Release 2 (9.2)
Part Number A96584-01
 Home |
 Book List |
 Contents |
 Index |
 Master Index |
 Feedback |
| Oracle Call Interface Programmer's Guide Release 2 (9.2) Part Number A96584-01 |
|
This chapter introduces advanced programming topics, including the following:
The thread safety feature of the Oracle database server and OCI libraries allows developers to use the OCI in a multithreaded environment. With thread safety, OCI code can be reentrant, with multiple threads of a user program making OCI calls without side effects from one thread to another.
|
Note: Thread safety is not available on every platform. Check your Oracle system-specific documentation for more information. |
The following sections describe how you can use the OCI to develop multithreaded applications.
The implementation of thread safety in the Oracle Call Interface has the following benefits and advantages:
In addition to client-server applications, where the client can be a multithreaded program, a typical use of multithreaded applications is in three-tier (also called client-agent-server) architectures. In this architecture the client is concerned only with presentation services. The agent (or application server) processes the application logic for the client application. Typically, this relationship is a many-to-one relationship, with multiple clients sharing the same application server.
The server tier in this scenario is an Oracle database. The applications server (agent) is very well suited to being a multithreaded application server, with each thread serving a client application. In an Oracle environment this application server is an OCI or precompiler program.
Threads are lightweight processes that exist within a larger process. Threads share the same code and data segments but have their own program counters, machine registers, and stack. Global and static variables are common to all threads, and a mutual exclusivity mechanism may be required to manage access to these variables from multiple threads within an application.
Once spawned, threads run asynchronously to one another. They can access common data elements and make OCI calls in any order. Because of this shared access to data elements, a mechanism is required to maintain the integrity of data being accessed by multiple threads.
The mechanism to manage data access takes the form of mutexes (mutual exclusivity locks), that ensure that no conflicts arise between multiple threads that are accessing shared resources within an application. In OCI, mutexes are granted on a per-environment-handle basis.
In order to take advantage of thread safety, an application must be running on a thread-safe platform. Then the application must tell the OCI layer that the application is running in multithreaded mode, by specifying OCI_THREADED for the mode parameter of the opening call to OCIEnvCreate().
Once OCIEnvCreate() is called with OCI_THREADED, all subsequent calls to OCIEnvCreate() must also be made with OCI_THREADED.
|
Note: Applications running on non-thread-safe platforms should not pass a value of OCI_THREADED to |
If an application is single-threaded, whether or not the platform is thread-safe, the application should pass a value of OCI_DEFAULT to OCIInitialize() or OCIEnvCreate(). Single-threaded applications which run in OCI_THREADED mode may incur performance hits.
If a multithreaded application is running on a thread-safe platform, the OCI library will manage mutexing for the application on a per-environment-handle basis. If the application programmer desires, this application can override this feature and maintain its own mutexing scheme. This is done by specifying a value of OCI_NO_MUTEX to the OCIEnvCreate() call.
The following three scenarios are possible, depending on how many connections exist in each environment handle, and how many threads will be spawned in each connection.
mode of OCIEnvCreate(). In this case the application must mutex OCI calls by made on the same environment handle itself. This has the advantage that the mutexing scheme can be optimized based on the application design. The programmer must also insure that only one OCI call is in process on the environment handle connection at any given time.OCIEnvCreate(). In this case, the OCI library automatically gets a mutex on every OCI call on the same environment handle.
If an application is mixing later release and 7.x OCI calls, and the application has been initialized as thread-safe (with the appropriate calls of the later release), it is not necessary to call opinit() to achieve thread safety. The application will get 7.x behavior on any subsequent 7.x function calls.
See cdemothr.c in the demo directory for an example of a multithreading application.
The OCIThread package provides a number of commonly used threading primitives for use by Oracle customers. It offers a portable interface to threading capabilities native to various platforms. It does not implement threading on platforms that do not have native threading capability.
OCIThread does not provide a portable implementation of multithreaded facilities. It only serves as a set of portable covers for native multithreaded facilities. Therefore, platforms that do not have native support for multithreading will only be able to support a limited implementation of OCIThread. As a result, products that rely on all of OCIThread's functionality will not port to all platforms. Products that must port to all platforms must use only a subset of OCIThread's functionality. This issue is discussed further in later sections of this document.
The OCIThread API is split into three main parts. Each part is described briefly here. The following subsections describe each in greater detail.
See Also:
|
These calls deal with the initialization and termination of OCIThread. Initialization of OCIThread initializes the OCIThread context that is a member of the OCI environment or user session handle. This context is required for other OCIThread calls.
The passive threading primitives include primitives to manipulate mutual exclusion (mutex) locks, thread ID's, and thread-specific data keys.
The reason that these primitives are described as passive is that while their specifications allow for the existence of multiple threads, they do not require it. This means that it is possible for these primitives to be implemented according to specification in both single-threaded and multithreaded environments.
As a result, OCIThread clients that use only these primitives will not require the existence of multiple threads in order to work correctly, that is, they will be able to work in single-threaded environments without branching code.
Active threading primitives include primitives dealing with the creation, termination, and other manipulation of threads.
The reason that these primitives are described as active is that they can only be used in true multithreaded environments. Their specifications explicitly require that it be possible to have multiple threads. If you need to determine at runtime whether or not you are in a multithreaded environment, call OCIThreadIsMulti() before calling an OCIThread active primitive.
The types and functions described in this section are associated with the initialization and termination of the OCIThread package. OCIThread must be properly initialized before any of its functionality can be used. OCIThread's process initialization function, OCIThreadProcessInit(), must be called with care, as described below.
The observed behavior of the initialization and termination functions is the same regardless of whether OCIThread is in single-threaded or multithreaded environment.
Most calls to OCIThread functions take the OCI environment or user session handle as a parameter. The OCIThread context is part of the OCI environment or user session handle and it must be initialized by calling OCIThreadInit(). Termination of the OCIThread context occurs by calling OCIThreadTerm().
|
Note: The |
The following functions are used to implement thread initialization and termination. Detailed descriptions of each function can be found in "Thread Management Functions" .
