Cursor (databases)

In database packages, a cursor comprises a control structure for the successive traversal (and potential processing) of records in a result set.

Cursors provide a mechanism by which a database client iterates over the records in a database. Using cursors, the client can get, put, and delete database records. Database programmers use cursors for processing individual rows returned by the database system for a query. Cursors address the problem of impedance mismatch, an issue that occurs in many programming languages^{[citation needed]}. Most procedural programming languages do not offer any mechanism for manipulating whole result-sets at once. In this scenario, the application must process rows in a result-set sequentially. Thus one can think of a database cursor as an iterator over the collection of rows in the result set.

Several SQL statements do not require the use of cursors. That includes the INSERT statement, for example, as well as most forms of the DELETE and UPDATE statements. Even a SELECT statement may not involve a cursor if it is used in the variation of SELECT INTO. A SELECT INTO retrieves at most a single row directly into the application.

Working with cursors

This section introduces the ways the SQL:2003 standard defines how to use cursors in applications in embedded SQL. Not all application bindings for relational database systems adhere to that standard, and some (such as CLI or JDBC) use a different interface.

A programmer makes a cursor known to the DBMS by using a DECLARE ... CURSOR statement and assigning the cursor a (compulsory) name:

 DECLARE cursor_name CURSOR FOR SELECT ... FROM ...

Before code can access the data, it must open the cursor with the OPEN statement. Directly following a successful opening, the cursor is positioned before the first row in the result set.

 OPEN cursor_name

Programs position cursors on a specific row in the result set with the FETCH statement. A fetch operation transfers the data of the row into the application.

 FETCH cursor_name INTO ...

Once an application has processed all available rows or the fetch operation is to be positioned on a non-existing row (compare scrollable cursors below), the DBMS returns a SQLSTATE '02000' (usually accompanied by an SQLCODE +100) to indicate the end of the result set.

The final step involves closing the cursor using the CLOSE statement:

 CLOSE cursor_name

After closing a cursor, a program can open it again, which implies that the DBMS re-evaluates the same query or a different query and builds a new result-set.

Scrollable cursors

Programmers may declare cursors as scrollable or not scrollable. The scrollability indicates the direction in which a cursor can move.

With a non-scrollable cursor, also known as forward-only, one can FETCH each row at most once, and the cursor automatically moves to the immediately following row. A fetch operation after the last row has been retrieved positions the cursor after the last row and returns SQLSTATE 02000 (SQLCODE +100).

A program may position a scrollable cursor anywhere in the result set using the FETCH SQL statement. The keyword SCROLL must be specified when declaring the cursor. The default is NO SCROLL, although different language bindings like JDBC may apply different default.

 DECLARE cursor_name sensitivity SCROLL CURSOR FOR SELECT ... FROM ...

The target position for a scrollable cursor can be specified relative to the current cursor position or absolute from the beginning of the result set.

 FETCH [ NEXT | PRIOR | FIRST | LAST ] FROM cursor_name

 FETCH ABSOLUTE n FROM cursor_name

 FETCH RELATIVE n FROM cursor_name

Scrollable cursors can potentially access the same row in the result set multiple times. Thus, data modifications (insert, update, delete operations) from other transactions could have an impact on the result set. A cursor can be SENSITIVE or INSENSITIVE to such data modifications. A sensitive cursor picks up data modifications impacting the result set of the cursor, and an insensitive cursor does not. Additionally, a cursor may be ASENSITIVE, in which case the DBMS tries to apply sensitivity as much as possible.

"WITH HOLD"

Cursors are usually closed automatically at the end of a transaction, i.e when a COMMIT or ROLLBACK (or an implicit termination of the transaction) occurs. That behavior can be changed if the cursor is declared using the WITH HOLD clause. (The default is WITHOUT HOLD.) A holdable cursor is kept open over COMMIT and closed upon ROLLBACK. (Some DBMS deviate from this standard behavior and also keep holdable cursors open over ROLLBACK.)

 DECLARE cursor_name CURSOR WITH HOLD FOR SELECT ... FROM ...

