2 SQL Server Gateway Features and Restriction

Return Values and Stored Procedures

By default, all stored procedures and functions do not return a return value to the user. To enable return values, set the HS_FDS_PROC_IS_FUNC parameter in the initialization parameter file.

In the following example, the employee name JOHN SMYTHE is passed to the SQL Server stored procedure REVISE_SALARY. The stored procedure retrieves the salary value from the SQL Server database to calculate a new yearly salary for JOHN SMYTHE. The revised salary returned in RESULT is used to update EMP in a table of an Oracle database:

DECLARE
  INPUT VARCHAR2(15);
  RESULT NUMBER(8,2);
BEGIN
  INPUT := 'JOHN SMYTHE';
  RESULT := REVISE_SALARY@MSQL(INPUT);
  UPDATE EMP SET SAL = RESULT WHERE ENAME =: INPUT;
END;
/

The procedural feature automatically converts non-Oracle data types to and from PL/SQL data types.

Result Sets and Stored Procedures

The Oracle Database Gateway for SQL Server provides support for stored procedures which return result sets.

By default, all stored procedures and functions do not return a result set to the user. To enable result sets, set the HS_FDS_RESULTSET_SUPPORT parameter in the initialization parameter file.

When accessing stored procedures with result sets through the Oracle Database Gateway for SQL Server, you will be in the sequential mode of Heterogeneous Services.

The Oracle Database Gateway for SQL Server returns the following information to Heterogeneous Services during procedure description:

• All the input arguments of the remote stored procedure

• None of the output arguments

• One out argument of type ref cursor (corresponding to the first result set returned by the stored procedure)

Client programs have to use the virtual package function dbms_hs_result_set.get_next_result_set to get the ref cursor for subsequent result sets. The last result set returned is the out argument from the procedure.

The limitations of accessing result sets are the following:

• Result sets returned by a remote stored procedure have to be retrieved in the order in which they were placed on the wire

• On execution of a stored procedure, all result sets returned by a previously executed stored procedure will be closed (regardless of whether the data has been completely retrieved or not)

In the following example, the SQL Server stored procedure is executed to fetch the contents of the emp and dept tables from SQL Server:

create procedure REFCURPROC (@arg1 varchar(255), @arg2 varchar(255) output)
as
select @arg2 = @arg1
select * from EMP
select * from DEPT
go

This stored procedure assigns the input parameter arg1 to the output parameter arg2, opens the query SELECT * FROM EMP in ref cursor rc1, and opens the query SELECT * FROM DEPT in ref cursor rc2.

OCIEnv *ENVH;
OCISvcCtx *SVCH;
OCIStmt *STMH;
OCIError *ERRH;
OCIBind *BNDH[3];
OraText arg1[20];
OraText arg2[255];
OCIResult *rset;
OCIStmt *rstmt;
ub2 rcode[3];
ub2 rlens[3];
sb2 inds[3];
OraText *stmt = (OraText *) "begin refcurproc@MSQL(:1,:2,:3); end;";
OraText *n_rs_stm = (OraText *)
  "begin :ret := DBMS_HS_RESULT_SET.GET_NEXT_RESULT_SET@MSQL; end;";

/* Prepare procedure call statement */

/* Handle Initialization code skipped */
OCIStmtPrepare(STMH, ERRH, stmt, strlen(stmt), OCI_NTV_SYNTAX, OCI_DEFAULT);

/* Bind procedure arguments */
inds[0] = 0;
strcpy((char *) arg1, "Hello World");
rlens[0] = strlen(arg1);
OCIBindByPos(STMH, &BNDH[0], ERRH, 1, (dvoid *) arg1, 20, SQLT_CHR,
             (dvoid *) &(inds[0]), &(rlens[0]), &(rcode[0]), 0, (ub4 *) 0, 
             OCI_DEFAULT);

inds[1] = -1;
OCIBindByPos(STMH, &BNDH[1], ERRH, 1, (dvoid *) arg2, 20, SQLT_CHR,
             (dvoid *) &(inds[1]), &(rlens[1]), &(rcode[1]), 0, (ub4 *) 0, 
             OCI_DEFAULT);

inds[2] = 0;
rlens[2] = 0;
OCIDescriptorAlloc(ENVH, (dvoid **) &rset, OCI_DTYPE_RSET, 0, (dvoid **) 0);
OCIBindByPos(STMH, &BNDH[2], ERRH, 2, (dvoid *) rset, 0, SQLT_RSET,
             (dvoid *) &(inds[2]), &(rlens[2]), &(rcode[2]),
             0, (ub4 *) 0, OCI_DEFAULT);