The passive threading primitives deal with the manipulation of mutex, thread ID's, and thread-specific data. Since the specifications of these primitives do not require the existence of multiple threads, they can be used both on multithreaded and single-threaded platforms.
The type OCIThreadMutex is used to represent a mutual exclusion lock (mutex). A mutex is typically used for one of two purposes:
Mutex pointers can be declared as parts of client structures or as stand-alone variables. Before they can be used, they must be initialized using OCIThreadMutexInit(). Once they are no longer needed, they must be destroyed using OCIThreadMutexDestroy(). A mutex pointer must not be used after it is destroyed.
A thread can acquire a mutex by using OCIThreadMutexAcquire(). This ensures that only one thread at a time is allowed to hold a given mutex. A thread that holds a mutex can release it by calling OCIThreadMutexRelease().
The type OCIThreadKey can be thought of as a process-wide variable that has a thread-specific value. What this means is that all the threads in a process can use any given key. However, each thread can examine or modify that key independently of the other threads. The value that a thread sees when it examines the key will always be the same as the value that it last set for the key. It will not see any values set for the key by the other threads.
The type of the value held by a key is a dvoid * generic pointer.
Keys can be created using OCIThreadKeyInit(). When a key is created, its value is initialized to NULL for all threads.
A thread can set a key's value using OCIThreadKeySet(). A thread can get a key's value using OCIThreadKeyGet().
The OCIThread key functions will save and retrieve data specific to the thread. When clients maintain a pool of threads and assign the threads to different tasks, it may not be appropriate for a task to use OCIThread key functions to save data associated with it.
Here is a scenario of how things can fail: A thread is assigned to execute the initialization of a task. During the initialization, the task stored some data related to it in the thread using OCIThread key functions.
After the initialization, the thread is returned back to the threads pool. Later, the threads pool manager assigned another thread to perform some operations on the task, and the task needs to retrieve the data it stored earlier in initialization. Since the task is running in another thread, it will not be able to retrieve the same data. Applications that use thread pools should be aware of this and be cautious when using OCIThread key functions.
OCIThreadKeyDestFunc is the type of a pointer to a key's destructor routine. Keys can be associated with a destructor routine when they are created (see OCIThreadKeyInit()).
A key's destructor routine will be called whenever a thread that has a non-NULL value for the key terminates.
The destructor routine returns nothing and takes one parameter. The parameter will be the value that was set for key when the thread terminated.
The destructor routine is guaranteed to be called on a thread's value in the key after the termination of the thread and before process termination. No more precise guarantee can be made about the timing of the destructor routine call; thus no code in the process may assume any post-condition of the destructor routine. In particular, the destructor is not guaranteed to execute before a join call on the terminated thread returns.
OCIThreadId is the type that will be used to identify a thread. At any given time, no two threads will ever have the same OCIThreadId. However, OCIThreadId values can be recycled; that is, once a thread dies, a new thread may be created that has the same OCIThreadId as the one that died. In particular, the thread ID must uniquely identify a thread T within a process, and it must be consistent and valid in all threads U of the process for which it can be guaranteed that T is running concurrently with U. The thread ID for a thread T must be retrievable within thread T. This will be done using OCIThreadIdGet().
The OCIThreadId type supports the concept of a NULL thread ID: the NULL thread ID will never be the same as the ID of an actual thread.
The following functions are used to manipulate mutexes, thread keys and thread IDs.
| See Also:
Complete descriptions of each function can be found in "Thread Management Functions" . |
The active threading primitives deal with the manipulation of actual threads. Because the specifications of most of these primitives require that it be possible to have multiple threads, they work correctly only in the enabled OCIThread; In the disabled OCIThread, they always return failure. The exception is OCIThreadHandleGet(); it may be called in a single-threaded environment, in which case it has no effect.
Active primitives should only be called by code running in a multithreaded environment. You can call OCIThreadIsMulti() to determine whether the environment is multithreaded or single-threaded.
Type OCIThreadHandle is used to manipulate a thread in the active primitives: OCIThreadJoin() and OCIThreadClose(). A thread handle opened by OCIThreadCreate() must be closed in a matching call to OCIThreadClose(). A thread handle is invalid after the call to OCIThreadClose().
The distinction between a thread ID and a thread handle in OCIThread usage follows the distinction between the thread ID and the thread handle on Windows NT. On many platforms, the underlying native types are the same.
The following functions are used to implement active threading.
| See Also:
Complete descriptions of the functions are available in "Thread Management Functions" |
This section summarizes some of the more important details relating to the use of OCIThread.
OCIThread only requires that the process initialization function (OCIThreadProcessInit()) be called when OCIThread is being used in a multithreaded application. Failing to call OCIThreadProcessInit() in a single-threaded application is not an error.
Separate calls to OCIThreadInit() will all return the same OCIThread context.
Remember that each call to OCIThreadInit() must eventually be matched by a call to OCIThreadTerm().
OCIThread client code written without using any active primitives can be compiled and used without modifications on both single-threaded and multithreaded platforms.
OCIThread client code written using active primitives will only work correctly on multithreaded platforms. In order to write a version of the same application to run on single-threaded platform, it is necessary to branch your code, whether by branching versions of the source file or by branching at runtime with the OCIThreadIsMulti() call.
