11 SQLJ in the Server

Database Connections Within the Server

The concept of connecting to a server is different when your SQLJ code is running within the server itself. There is no explicit database connection. By default, an implicit channel to the database is used for any Java program running in the server. You do not have to initialize this connection, as it is automatically initialized for SQLJ programs. You do not have to register or specify a driver, create a connection instance, specify a default connection context, specify any connection objects for any of your #sql statements, or close the connection.

DefaultContext.setDefaultContext(null);

Coding Issues Within the Server

There are a few coding issues to consider when your code will run within the target server using the server-side internal driver. Note the following:

• Result sets issued by the internal driver persist across calls, and their finalizers do not release their cursors. Because of this, it is especially important to close all iterators to avoid running out of available cursors, unless you have a particular reason for keeping an iterator open, such as when it is actually used across calls.

• The internal driver does not support auto-commit functionality, so the auto-commit setting is ignored within the server. Use explicit COMMIT or ROLLBACK statements to implement or cancel your data updates:

  #sql { COMMIT };
  ...
  #sql { ROLLBACK };
  

• For ISO standard code generation, if you use SQLJ code that interacts with JDBC code and you use a nondefault connection context instance, then you must eventually close the connection context instance in order to clean up statements cached there, unless you use the same connection context instance for the duration of your session. Following is an example:

  DefaultContext ctx = new DefaultContext(conn); // conn is JDBC connection
  #sql [ctx] { SQL operation };
  ...
  ctx.close(sqlj.runtime.ConnectionContext.KEEP_CONNECTION);
  ...
  

  If you do not close the connection context instance, you are likely to run out of statement handles in your session. Also be aware that simply closing the underlying JDBC connection object does not reclaim statement handles, which differs from the behavior when the application executes on a client.

  For the default Oracle-specific code generation, statements are cached in the underlying JDBC statement cache and can be automatically reclaimed.

• With Oracle-specific code generation for code that will run in the server, use an explicit ExecutionContext instance. This ensures that your application can fully interoperate with applications translated with ISO standard SQLJ code generation.

  If you use one thread per connection, which translates to one thread per Oracle session, it is sufficient to use one static instance, as in the following example:

  public static ExecutionContext ec = new ExecutionContext();
  ...
  #sql [ec] { SQL operation };    // use ec for all operations
  

  If you use multiple threads per connection, then you must use a separate execution context instance for each method invocation.

Default Output Device in the Server

The default standard output device in Oracle Java virtual machine (JVM) is the current trace file. If you want to reroute all standard output from a program executing in the server, for example, output from any System.out.println() calls, to a user screen, then you can execute the SET_OUTPUT() procedure of the DBMS_JAVA package as in the following example. Input the buffer size in bytes (10,000 bytes in this case).

sqlplus> execute dbms_java.set_output(10000);

Output exceeding the buffer size will be lost.

If you want your code executing in the server to expressly write output to the user screen, then you can also use the PL/SQL DBMS_OUTPUT.PUT_LINE() procedure instead of the Java System.out.println() method. The PUT_LINE() procedure is overloaded, accepting either VARCHAR2, NUMBER, or DATE as input to specify what is printed.

Name Resolution in the Server

Class loading and name resolution in the server follow a very different paradigm than on a client, because the environments themselves are very different.

Java name resolution in Oracle JVM includes the following:

• Class resolver specs, which are schema lists to search in resolving a class schema object (functionally equivalent to the classpath on a client)

• The resolver, which maintains mappings between class schema objects that reference each other in the server

A class schema object must be resolved before Java objects of the class can be instantiated or methods of the class can be executed. A class schema object is said to be resolved when all of its external references to Java names are bound. In general, all the classes of a Java program should be compiled or loaded before they can be resolved. This is because Java programs are typically written in multiple source files that can reference each other recursively.

When all the class schema objects of a Java program in the server are resolved and none of them have been modified since being resolved, the program is effectively prelinked and ready to run.

