transaction is completed.

This behavior is by design, and is required to ensure the basic functionality of transaction processing.

The following coding techniques can be used to minimize contention for record locks:

• Open all tables before issuing the BEGIN TRANSACTION command. This prevents the entire table from being locked as soon as it is opened.
• Buffering should be enabled and all edits performed before the transaction starts. The BEGIN TRANSACTION command should be issued immediately before a TABLEUPDATE() or a SQLCOMMIT() function. Optimistic row buffering provides the shortest locking time of all buffering types. However, table buffering must be used if you wish to rollback multiple transactions.
• Avoid the use of commands with a FOR scope within the confines of a transaction, since these commands lock the entire database. If you must use commands such as REPLACE within a transaction, use a SCAN loop to perform individual commands instead.
• Close the transaction as soon as possible after the update has been applied to the table.