The following code sample illustrates the use of OCIThread.
| See Also:
For a listing of the complete demonstration programs, see Appendix B, "OCI Demonstration Programs" |
static OCIEnv *envhp; static OCIError *errhp; void parent(argc, argv) sb4 argc; text **argv; { OCIThreadId *tidArr[5]; OCIThreadHandle *tHndArr[5]; ub4 i; OCIThreadKey *key; (void) OCIInitialize((ub4) OCI_DEFAULT, (dvoid *)0, (dvoid * (*)(dvoid *, size_t)) 0, (dvoid * (*)(dvoid *, dvoid *, size_t))0, (void (*)(dvoid *, dvoid *)) 0 ); (void) OCIEnvInit( (OCIEnv **) &envhp, OCI_DEFAULT, (size_t) 0, (dvoid **) 0 ); (void) OCIHandleAlloc( (dvoid *) envhp, (dvoid **) &errhp, OCI_HTYPE_ERROR, (size_t) 0, (dvoid **) 0); OCIThreadProcessInit(); OCIThreadInit(envhp, errhp); OCIThreadKeyInit(envhp, errhp, &key, (OCIThreadKeyDestFunc) NULL); for (i=0; i<5; i++) { OCIThreadIdInit(envhp, errhp, &(tidArr[i])); OCIThreadHndInit(envhp, errhp, &(tHndArr[i])); } for (i=0; i<5; i++) OCIThreadCreate(envhp, errhp, child, (dvoid *)key, tidArr[i], tHndArr[i]); for (i=0; i<5; i++) { OCIThreadJoin(envhp, errhp, tHndArr[i]); OCIThreadClose(envhp, errhp, tHndArr[i]); } for (i=0; i<5; i++) { OCIThreadIdDestroy(envhp, errhp, &(tidArr[i])); OCIThreadHndDestroy(envhp, errhp, &(tHndArr[i])); } OCIThreadKeyDestroy(envhp, errhp, &key); OCIThreadTerm(envhp, errhp); } void child(arg) dvoid *arg; { OCIThreadKey *key = (OCIThreadKey *)arg; OCIThreadId *tid; dvoid *keyval; OCIThreadIdInit(envhp, errhp, &tid); OCIThreadIdGet(envhp, errhp, tid); if (OCIThreadKeySet(envhp, errhp, key, (dvoid *)tid) != OCI_SUCCESS) printf("Could not set value for key\n"); if (OCIThreadKeyGet(envhp, errhp, key, &keyval) !=OCI_SUCCESS) printf("Could not retrieve value for key\n"); if (keyval != (dvoid *)tid) printf("Incorrect value from key after setting it\n"); /* we must destroy thread id */ OCIThreadIdDestroy(envhp, errhp, &tid); }
Connection pooling is the use of a group (the pool) of reusable physical connections by several sessions, in order to balance loads. The management of the pool is done by OCI, not the application. Applications that can use connection pooling include middle-tier applications for web application servers and e-mail servers.
A sample usage of this feature would be in a web application server connected to a back-end Oracle database. Suppose that a web application server gets several concurrent requests for data from the database server. Typically, it would have to explicitly manage the connections to the database. However, by using this functionality, it can leave that task to OCI. The application can create a pool (or a set of pools) in each environment during initialization.
Oracle has several transaction monitor capabilities such as the fine grained management of database sessions and connections. This is done by separating the notion of database sessions (user handles) from connections (server handles). Using these OCI calls for session switching and session migration, it is possible for an application server or transaction monitor to multiplex several sessions over fewer physical connections, thus achieving a high degree of scalability by pooling of connections and back-end Oracle server processes.
Connection pooling further simplifies the session and connection separation by hiding the management of the physical connection pool from the end user, who has only to create the necessary database sessions. The connection pool itself is normally configured with a shared pool of physical connections, translating to a back-end server pool containing an identical number of dedicated server processes.
The number of physical connections is less than the number of database sessions in use by the application.The number of physical connections and back-end server processes are also reduced by using connection pooling. Thus many more database sessions can be multiplexed.
Connection pooling on the middle-tier is similar to what shared server offers on the back end. Connection pooling makes a dedicated server instance behave like a shared server instance by managing the session multiplexing logic on the middle tier.
Thus, the pooling of dedicated server processes including incoming connections into the dedicated server processes is controlled by the connection pool on the middle tier. The main difference between connection pooling and a shared server is that in the latter case, the connection from the client is normally to a dispatcher in the database instance. The dispatcher is responsible for directing the client request to an appropriate shared server. On the other hand, the physical connection from the connection pool is established directly from the middle-tier to the dedicated server process in the back-end server pool.
Connection pooling is beneficial only if the middle tier itself is multithreaded. Each thread could maintain a session to the database. The actual connections to the database are maintained by the connection pool and these connections (including the pool of dedicated database server processes) are shared among all the threads in the middle tier.
What connection pooling offers is stateless connections and stateful sessions. Users who need to work with stateless sessions should see "Session Pooling".
This advanced concept can be used for different database connections. Multiple connection pools can also be used when different priorities are assigned to users. Different service level guarantees can be implemented using connection pooling.
The following figure illustrates connection pooling, as it is described above:
The steps in using connection pooling in your application are:
Connection pooling requires that the pool handle OCI_HTYPE_CPOOL be allocated by OCIHandleAlloc(). Multiple pools can be created for a given environment handle.
For a single connection pool, here is an allocation example:
OCICPool *poolhp; OCIHandleAlloc((dvoid *) envhp, (dvoid **) &poolhp, OCI_HTYPE_CPOOL, (size_t) 0, (dvoid **) 0));
The function OCIConnectionPoolCreate() initializes the connection pool handle. It has these IN parameters:
connMin, the minimum number of connections to be opened when the pool is created.connIncr, the incremental number of connections to be opened when all the connections are busy and a call needs a connection. This increment is used only when the total number of open connections is less than the maximum number of connections that can be opened in that pool.connMax, the maximum number of connections that can be opened in the pool. When the maximum number of connections are open in the pool, and all the connections are busy, if a call needs a connection, it will wait till it gets one. However, if the OCI_ATTR_CONN_NOWAIT attribute is set for the pool, an error is returned.poolUsername, and a poolPasswd, to allow user sessions to transparently migrate between connections in the pool.All the above attributes can be configured dynamically. So the application has the flexibility of reading the current load (number of open connections and number of busy connections) and tuning these attributes appropriately.