SQL Names Versus Java Names

SQL names, such as names of source, class, and resource schema objects, are not global in the way that Java names are global. The Java Language Specification (JLS) directs that package names use Internet naming conventions to create globally unique names for Java programs. By contrast, a fully qualified SQL name is interpreted only with respect to the current schema and database. For example, the SCOTT.FIZZ name in one database does not necessarily denote the same program as SCOTT.FIZZ in another database. In fact, SCOTT.FIZZ in one database can even call SCOTT.FIZZ in another database.

Because of this inherent difference, SQL names must be interpreted and processed differently than Java names. SQL names are relative names and are interpreted from the point of view of the schema where a program is executed. This is central to how the program binds local data stored at that schema. Java names are global names, and the classes that they designate can be loaded at any execution site, with reasonable expectation that those classes will be classes that were used to compile the program.

Loading Classes and Resources into the Server

Once you run the translator on the client, use the Oracle loadjava client-side utility to load class and resource files into schema objects in the server. Either specify the class and resource files, if any, individually on the loadjava command line, or put them into a JAR file and specify the JAR file on the command line. A separate schema object is created for each .class or .ser file in the JAR file or on the command line.

Consider an example where you do the following:

1. Translate and compile Foo.sqlj, which includes an iterator declaration for MyIter, using ISO standard code generation.

2. Enable the -ser2class option when you translate Foo.sqlj.

3. Archive the resulting files, Foo.class, MyIter.class, Foo_SJProfileKeys.class, and Foo_SJProfile0.class, into Foo.jar.

Then run loadjava with the following command line (plus any options you want to specify). This examples uses the default JDBC Oracle Call Interface (OCI) driver:

% loadjava -user scott Foo.jar
Password: password

Alternatively, you can use the original files:

% loadjava -user scott Foo.class MyIter.class Foo_SJProfileKeys.class Foo_SJProfile0.class
Password: password

or:

% loadjava -user scott Foo*.class MyIter.class
Password: password

You can use the JDBC Thin driver for loading as follows (specifying the -thin option and an appropriate URL):

% loadjava -thin -user scott@localhost:1521/myservice Foo.jar
Password: password

Although the loadjava utility is recommended for loading your SQLJ and Java applications into the server, you can also use SQL CREATE JAVA commands such as the following:

CREATE OR REPLACE <AND RESOLVE> JAVA CLASS <NAMED name>;

CREATE OR REPLACE JAVA RESOURCE <NAMED name>;

Naming of Loaded Class and Resource Schema Objects

This section discusses how schema objects for classes and profiles are named when you load classes and profiles into the server. However, remember that profiles are created only for ISO standard code generation.

For ISO standard code generation, if the SQLJ -ser2class option was enabled when you translated your application on the client, then profiles were converted to .class files and will be loaded into class schema objects in the server. If -ser2class was not enabled, then profiles were generated as .ser serialized resource files and will be loaded into resource schema objects in the server.

In the following discussion, it is assumed that you use only the default connection context class for any application that will run in the server. Therefore, there will be only one profile.

Full Names and Short Names

There are two forms of schema object names in the server, full names and short names. Full names are fully qualified and are used as the schema object names whenever possible. If any full name is longer than 31 characters, however, or contains characters that are illegal or cannot be converted to characters in the database character set, then Oracle Database 11g converts the full name to a short name to use as the name of the schema object, keeping track of both names and how to convert between them. If the full name is 31 characters or less and has no illegal or inconvertible characters, then the full name is used as the schema object name.

For more information about these and about other file naming considerations, including DBMS_JAVA procedures to retrieve a full name from a short name, and vice versa, refer to Oracle Database Java Developer's Guide

Full Names of Loaded Classes

Loaded classes will include profile files if you use ISO standard code generation and enable the -ser2class flag. The full name of the class schema object produced when you load a .class file into the server is determined by the package and class name in the original source code. Any path information you supply on the command line or in the JAR file is irrelevant in determining the name of the schema object. For example, if Foo.class consists of the Foo class, which was specified in the source code as being in the x.y package, then the full name of the resulting class schema object is as follows:

x/y/Foo

Note that the .class extension is dropped.