/* Execute procedure */
OCIStmtExecute(SVCH, STMH, ERRH, 1, 0, (CONST OCISnapshot *) 0,
               (OCISnapshot *) 0, OCI_DEFAULT);

/* Convert result set to statement handle */
OCIResultSetToStmt(rset, ERRH);
rstmt = (OCIStmt *) rset;

/* After this the user can fetch from rstmt */
/* Issue get_next_result_set call to get handle to next_result set */
/* Prepare Get next result set procedure call */

OCIStmtPrepare(STMH, ERRH, n_rs_stm, strlen(n_rs_stm), OCI_NTV_SYNTAX,
               OCI_DEFAULT);

/* Bind return value */
OCIBindByPos(STMH, &BNDH[1], ERRH, 1, (dvoid *) rset, 0, SQLT_RSET,
             (dvoid *) &(inds[1]), &(rlens[1]), &(rcode[1]),
             0, (ub4 *) 0, OCI_DEFAULT);

/* Execute statement to get next result set*/
OCIStmtExecute(SVCH, STMH, ERRH, 1, 0, (CONST OCISnapshot *) 0,
               (OCISnapshot *) 0, OCI_DEFAULT);

/* Convert next result set to statement handle */
OCIResultSetToStmt(rset, ERRH);
rstmt = (OCIStmt *) rset;

/* Now rstmt will point to the second result set returned by the
remote stored procedure */

/* Repeat execution of get_next_result_set to get the output arguments */

create or replace package rcpackage is  type RCTYPE is ref cursor;end rcpackage;/

declare
  rc1 rcpackage.rctype;
  rec1 loc_emp%rowtype;
  rc2 rcpackage.rctype;
  rec2 loc_dept%rowtype;
  rc3 rcpackage.rctype;
  rec3 outargs%rowtype;
  out_arg varchar2(255);

begin

  -- Execute procedure
  out_arg := null;
  refcurproc@MSQL('Hello World', out_arg, rc1);

  -- Fetch 20 rows from the remote emp table and insert them into loc_emp
  for i in 1 .. 20 loop
    fetch rc1 into rec1;
    insert into loc_emp (rec1.empno, rec1.ename, rec1.job,
    rec1.mgr, rec1.hiredate, rec1.sal, rec1.comm, rec1.deptno);
  end loop;

  -- Close ref cursor
  close rc1;

  -- Get the next result set returned by the stored procedure
  rc2 := dbms_hs_result_set.get_next_result_set@MSQL;

  -- Fetch 5 rows from the remote dept table and insert them into loc_dept
  for i in 1 .. 5 loop
    fetch rc2 into rec2;
    insert into loc_dept values (rec2.deptno, rec2.dname, rec2.loc);
  end loop;

  --Close ref cursor
  close rc2;

  -- Get the output arguments from the remote stored procedure
  -- Since we are in sequential mode, they will be returned in the
  -- form of a result set
  rc3 := dbms_hs_result_set.get_next_result_set@MSQL;

  -- Fetch them and insert them into the outarguments table
  fetch rc3 into rec3;
  insert into outargs (rec3.outarg, rec3.retval);

  -- Close ref cursor
  close rc3;

end;
/

Implicit Transactions (Chained Mode)

The gateway supports the ANSI-standard implicit transactions. SQL Server stored procedures must be written for this mode. Running implicit transactions allows the gateway to extend the Oracle two-phase commit protection to transactions updating Oracle and SQL Server databases.

Column Definitions

By default, a SQL Server table column cannot contain null values unless NULL is specified in the column definition. SQL Server assumes all columns cannot contain null values unless you set a SQL Server option to override this default.