If the pool attributes (connMax, connMin, connIncr) are to be changed dynamically, OCIConnectionPoolCreate() must be called with mode set to OCI_CPOOL_REINITIALIZE.
The OUT parameters poolName and poolNameLen will contain values to be used in subsequent OCIServerAttach() and OCILogon2() calls in place of the database name and the database name length arguments.
There is no limit on the number of pools that can be created by an application. Middle tier applications can take advantage of this feature and create multiple pools to connect to the same server or to different servers, to balance the load based on the specific needs of the application.
Here is an example of this call:
OCIConnectionPoolCreate((OCIEnv *)envhp, (OCIError *)errhp, (OCICPool *)poolhp, &poolName, &poolNameLen, (text *)database,strlen(database), (ub4) conMin, (ub4) conMax, (ub4) conIncr, (text *)pooluser,strlen(pooluser), (text *)poolpasswd,strlen(poolpasswd), OCI_DEFAULT));
There are three interfaces that can be used to log on to the database in connection pooling mode. Each one is described below. The application will need to log on to the database for each thread, using one of the following interfaces.
(1) OCILogon2()
This is the simplest interface. Use this interface when you need a simple Connection Pool connection and do not need to alter any attributes of the session handle. This interface can also be used to make proxy connections to the database.
Here is an example using OCILogon2():
for (i = 0; i < MAXTHREADS; ++i) { OCILogon2(envhp, errhp, &svchp[i], "scott", 5, "tiger", 5, poolName, poolNameLen, OCI_LOGON2_CPOOL)); }
In order to use this interface to get a proxy connection, set the password parameter to NULL.
(2) OCISessionGet()
This is the recommended interface. It gives the user the additional option of using external authentication methods, such as certificates, distinguished name, and so on. OCISessionGet() is the recommended uniform function call to retrieve a session.
Here is an example using OCISessionGet():
for (i = 0; i < MAXTHREADS; ++i) { OCISessionGet(envhp, errhp, &svchp, authp, (OraText *) poolName, strlen(poolName), NULL, 0, NULL, NULL, NULL, OCI_SESSGET_CPOOL) }
(3) OCIServerAttach() and OCISessionBegin():
Another interface can be used if the application needs to set any special attributes on the user session handle and server handle. For such a requirement, applications need to do this:
Allocate all the handles (connection pool handle, server handles, session handles and service context handles).
OCIServerAttach() with mode set to OCI_CPOOL.OCISessionBegin() with mode set to OCI_DEFAULT or OCI_MIGRATE.The OCI_MIGRATE flag will be set internally in any case. Credentials can be set to OCI_CRED_RDBMS or OCI_CRED_PROXY. If the credentials are set to OCI_CRED_PROXY, only username needs to be set on the session handle. (no explicit primary session needs to be created and OCI_ATTR_MIGSESSION need not be set).
With OCI_CPOOL mode (connection pooling), the session memory (UGA) in the back-end database will come out of the SGA. This may require some SGA tuning on the back-end database to have a larger SGA if your application consumes more session memory than the SGA can accommodate. The memory tuning requirements for the back-end database will be similar to configuring the LARGE POOL in case of a shared server back end except that the instance is still in dedicated mode.
| See Also:
For more information, see the section on configuring Shared Server in the Oracle9i Database Performance Tuning Guide and Reference |
If you are still running into the SGA limitation, you should consider:
The application should avoid using dedicated database links on the back end with connection pooling.
If the back end is a dedicated server, effective connection pooling will not be possible because sessions using dedicated database links will be tied to a physical connection rendering that same connection unusable by other sessions. If your application uses dedicated database links and you do not see effective sharing of back-end processes among your sessions, you should consider using shared database links.
| See Also:
For more information about distributed databases, see the section on shared database links in Oracle9i Database Administrator's Guide |
Corresponding to the logon calls, there are three different interfaces to use to log off from the database in connection pooling mode.
(1) OCILogoff():
If OCILogon2() was used to make the connection, OCILogoff() must be used to log off.
(2) OCISessionRelease()
If OCISessionGet() was called to make the connection, then OCISessionRelease() must be called to log off.
(3) OCISessionEnd() and OCIServerDetach()
If OCIServerAttach() and OCISessionBegin() were called to make the connection and start up the session, then OCISessionEnd() must be called to end the session and OCIServerDetach() must be called to release the connection.
Use OCIConnectionPoolDestroy() to destroy the connection pool.
The pool handle is freed using OCIHandleFree().
These last three actions are illustrated in this code fragment:
for (i = 0; i < MAXTHREADS; ++i) { checkerr(errhp, OCILogoff((dvoid *) svchp[i], errhp)); } checkerr(errhp, OCIConnectionPoolDestroy(poolhp, errhp, OCI_DEFAULT)); checkerr(errhp, OCIHandleFree((dvoid *)poolhp, OCI_HTYPE_CPOOL));
See Also:
|
Better response time is obtained if you use OCI client-side prefetch buffers. After calling OCIStmtExecute() for a scrollable cursor, you can call OCIStmtFetch2() using OCI_FETCH_LAST to obtained the size of the result set. Then you can set OCI_ATTR_PREFETCH_ROWS to about 20% of the size and set OCI_ATTR_PREFETCH_MEMORY in case some result sets could take a large amount of memory.