If Foo.sqlj declares an iterator, MyIter, then the full name of its class schema object is as follows (unless it is a nested class, in which case it will not have its own schema object):

x/y/MyIter

Furthermore, if you are using ISO standard code generation:

• The related profile-keys class file, generated by SQLJ when you translate Foo.sqlj, is Foo_SJProfileKeys.class. Therefore, the full name of its class schema object is:

  x/y/Foo_SJProfileKeys
  

• If the -ser2class option was enabled when you translated your application, then the resulting profile is generated in Foo_SJProfile0.class. Therefore, the full name of the class schema object is:

  x/y/Foo_SJProfile0
  

Full Names of Loaded Resources

This discussion is relevant only if you are using ISO standard code generation and did not enable the -ser2class option when you translated your application, or if you use other Java serialized resource (.ser) files in your application.

The naming of resource schema objects is handled differently from class schema objects. Their names are not determined from the contents of the resources. Instead, their full names are identical to the names that appear in a JAR file or on the loadjava command line, including path information. Also note that the .ser extension is not dropped.

It is important to note that because resource names are used to locate the resources at run time. Their names must include the correct path information. In the server, the correct full name of a resource is identical to the relative path and file name that Java would use to look it up on a client.

In the case of a SQLJ profile, this is a subdirectory under the directory specified by the translator -d option, according to the package name. If the -d option, used to specify the top-level output directory for generated .class and .ser files, is set to /mydir and the application is in the abc.def package, then .class and .ser files generated during translation will be placed in the /mydir/abc/def directory.

At run time, /mydir would presumably be in your classpath and Java will look for your application components in the abc/def directory underneath it. Therefore, when you load this application into the server, you must run loadjava or jar from the -d directory so that the path you specify on the command line to find the files also indicates the package name, as follows (where % is the system prompt):

% cd /mydir
% loadjava <...options...> abc/def/*.class abc/def/*.ser

Alternatively, to use a JAR file:

% cd /mydir
% jar -cvf myjar.jar abc/def/*.class abc/def/*.ser
% loadjava <...options...> myjar.jar

If your application is App and your profile is App_SJProfile0.ser, then either of the preceding examples will correctly result in the following full name of the created resource schema object:

abc/def/App_SJProfile0.ser

Note that .ser is retained.

Note also that if you set -d to a directory whose hierarchy has no other contents (which is advisable), you can simply run the JAR utility as follows to recursively get your application components:

% cd /mydir
% jar -cvf myjar.jar *
% loadjava <...options...> myjar.jar

Publishing the Application After Loading Class and Resource Files

Before using your SQLJ code in the server, you must publish the top-level methods, as is true of any Java code you use in the server. Publishing includes writing call descriptors, mapping data types, and setting parameter modes.

Summary: Running a Client Application in the Server

This section summarizes the typical steps of running a client application in the server. As an example, it uses a demo application called NamedIterDemo.

1. Create a JAR file for your application components. For NamedIterDemo, the components include SalesRec.class as well as the application class and profile, if any.

   You can create JAR file niter-server.jar as follows:

   % jar cvf niter-server.jar Named*.class Named*.ser SalesRec.class connect.properties
   

   But remember that .ser files are only relevant for ISO standard code generation.

2. Load the JAR file into the server.

   Use loadjava as follows. This example instructs loadjava to use the OCI driver in loading the files. The -resolve option results in the class files being resolved.

   % loadjava -oci -resolve -force -user scott niter-server.jar
   Password: password
   

3. Create a SQL wrapper in the server for your application.

   For example, run a SQL*Plus script that executes the following:

   set echo on
   set serveroutput on
   set termout on
   set flush on
   
   execute dbms_java.set_output(10000);
   
   create or replace procedure SQLJ_NAMED_ITER_DEMO as language java 
   name 'NamedIterDemo.main (java.lang.String[])';
   /
   

   The DBMS_JAVA.SET_OUTPUT() routine reroutes default output to your screen, instead of to a trace file. The input parameter is the buffer size in bytes.