When a COMMIT occurs, a holdable cursor is positioned before the next row. Thus, a positioned UPDATE or positioned DELETE statement will only succeed after a FETCH operation occurred first in the transaction.

Note that JDBC defines cursors as holdable per default. This is done because JDBC also activates auto-commit per default. Due to the usual overhead associated with auto-commit and holdable cursors, both features should be explicitly deactivated at the connection level.

Positioned update/delete statements

Cursors can not only be used to fetch data from the DBMS into an application but also to identify a row in a table to be updated or deleted. The SQL:2003 standard defines positioned update and positioned delete SQL statements for that purpose. Such statements do not use a regular WHERE clause with predicates. Instead, a cursor identifies the row. The cursor must be opened and positioned on a row already using the FETCH statement.

 UPDATE table_name
 SET    ...
 WHERE  CURRENT OF cursor_name

 DELETE
 FROM   table_name
 WHERE  CURRENT OF cursor_name

The cursor must operate on an updatable result set in order to successfully execute a positioned update or delete statement. Otherwise, the DBMS would not know how to apply the data changes to the underlying tables referred to in the cursor.

Cursors in distributed transactions

Using cursors in distributed transactions (X/Open XA Environments), which are controlled using a transaction monitor, is no different than cursors in non-distributed transactions.

One has to pay attention when using holdable cursors, however. Connections can be used by different applications. Thus, once a transaction has been ended and committed, a subsequent transaction (running in a different application) could inherit existing holdable cursors. Therefore, an application developer has to be aware of that situation.

Cursors in XQuery

The XQuery language allows cursors to be created using the subsequence() function.

The format is:

let $displayed-sequence := subsequence($result, $start, $item-count)

Where $result is the result of the initial XQuery, $start is the item number to start and $item-count is the number of items to return.

Equivalently this can also be done using a predicate:

let $displayed-sequence := $result[$start to $end]

Where $end is the end sequence.

For complete examples see the XQuery Wikibook.

Disadvantages of cursors

The following information may vary from database system to database system.

Fetching a row from the cursor may result in a network round trip each time. This uses much more network bandwidth than would ordinarily be needed for the execution of a single SQL statement like DELETE. Repeated network round trips can severely impact the speed of the operation using the cursor. Some DBMSs try to reduce this impact by using block fetch. Block fetch implies that multiple rows are sent together from the server to the client. The client stores a whole block of rows in a local buffer and retrieves the rows from there until that buffer is exhausted.

Cursors allocate resources on the server, for instance locks, packages, processes, temporary storage, etc. For example, Microsoft SQL Server implements cursors by creating a temporary table and populating it with the query's result-set. If a cursor is not properly closed (deallocated), the resources will not be freed until the SQL session (connection) itself is closed. This wasting of resources on the server can not only lead to performance degradations but also to failures.

See also

• Iterator

References

• Christopher J. Date: Database in Depth, O'Reilly & Associates, ISBN 0-596-10012-4
• Thomas M. Connolly, Carolyn E. Begg: Database Systems, Addison-Wesley, ISBN 0-321-21025-5
• Ramiz Elmasri, Shamkant B. Navathe: Fundamentals of Database Systems, Addison-Wesley, ISBN 0-201-54263-3
• Neil Matthew, Richard Stones: Beginning Databases with PostgreSQL: From Novice to Professional, Apress, ISBN 1-59059-478-9
• Thomas Kyte: Expert One-On-One: Oracle, Apress, ISBN 1-59059-525-4
• Kevin Loney: Oracle Database 10g: The Complete Reference, Oracle Press, ISBN 0-07-225351-7

External links

• Descriptions from Portland Pattern Repository
• PostgreSQL Documentation
• Berkeley DB Reference Guide: Cursor operations
• ResultSet in JDBC
• Q3SqlCursor Class Reference
• OCI Scrollable Cursor
• function oci_new_cursor
• MySQL's Cursor Documentation
• Cursors in DB2 CLI applications; Cursors in DB2 SQL stored procedures