Examples of connection pooling can be found here:
Here is another example of connection pooling:
#include <oci.h> #define MAXTHREAD 10 static OCIError *errhp; static OCIEnv *envhp; static OCICPool *poolhp; static int employeeNum[MAXTHREAD]; static OraText *poolName; static sb4 poolNameLen; static text *database = (text *)""; static text *username =(text *)"SCOTT"; static text *password =(text *)"TIGER"; static text *appusername =(text *)"APPUSER"; static text *apppassword =(text *)"APPPASSWD"; static ub4 conMin = 2; static ub4 conMax = 5; static ub4 conIncr = 1; static void checkerr (OCIError *errhp, sword status); static void threadFunction (dvoid *arg); int main (void) { int i = 0; OCIEnvCreate (&envhp, OCI_THREADED, (dvoid *)0, (dvoid * (*)()) 0, (dvoid * (*)()) 0, (dvoid (*)()) 0, 0, (dvoid *)0); (void) OCIHandleAlloc((dvoid *) envhp, (dvoid **) &errhp, OCI_HTYPE_ERROR, (size_t) 0, (dvoid **) 0); (void) OCIHandleAlloc((dvoid *) envhp, (dvoid **) &poolhp, OCI_HTYPE_CPOOL, (size_t) 0, (dvoid **) 0); /* CREATE THE CONNECTION POOL */ checkerr (errhp, OCIConnectionPoolCreate(envhp, errhp,poolhp, &poolName, &poolNameLen, database,strlen(database), conMin, conMax, conIncr, appusername,strlen(appusername), apppassword,strlen(apppassword),OCI_DEFAULT)); /* Multiple threads using the connection pool */ { OCIThreadId *thrid[MAXTHREAD]; OCIThreadHandle *thrhp[MAXTHREAD]; OCIThreadProcessInit (); checkerr (errhp, OCIThreadInit (envhp, errhp)); for (i = 0; i < MAXTHREAD; ++i) { checkerr (errhp, OCIThreadIdInit (envhp, errhp, &thrid[i])); checkerr (errhp, OCIThreadHndInit (envhp, errhp, &thrhp[i])); } for (i = 0; i < MAXTHREAD; ++i) { employeeNum[i]=i; checkerr (errhp, OCIThreadCreate (envhp, errhp, threadFunction, (dvoid *) &employeeNum[i], thrid[i], thrhp[i])); } for (i = 0; i < MAXTHREAD; ++i) { checkerr (errhp, OCIThreadJoin (envhp, errhp, thrhp[i])); checkerr (errhp, OCIThreadClose (envhp, errhp, thrhp[i])); checkerr (errhp, OCIThreadIdDestroy (envhp, errhp, &(thrid[i]))); checkerr (errhp, OCIThreadHndDestroy (envhp, errhp, &(thrhp[i]))); } checkerr (errhp, OCIThreadTerm (envhp, errhp)); } /* ALL THE THREADS ARE COMPLETE */ checkerr(errhp, OCIConnectionPoolDestroy(poolhp, errhp, OCI_DEFAULT)); checkerr(errhp, OCIHandleFree((dvoid *)poolhp, OCI_HTYPE_CPOOL)); checkerr(errhp, OCIHandleFree((dvoid *)errhp, OCI_HTYPE_ERROR)); } /* end of main () */ static void threadFunction (dvoid *arg) { int empno = *(int *)arg; OCISvcCtx *svchp = (OCISvcCtx *) arg; text insertst1[256]; OCIStmt *stmthp = (OCIStmt *)0; checkerr(errhp,OCILogon2(envhp, errhp, &svchp, (CONST OraText *)username, strlen(username), (CONST OraText *)password, strlen(password), (CONST OraText *)poolName, poolNameLen, OCI_CPOOL)); sprintf(insertst1,"INSERT INTO emp(empno, ename, job, sal, deptno) values\ (%d,'abc','MANAGER',122,20)",empno); OCIHandleAlloc(envhp, (dvoid **)&stmthp, OCI_HTYPE_STMT, (size_t)0, (dvoid **)0); checkerr(errhp, OCIStmtPrepare (stmthp, errhp, (text *)insertst1, (ub4)strlen(insertst1), OCI_NTV_SYNTAX, OCI_DEFAULT)); checkerr(errhp, OCIStmtExecute (svchp, stmthp, errhp, (ub4)1, (ub4)0, (OCISnapshot *)0, (OCISnapshot *)0, OCI_DEFAULT )); checkerr(errhp, OCITransCommit(svchp,errhp,(ub4)0)); checkerr(errhp, OCIHandleFree((dvoid *) stmthp, OCI_HTYPE_STMT)); checkerr(errhp, OCILogoff((dvoid *) svchp, errhp)); } /* end of threadFunction (dvoid *) */ void checkerr(errhp, status) OCIError *errhp; sword status; { text errbuf[512]; sb4 errcode = 0; switch (status) { case OCI_SUCCESS: break; case OCI_SUCCESS_WITH_INFO: (void) printf("Error - OCI_SUCCESS_WITH_INFO\n"); break; case OCI_NEED_DATA: (void) printf("Error - OCI_NEED_DATA\n"); break; case OCI_NO_DATA: (void) printf("Error - OCI_NODATA\n"); break; case OCI_ERROR: (void) OCIErrorGet((dvoid *)errhp, (ub4) 1, (text *) NULL, &errcode, errbuf, (ub4) sizeof(errbuf), OCI_HTYPE_ERROR); (void) printf("Error - %.*s\n", 512, errbuf); break; case OCI_INVALID_HANDLE: (void) printf("Error - OCI_INVALID_HANDLE\n"); break; case OCI_STILL_EXECUTING: (void) printf("Error - OCI_STILL_EXECUTE\n"); break; case OCI_CONTINUE: (void) printf("Error - OCI_CONTINUE\n"); break; default: break; } }
Session pooling means that the application will create and maintain a group of stateless sessions to the database. These sessions will be handed over to thin clients as requested. If no sessions are available, a new one may be created. When the client is done with the session, the client will release it to the pool. Thus, the number of sessions in the pool can increase dynamically.
Some of the sessions in the pool may be 'tagged' with certain properties. For instance, a user may request for a default session, set certain attributes on it, then label it or 'tag' it and return in to the pool. That user, or some other user, can require a session with the same attributes, and thus request for a session with the same tag. There may be several sessions in the pool with the same tag. The 'tag' on a session can be changed or reset.
| See Also:
"Using Tags in Session Pools" for a discussion of using tags. |
Proxy sessions, too, can be created and maintained through this interface.