4. Execute the wrapper.

   For example:

   sqlplus> call SQLJ_NAMED_ITER_DEMO();
   

Loading SQLJ Source Code into the Server

Use the Oracle loadjava client-side utility on a .sqlj file, instead of on a .class file, to load source into the server. If you enable the loadjava -resolve option in loading a .sqlj file, then the server-side embedded translator is run to perform the translation and compilation of your application as it is loaded. Otherwise, the source is loaded into a source schema object without any translation. However, in this case, the source is implicitly translated and compiled the first time an attempt is made to use a class defined in the source. Such implicit translation might seem surprising at first, because there is nothing comparable in client-side SQLJ.

For example, run loadjava as follows from the system prompt:

% loadjava -user scott -resolve Foo.sqlj
Password: password

Alternatively, you can use the JDBC Thin driver to load:

% loadjava -thin -user scott@localhost:1521/myservice -resolve Foo.sqlj
Password: password

Either of these will result in appropriate class schema objects being created in addition to the source schema object.

Before running loadjava, however, you must set SQLJ options appropriately. Note that encoding can be set on the loadjava command line, instead of through the server-side SQLJ encoding option, as follows:

% loadjava -user scott -resolve -encoding SJIS Foo.sqlj
Password: password

The loadjava script, which runs the actual utility, is in the bin subdirectory under your ORACLE_HOME directory. This directory should already be in your path once Oracle has been installed.

Although the loadjava utility is recommended for loading your SQLJ and Java applications into the server, you can also use SQL CREATE JAVA commands such as the following:

CREATE OR REPLACE <AND COMPILE> JAVA SOURCE <NAMED srcname> <AS loadname>;

If you specify AND COMPILE for a .sqlj file, then the source is translated and compiled at that time, creating class schema objects as appropriate in addition to the source schema object. Otherwise, it is not translated and compiled. In this case, only the source schema object is created. In this latter case, however, the source is implicitly translated and compiled the first time an attempt is made to use a class contained in the source.

Option Support in the Server Embedded Translator

The following options are available in the server-side SQLJ translator:

• encoding

• online

• debug

This section discusses these options, after leading off with some discussion of fixed settings in server-side SQLJ. There is also discussion of the loadjava utility and its -resolve option.

Fixed Settings in the Server-Side SQLJ Translator

The following settings, supported by SQLJ translator options on a client, are fixed in the server-side translator:

• Both online semantics-checking and offline parsing are enabled in the server by default, equivalent to the default -parse=both setting on a client. You can override this to disable online semantics-checking through the online option, but cannot disable offline parsing.

  See Also:

  "Online Semantics-Checking Versus Offline Parsing"

• Oracle-specific code generation is used in the server, equivalent to the default -codegen=oracle setting on a client. This is a fixed setting.

  See Also:

  "Oracle-Specific Code Generation (No Profiles)"

• Class schema objects created during server-side translation reference line numbers that map to the SQLJ source code. This is equivalent to enabling the -linemap option when you translate on a client.

  See Also:

  "Line-Mapping to SQLJ Source File (-linemap)"

The encoding Option

This option determines any encoding used to interpret your source code when it is loaded into the server. The encoding option is used at the time the source is loaded, regardless of whether it is also compiled. Alternatively, when using loadjava to load your SQLJ application into the server, you can specify encoding on the loadjava command line. Any loadjava command-line setting for encoding overrides this encoding option.

The online Option

A true setting for the online option (the default value) enables online semantics-checking. Semantics-checking is performed relative to the schema in which the source is loaded. You do not specify an exemplar schema, as you do for online-checking on a client. If the online option is set to false, offline checking is performed.

In either case, the default checker is oracle.sqlj.checker.OracleChecker, which will choose an appropriate checker according to your JDBC driver version and Oracle version.

The online option is used at the time the source is translated and compiled. If you load it with the loadjava -resolve option enabled, then this will occur immediately. Otherwise it will occur the first time an attempt is made to use a class defined in the source, resulting in implicit translation and compilation.

The debug Option

Setting this option to true instructs the server-side Java compiler to output debugging information when a .sqlj or .java source file is compiled in the server. This is equivalent to using the -g option when running the standard javac compiler on a client.