For an Oracle table, null values are allowed in a column unless NOT NULL is specified in the column definition.

Naming Rules

Naming rule issues include the following:

• Rules for Naming Objects

• Case Sensitivity

SQL> SELECT * FROM "emp"@MSQL;

SQL> SELECT * FROM "Scott"."emp"@MSQL;

SQL> SELECT * FROM SCOTT."emp"@MSQL;

SQL> SELECT * FROM scott."emp"@MSQL;

SQL> CREATE VIEW EMP (EMPNO, ENAME, SAL, HIREDATE)
      AS SELECT "empno", "ename", "sal", "hiredate"
      FROM "emp"@MSQL;

SQL> SELECT EMPNO, ENAME FROM EMP;

Data Types

Data type issues include the following:

• Binary Literal Notation

• Bind Variables With LONG Columns

• Data Type Conversion

SQL> INSERT INTO BINARY_TAB@MSQL VALUES ('0xff')

SELECT DATE_COL FROM TEST@MSQL WHERE DATE_COL = "1-JAN-2004";

SELECT DATE_COL FROM TEST@MSQL WHERE DATE_COL = TO_DATE("1-JAN-2004")

Queries

Query issues include the following:

• Row Selection

• Empty Strings

• Empty Bind Variables

SELECT * from "emp"@MSQL where "ename" IS NULL;

Locking

The locking model for an SQL Server database differs significantly from the Oracle model. The gateway depends on the underlying SQL Server behavior, so Oracle applications that access SQL Server through the gateway can be affected by the following possible scenarios:

• Read access might block write access

• Write access might block read access

• Statement-level read consistency is not guaranteed

  See Also:

  SQL Server documentation for information about the SQL Server locking model.

Multiple Open Statements

Accessing SQL Server has the limitation that one open statement or cursor is allowed for each connection. If a second statement or cursor needs to open in the same transaction to access SQL Server, it requires a new connection.

Because of this limitation multiple open statements or cursors within the same transaction can lock each other because they use different connections to SQL Server.

To avoid this restriction, issue a commit, or modify the logic, or both.

Transactional Integrity

The gateway cannot guarantee transactional integrity in the following cases:

• When a statement that is processed by the gateway causes an implicit commit in the target database

• When the target database is configured to work in autocommit mode

  Note:

  Oracle strongly recommends the following:

  • If you know that executing a particular statement causes an implicit commit in the target database, then ensure that this statement is executed in its own transaction.

The gateway sets Autocommit Mode to Off when a connection is established to the SQL Server database.

Transaction Capability

The gateway does not support savepoints. If a distributed update transaction is under way involving the gateway, and a user attempts to create a savepoint, the following error occurs:

ORA-02070: database dblink does not support savepoint in this context

By default, the gateway is configured as COMMIT_CONFIRM.

COMMIT or ROLLBACK in PL/SQL Cursor Loops Closes Open Cursors

Any COMMIT or ROLLBACK issued in a PL/SQL cursor loop closes all open cursors, which can result in the following error:

ORA-1002:  fetch out of sequence 

To prevent this error, move the COMMIT or ROLLBACK statement outside the cursor loop.

Stored Procedures

Changes issued through stored procedures that embed commits or rollbacks cannot be controlled by the Oracle transaction manager or Oracle COMMIT or ROLLBACK commands.

When accessing stored procedures with result sets through the Oracle Database Gateway for SQL Server, you must work in the sequential mode of Heterogeneous Services.

When accessing stored procedures with multiple result sets through the Oracle Database Gateway for SQL Server, you must read all the result sets before continuing.

Output parameters of stored procedures must be initialized to an empty string.

Pass-Through Feature

If the SQL statements being passed through the gateway result in an implicit commit at the SQL Server database, the Oracle transaction manager is unaware of the commit and an Oracle ROLLBACK command cannot be used to roll back the transaction.

DDL Statements

SQL Server requires some DDL statements to be executed in their own transaction, and only one DDL statement can be executed in a given transaction.