The behavior of the application when no free sessions are available and the pool has reached it's maximum size, will depend on certain attributes. A new session may be created or an error returned, or the thread may just block and wait for a session to become free.
The main benefit of this type of pooling will be performance. Making a connection to the database is a time-consuming activity, especially when the database is remote. Thus, instead of a client spending time connecting to the server, authenticating its credentials, and then receiving a valid session, it can just pick one from the pool.
Session pooling has the following features:
A session pool can be either homogeneous or heterogeneous. Homogeneous session pooling means that sessions in the pool are alike with respect to authentication (have the same username and password and privileges). Heterogeneous session pooling means that you must provide authentication information because the sessions can have different security attributes and privileges.
The tags provide a way for users to customize sessions in the pool. A client may get a default or untagged session from a pool, set certain attributes on the session (such as NLS settings), and return the session to the pool, labeling it with an appropriate tag in the OCISessionRelease() call.
The user, or some other user, may request a session with the same tags in order to have a session withe the same attributes, and can do so by providing the same tag in the OCISessionGet() call.
| See Also:
|
Two handle types have been added for session pooling:
This is the session pool handle. It is allocated using OCIHandleAlloc(). It needs to be passed to OCISessionPoolCreate(), and OCISessionPoolDestroy(). It has the attribute type OCI_HTYPE_SPOOL.
An example of the OCIHandleAlloc() call follows:
OCISPool *spoolhp; OCIHandleAlloc((dvoid *) envhp, (dvoid **) &spoolhp, OCI_HTYPE_SPOOL, (size_t) 0, (dvoid **) 0));
For an environment handle, multiple session pools can be created.
This is the authentication information handle. It is allocated using OCIHandleAlloc(). It is passed to OCISessionGet(). It supports all the attributes that are supported for user session handle. Please refer to user session handle attributes for more information. The authentication information handle has the attribute type OCI_HTYPE_AUTHINFO.
An example of the OCIHandleAlloc() call follows:
OCIAuthInfo *authp; OCIHandleAlloc((dvoid *) envhp, (dvoid **) &authp, OCI_HTYPE_AUTHINFO, (size_t) 0, (dvoid **) 0));
See Also:
|
The steps in writing a simple session pooling application which uses a username and password are:
OCIHandleAlloc() for an OCISPool handle. Multiple session pools can be created for an environment handle.OCISessionPoolCreate() with mode set to OCI_DEFAULT (for a new session pool). Refer to the function for a discussion of the other modes.OCIHandleAlloc().OCIAttrSet().mode set to OCI_SESSGET_SPOOL.OCISessionRelease().OCIHandleFree().OCISessionPoolDestroy().Here are the usages for OCI calls for session pooling.
Session pooling requires that the pool handle OCI_HTYPE_SPOOL be allocated by calling OCIHandleAlloc().
Multiple pools can be created for a given environment handle. For a single session pool, here is an allocation example:
OCISPool *poolhp; OCIHandleAlloc((dvoid *) envhp, (dvoid **) &poolhp, OCI_HTYPE_SPOOL, (size_t) 0, (dvoid **) 0));
The function OCISessionPoolCreate() can be used to create the session pool. Here is an example of how to use this call:
OCISessionPoolCreate(envhp, errhp,poolhp, &poolName, &poolNameLen, database,(sb4)strlen((const signed char *)database), conMin, conMax, conIncr, appusername,(sb4)strlen((const signed char *)appusername), apppassword,(sb4)strlen((const signed char *)apppassword), OCI_DEFAULT);
There are two interfaces that can be used to logon to the database in session pooling mode.
(1) OCILogon2():
This is the simplest interface. However, it does not give the user the option of using tagging.
Here is an example of how OCILogon2() can be used to log on to the database in session pooling mode:
for (i = 0; i < MAXTHREADS; ++i) { OCILogon2(envhp, errhp, &svchp[i], "scott", 5, "tiger", 5, poolName, poolNameLen, OCI_LOGON2_SPOOL)); }
(2) OCISessionGet():
This is the recommended interface. It gives the user the option of using tagging to label sessions in the pool, and thus make it easier to retrieve specific sessions.
An example of using OCISessionGet() follows:
OCISessionGet(envhp, errhp, &svchp, authInfop, (OraText *)database,strlen(database), tag, strlen(tag), &retTag, &retTagLen, &found, OCI_SESSGET_SPOOL)
Corresponding to the preceding logon calls, there are two interfaces to use to log off from the database in session pooling mode.
(1) OCILogoff():
If OCILogon2() was used to make the connection, OCILogoff() must be used to log off.
(2) OCISessionRelease()
If OCISessionGet() was called to make the connection, then OCISessionRelease() must be called to log off.
OCISessionPoolDestroy() must be called to destroy the session pool. Here is an example of how this call can be made:
OCISessionPoolDestroy(poolhp, errhp, OCI_DEFAULT);
OCIHandleFree() must be called to free the session pool handle. Here is how this call can be made:
OCIHandleFree((dvoid *)poolhp, OCI_HTYPE_SPOOL);
Here is an example of session pooling:
Statement caching refers to the feature, first introduced in release 9.2, that provides and manages a cache of statements for each session. In the server, it means that cursors are ready to be used without the need to parse the statement again. Statement caching can be used with connection pooling and with session pooling, and will improve performance and scalability. It can be used without session pooling as well, and it works with TAF. The OCI calls that implement statement caching are:
Users perform the usual OCI steps to logon. The call to obtain a session will have a mode that specifies whether statement caching is enabled for the session. Initially the statement cache will be empty. Developers will try to find a statement in the cache using the statement text. If the statement exists the API will return a previously prepared statement handle, otherwise it will return an newly prepared statement handle.
The application developer can perform binds and define and then simply execute and fetch the statement before returning the statement back to the cache. In the latter case, where the statement handle was not found, the developer will need to set different attributes on the handle in addition to the other steps.