Source is compiled during loading if you use the loadjava -resolve option, right after SQLJ translation in the case of a .sqlj file. If you do not use the -resolve option, then implicit translation and compilation occurs the first time an attempt is made to use a class defined in the source.

Setting SQLJ Options in the Server

There is no command line and there are no properties files when running the SQLJ translator in the server. Information about translator and compiler options is held in each schema in a table named JAVA$OPTIONS. Manipulate options in this table through the following functions and procedures of the DBMS_JAVA package:

• DBMS_JAVA.GET_COMPILER_OPTION()

• DBMS_JAVA.SET_COMPILER_OPTION()

• DBMS_JAVA.RESET_COMPILER_OPTION()

Use set_compiler_option() to specify separate option settings for individual packages or sources. It takes the following as input, with each parameter enclosed by single-quotes:

• Package name, using dotted names, or source name

  Specify this as a full name, not a short name. If you specify a package name, then the option setting applies to all sources in that package and subpackages, except where you override the setting for a particular subpackage or source.

• Option name

• Option setting

Execute the DBMS_JAVA routines using SQL*Plus as follows:

sqlplus> execute dbms_java.set_compiler_option('x.y', 'online', 'true');
sqlplus> execute dbms_java.set_compiler_option('x.y.Create', 'online', 'false');

These two commands enable online checking for all sources in the x.y package, then override that for the Create source by disabling online checking for that particular source.

Similarly, set encoding for the x.y package to SJIS as follows:

sqlplus> execute dbms_java.set_compiler_option('x.y', 'encoding', 'SJIS');

Server-Side Option Notes

Be aware of the following:

• The set_compiler_option() parameter for package and source names uses dotted names, such as abc.def as a package name, even though schema object names use slash syntax, such as abc/def as a package name.

• When you specify a package name, be aware that the option will apply to any included packages as well. A setting of a.b.MyPackage sets the option for any source schema objects whose names are of the following form:

  a/b/MyPackage/subpackage/... 
  

• Specifying '' (empty set of single-quotes) as a package name makes the option apply to the root and all subpackages, effectively making it apply to all packages in your schema.

Naming of Loaded Source and Generated Class and Resource Schema Objects

When you use the server-side SQLJ translator, such as when you use loadjava on a .sqlj file with the -resolve option enabled, the output generated by the server-side translator is essentially identical to what would be generated on a client. This output consists of a compiled class for each class you defined in the source and a compiled class for each iterator and connection context class.

As a result, the following schema objects will be produced when you load a .sqlj file into the server with loadjava and have it translated and compiled:

• A source schema object for the original source code

• A class schema object for each class you defined in the source

• A class schema object for each iterator or connection context class you declared in the source

  But presumably you will not need to declare connection context classes in code that will run in the server, unless it is to specify type maps for user-defined types.

The full names of these schema objects are determined as described in the following subsections. Use the loadjava -verbose option for a report of schema objects produced and what they are named.

Full Name of Source

When you load a source file into the server, regardless of whether it is translated and compiled, a source schema object is produced. The full name of this schema object is determined by the package and class names in the source code. Any path information you supply to loadjava on the command line is irrelevant to the determination of the name of the schema object.

For example, if Foo.sqlj defines the Foo class in the x.y package and defines or declares no other classes, then the full name of the resulting source schema object is:

x/y/Foo

Note that the .sqlj extension is dropped.

If you define additional classes or declare iterator or connection context classes, then the source schema object is named according to the first public class definition or declaration encountered, or, if there are no public classes, the first class definition. In the server, there can be more than one public class definition in a single source.

For example, if Foo.sqlj is still in the x.y package, defines public class Bar first and then class Foo, and has no public iterator or connection context class declarations preceding the definition of Bar, then the full name of the resulting source schema object is:

x/y/Bar

However, if the declaration of public iterator class MyIter precedes the Bar and Foo class definitions, then the full name of the resulting source schema object is:

x/y/MyIter

Full Names of Generated Classes

Class schema objects are generated for each class you defined in the source, each iterator you declared, and the profile-keys class. The naming of the class schema objects is based on the class names and the package name from the source code.