If you use these DDL statements in a SQL Server stored procedure and you execute the stored procedure through the gateway using the procedural feature, or, if you execute the DDL statements through the gateway using the pass-through feature, an error condition might result. This is because the procedural feature and the pass-through feature of the gateway cannot guarantee that the DDL statements are executed in their own separate transaction.

The following SQL Server DDL statements can cause an error condition if you attempt to pass them with the gateway pass-through feature, or if you execute a SQL Server stored procedure that contains them:

SQL Syntax

This section lists restrictions on the following SQL syntax:

• WHERE CURRENT OF Clause

• CONNECT BY Clause

• Functions in Subqueries

• Parameters in Subqueries

• Data Dictionary Table and Views in UPDATE Statement

• ROWID

• TO_DATE

• EXPLAIN PLAN Statement

• Callback Support

Functions

The following restrictions apply to using functions:

• Unsupported functions cannot be used in statements that refer to LONG columns.

• When negative numbers are used as the second parameter in a SUBSTR function, incorrect results are returned. This is due to incompatibility between the Oracle SUBSTR function and the equivalent in SQL Server.

SQL*Plus COPY Command with Lowercase Table Names

You need to use double quotes to wrap around lowercase table names.

For example:

copy from tkhouser/tkhouser@inst1 insert loc_tkhodept using select * from "tkhodept"@holink2;

Database Links

The gateway is not multithreaded and cannot support shared database links. Each gateway session spawns a separate gateway process and connections cannot be shared.

Encrypted Format Login

The Oracle9i database (Release 9.2 and earlier) supported an Oracle initialization parameter, DBLINK_ENCRYPT_LOGIN. When this parameter is set to TRUE, the password for the login user ID is not sent over the network.

If this parameter is set to TRUE in the initialization parameter file used by the Oracle9i database, you must change the setting to FALSE, the default setting, to allow Oracle9i to communicate with the gateway.

In the current release, Oracle Database 11g, Release 11.1, the DBLINK_ENCRYPT_LOGIN initialization parameter is obsolete, so you need not check it.

Date Arithmetic

The following SQL expressions do not function correctly with the gateway:

date + number
number + date
date - number
date1 - date2

Statements with the preceding expressions are sent to the SQL Server database without any translation. Since SQL Server does not support these date arithmetic functions, the statements return an error.

SQL Server IMAGE, TEXT and NTEXT Data Types

The following restrictions apply when using IMAGE, TEXT, and NTEXT data types:

• An unsupported SQL function cannot be used in a SQL statement that accesses a column defined as SQL Server data type IMAGE, TEXT, or NTEXT.

• You cannot use SQL*Plus to select data from a column defined as SQL Server data type IMAGE, TEXT, or NTEXT when the data is greater than 80 characters in length. Oracle recommends using Pro*C or Oracle Call Interface to access such data in a SQL Server database.

• IMAGE, TEXT, and NTEXT data cannot be read through pass-through queries.

• If a SQL statement is accessing a table including an IMAGE, TEXT, or NTEXT column, the statement will be sent to SQL Server as two separate statements. One statement to access the IMAGE, TEXT or NTEXT column, and a second statement for the other columns in the original statement. This will result in two connections to SQL Server due to a limitation in the Microsoft ODBC driver which only allows one statement for each connection, which can cause a hang depending on the sequence of SQL statements. If this happens, try issuing a commit and separating the statements in different transactions.

The gateway does not support the PL/SQL function COLUMN_VALUE_LONG of the DBMS_SQL package.

String Functions

If you concatenate numeric literals using the "||" or CONCAT operator when using the gateway to query a SQL Server database, the result is an arithmetic addition. For example, the result of the following statement is 18:

SQL> SELECT 9 || 9 FROM DUAL@MSQL;

The result is 99 when using Oracle to query an Oracle database.

Schema Names and PL/SQL

If you do not prefix a SQL Server database object with its schema name in a SQL statement within a PL/SQL block, the following error message occurs:

ORA-6550 PLS-201 Identifier table_name must be declared.

Change the SQL statement to include the schema name of the object.

Data Dictionary Views and PL/SQL

You cannot refer to data dictionary views in SQL statements that are inside a PL/SQL block.

Stored Procedures

Return values of stored procedures which return result sets are incorrect.