OCIStmtPrepare2() will also take a mode which will determine if the developer wants a prepared statement handle or a null statement handle if the statement is not found in the cache.
The pseudo code will look like:
OCISessionBegin( userhp, ... OCI_STMT_CACHE) ; OCIAttrset(svchp, userhp, ...); /* Set the user handle in the service context */ OCIStmtPrepare2(svchp, &stmthp, stmttext, key, ...); OCIBindByPos(stmthp, ...); OCIDefineByPos(stmthp, ...); OCIStmtExecute(svchp, stmthp, ...); OCIStmtFetch(svchp, ...); OCIStmtRelease(stmthp, ...); ...
The concepts remain the same, except that the statement cache is enabled at the session pool layer rather than at the session layer. If a session pool has statement cache enabled, then all the statements in all the sessions in the pool will be cached, else all the statements in all the sessions remain not cached.
Here are some notes to follow:
OCIStmtPrepare2() instead of OCIStmtPrepare(). If you are using OCIStmtPrepare(), you are strongly urged not to use a statement handle across different service contexts. Doing so will raise an error if the statement has been obtained by OCIStmtPrepare2(). Migration of a statement handle to a new service context actually closes the cursor associated with the old session and therefore no sharing is achieved. Client-side sharing is also not obtained, because OCI will free all buffers associated with the old session when the statement handle is migrated.OCIStmtPrepare2(), even if the session does not have a statement cache, will also allocate the statement handle and therefore applications using only OCIStmtPrepare2() should not call OCIHandleAlloc() for the statement handle.OCIStmtPrepare2() must be followed with OCIStmtRelease() after the user is done with the statement handle. If statement caching is used, this will release the statement to the cache. If statement caching is not used, the statement will be deallocated. Do not call OCIHandleFree() to free the memory.OCIStmtPrepare2() is made with the OCI_STMTCACHE_SEARCH_ONLY mode and a NULL statement was returned (statement was not found), the subsequent call to OCIStmtRelease() is not required and should not be performed.OCIStmtRelease() for a statement that was prepared using OCIStmtPrepare().Here is an example of statement caching:
The Oracle Call Interface has the capability to execute user-specific code in addition to OCI calls. This functionality can be used for:
The OCI callback feature has been added by providing support for calling user code before or after executing the OCI calls. Functionality has also been provided to allow the user-defined code to be executed instead of executing the OCI code.
The user callback code can also be registered dynamically without modifying the source code of the application. The dynamic registration is implemented by loading up to five user-created dynamically linked libraries after the initialization of the environment handle during the OCIEnvCreate() call. These user-created libraries (such as Dynamic Link Libraries (DLLs) on NT, or shared libraries on SolarisTM Operating Environment) register the user callbacks for the selected OCI calls transparently to the application.
For a listing of the complete demonstration programs that illustrate the OCI user callback feature, see Appendix B, "OCI Demonstration Programs".
An application can register user callback libraries with the OCIUserCallbackRegister() function. Callbacks are registered in the context of the environment handle. An application can retrieve information about callbacks registered with a handle with the OCIUserCallbackGet() function.
| See Also:
For detailed descriptions of these functions and their parameters, refer to the descriptions of |
A user-defined callback is a subroutine that is registered against an OCI call and an environment handle. It can be specified to be either an entry callback, a replacement callback, or an exit callback.
If a replacement or exit callback returns anything other than OCI_CONTINUE, then the return code from the callback is returned from the associated OCI call.
A user callback can return OCI_INVALID_HANDLE when either an invalid handle or an invalid context is passed to it.
A user callback is registered using the OCIUserCallbackRegister() call.
| See Also:
See |
Currently, OCIUserCallbackRegister() is only registered on the environment handle. The user's callback function of typedef OCIUserCallback is registered along with its context for the OCI call identified by the OCI function code, fcode. The type of the callback, whether entry, replacement, or exit, is specified by the when parameter.
For example, the stmtprep_entry_dyncbk_fn entry callback function and its context dynamic_context, are registered against the environment handle hndlp for the OCIStmtPrepare() call by calling the OCIUserCallbackRegister() function with the following parameters.
OCIUserCallbackRegister( hndlp, OCI_HTYPE_ENV, errh, stmtprep_entry_dyncbk_fn, dynamic_context, OCI_FNCODE_STMTPREPARE, OCI_UCBTYPE_ENTRY (OCIUcb*) NULL);
The user callback function has to follow the following syntax:
typedef sword (*OCIUserCallback) (dvoid *ctxp, /* context for the user callback*/ dvoid *hndlp, /* handle for the callback, env handle for now */ ub4 type, /* type of handlp, OCI_HTYPE_ENV for this release */ ub4 fcode, /* function code of the OCI call */ ub1 when, /* type of the callback, entry or exit */ sword returnCode, /* OCI return code */ ub4 *errnop, /* Oracle error number */ va_list arglist); /* parameters of the oci call */
In addition to the parameters described in the OCIUserCallbackRegister() call, the callback is called with the return code, errnop, and all the parameters of the original OCI as declared by the prototype definition.
The return code is always passed in as OCI_SUCCESS and *errnop is always passed in as 0 for the first entry callback. Note that *errnop refers to the content of errnop because errnop is an IN/OUT parameter.
If the callback does not want to change the OCI return code, then it must return OCI_CONTINUE, and the value returned in *errnop is ignored. If on the other hand, the callback returns any other return code than OCI_CONTINUE, the last returned return code becomes the return code for the call. At the this point, the value of *errnop returned is set in the error handle, or in the environment handle if the error information is returned in the environment handle because of the absence of the error handle for certain OCI calls such as OCIHandleAlloc().
For replacement callbacks, the returnCode is the non-OCI_CONTINUE return code from the previous callback or OCI call and *errnop is the value of the error number being returned in the error handle. This allows the subsequent callback to change the return code or error information if needed.
The processing of replacement callbacks is different in that if it returns anything other than OCI_CONTINUE, then subsequent replacement callbacks and OCI code is bypassed and processing jumps to the exit callbacks.