This discussion continues the example in "Full Name of Source". Presume your source code specifies the x.y package, defines public class Bar then class Foo, then declares public iterator class MyIter. The full names of the class schema objects for the classes you define and declare are as follows:

x/y/Bar
x/y/Foo
x/y/MyIter

Note that .class is not appended.

The name of the original source file, as well as any path information you specify when loading the source into the server, is irrelevant in determining the names of the generated classes. If you define inner classes or anonymous classes in your code, then they are named according to the conventions of the standard javac compiler.

Error Output from the Server Embedded Translator

SQLJ error processing in the server is similar to general Java error processing in the server. SQLJ errors are directed into the USER_ERRORS table of the user schema. You can SELECT from the TEXT column of this table to get the text of a given error message.

However, if you use loadjava to load your SQLJ source, then loadjava also captures and writes the error messages from the server-side translator.

Informational messages and suppressible warnings are withheld by the server-side translator in a way that is equivalent to the operation of the client-side translator with a -warn=noportable,noverbose setting, which is the default.

Publishing the Application After Loading Source Files

Before using your SQLJ code in the server, you must publish the top-level methods, as is true of any Java code you use in the server. Publishing includes writing call descriptors, mapping data types, and setting parameter modes. For information, refer to Oracle Database Java Developer's Guide.

Java Multithreading in the Server

Programs that use Java multithreading can execute in Oracle Database 11g without modification. However, while client-side programs use multithreading to improve throughput for users, there are no such benefits when Java-multithreaded code runs in the server. If you are considering porting a multithreaded application into the server, be aware of the following important differences in the functionality of multithreading in Oracle JVM, as opposed to in client-side JVMs:

• Threads in the server run sequentially, not simultaneously.

• In the server, threads within a call die at the end of the call.

• Threads in the server are not preemptively scheduled. If one thread goes into an infinite loop, then no other threads can run.

Do not confuse Java multithreading in Oracle Database 11g with general Oracle server multithreading. The latter refers to simultaneous Oracle sessions, not Java multithreading. In the server, scalability and throughput are gained by having many individual users, each with his own session, executing simultaneously. The scheduling of Java execution for maximum throughput, such as for each call within a session, is performed by Oracle Database 11g, and not by Java.

Recursive SQLJ Calls in the Server

SQLJ generally does not allow multiple SQLJ statements to use the same execution context instance simultaneously. Specifically, a statement trying to use an execution context instance that is already in use will be blocked until the first statement completes.

However, this functionality would be less desirable in the server than on a client. This is because different stored procedures or functions, which all typically use the default execution context instance, can inadvertently try to use this same execution context instance simultaneously in recursive situations. For example, one stored procedure might use a SQLJ statement to call another stored procedure that uses SQLJ statements. When these stored procedures are first created, there is probably no way of knowing when such situations might arise, so it is doubtful that particular execution context instances are specified for any of the SQLJ statements.

To address this situation, SQLJ does allow multiple SQLJ statements to use the same execution context instance simultaneously if this results from recursive calls.

Consider an example of a recursive situation to see what happens to status information in the execution context instance. Presume that all statements use the default connection context instance and its default execution context instance. If stored procedure proc1 has a SQLJ statement that calls stored procedure proc2, which also has SQLJ statements, then the statements in proc2 will each be using the execution context instance while the procedure call in proc1 is also using it.

Each SQLJ statement in proc2 results in status information for that statement being written to the execution context instance, with the opportunity to retrieve that information after completion of each statement, as desired. The status information from the statement in proc1 that calls proc2 is written to the execution context instance only after proc2 has finished executing, program flow has returned to proc1, and the operation in proc1 that called proc2 has completed.

To avoid confusion about execution context status information in recursive situations, execution context methods are carefully defined to update status information about a SQL operation only after the operation has completed.

Verifying that Code is Running in the Server

A convenient way to verify that your code is actually running in the server is to use the static getProperty() method of the java.lang.System class to retrieve the oracle.server.version Java property. If this property contains a version number, then you are running in the server. If it is null, then you are not. Here is an example:

...
if (System.getProperty("oracle.server.version") != null 
{
   // (running in server)
}
...