Note that if the replacement callbacks return OCI_CONTINUE to allow processing of OCI code, then the return code from entry callbacks is ignored.
All the original parameters of the OCI call are passed to the callback as variable parameters and the callback must retrieve them using the va_arg macros. The callback demonstration programs provide examples.
A null value can be registered to de-register a callback. That is, if the value of the callback (OCIUserCallback) is NULL in the OCIUserCallbackRegister() call, then the user callback is de-registered.
When using the thread-safe mode, the OCI program acquires all mutexes before calling the user callbacks.
This pseudocode describes the overall processing of a typical OCI call:
OCIXyzCall() { Acquire mutexes on handles; retCode = OCI_SUCCESS; errno = 0; for all ENTRY callbacks do { EntryretCode = (*entryCallback)(..., retcode, &errno, ...); if (retCode != OCI_CONTINUE) { set errno in error handle or environment handle; retCode = EntryretCode; } } for all REPLACEMENT callbacks do { retCode = (*replacementCallback) (..., retcode, &errno, ...); if (retCode != OCI_CONTINUE) { set errno in error handle or environment handle goto executeEXITCallback; } } retCode = return code for XyzCall; /* normal processing of OCI call */ errno = error number from error handle or env handle; executeExitCallback: for all EXIT callbacks do { exitRetCode = (*exitCallback)(..., retCode, &errno,...); if (exitRetCode != OCI_CONTINUE) { set errno in error handle or environment handle; retCode = exitRetCode; } } release mutexes; return retCode }
If the callbacks are a total replacement of the OCI code, then they usually maintain their own error information in the call context and use that to return error information in bufp and errnop parameters of the replacement callback of the OCIErrorGet() call.
If on the other hand, the callbacks are either partially overriding OCI code, or just doing some other post processing, then they can use the exit callback to modify the error text and errnop parameters of the OCIErrorGet() by their own error message and error number. Note that the *errnop passed into the exit callback is the error number in the error or the environment handle.
If an entry callback wants to return an error to the caller of the OCI call, then it must register a replacement or exit callback. This is because if the OCI code is executed, then the error code from the entry callback is ignored. Therefore the entry callback should pass the error to the replacement or exit callback through its own context.
Because user callbacks are expected to be used for monitoring OCI behavior or to access other data sources, it is desirable that the registration of the callbacks be done transparently and non-intrusively. This is accomplished by loading user-created dynamically linked libraries at OCI initialization time. These dynamically linked libraries are called packages. The user-created packages register the user callbacks for the selected OCI calls. These callbacks can further register or de-register user callbacks as needed when receiving control at runtime.
A makefile (ociucb.mk on SolarisTM Operating Environment) is provided with the OCI demonstration programs to create the package. The exact naming and location of this package is operating system dependent. The source code for the package must provide code for special callbacks that are called at OCI initialization and environment creation times.
The loading of the package is controlled by setting an operating system environment variable, ORA_OCI_UCBPKG. This variable names the packages in a generic way. The packages must be located in the $ORACLE_HOME/lib directory.
The ORA_OCI_UCBPKG variable can contain a semicolon separated list of package names. The packages are loaded in the order they are specified in the list.
For example, previously one specified the package as:
setenv ORA_OCI_UCBPKG mypkg
Now, you can still specify the package as above, but in addition multiple packages can be specified as:
setenv ORA_OCI_UCBPKG "mypkg;yourpkg;oraclepkg;sunpkg;msoftpkg"
All these packages are loaded in order. That is, mypkg is loaded first and msoftpkg is loaded last.
A maximum of five packages can be specified.
Previously a package had to specify the source code for the OCIEnvCallback() function. Now the OCIEnvCallback() function is obsolete. Instead, the package source must provide two functions. The first function has to be named as packagename suffixed with the word Init. For example, if the package is named foo, then the source file (for example, but not necessarily, foo.c) should contain a fooInit() function with a call to OCISharedLibInit() function specified exactly as:
sword fooInit(metaCtx, libCtx, argfmt, argc, argv) dvoid * metaCtx; /* The metacontext */ dvoid * libCtx; /* The context for this package. */ ub4 argfmt; /* package argument format */ sword argc; /* package arg count*/ dvoid * argv[]; /* package arguments */ { return (OCISharedLibInit(metaCtx, libCtx, argfmt, argc, argv, fooEnvCallback)); }
The last parameter of the OCISharedLibInit() function, fooEnvCallback(), in this case, is the name of the second function. It can be named anything, but by convention it can be named packagename suffixed with the word EnvCallback.
This function is a replacement for OCIEnvCallback(). Now all the dynamic user callbacks must be registered in this function. The function must be of type OCIEnvCallbackType, which is specified as:
typedef sword (*OCIEnvCallbackType)(OCIEnv *env, ub4 mode, size_t xtramem_sz, dvoid *usrmemp, OCIUcb *ucbDesc);
When an environment handle is created, then this callback function is called at the very end. The env parameter is the newly created environment handle.
The mode, xtramem_sz, and usrmemp are the parameters passed to the OCIEnvCreate() call. The last parameter, ucbDesc, is a descriptor that is passed to the package. The package uses this descriptor to register the user callbacks as described later.
A sample ociucb.c file is provided in the demo directory. The makefile ociucb.mk is also provided (on SolarisTM Operating Environment) in the demo directory to create the package. Please note that this may be different on other platforms. The demo directory also contains full user callback demo programs (cdemoucb.c, cdemoucbl.c,) illustrating this.
User callbacks can both be registered statically in the application itself or dynamically at runtime in the DLLs. A mechanism is needed to allow the application to override a previously registered callback and then later invoke the overridden one in the newly registered callback to preserve the behavior intended by the dynamic registrations. This can result in chaining of user callbacks.
For this purpose, the OCIUserCallbackGet() function is provided to find out which function and context is registered for an OCI call.
| See Also:
See |
Because Oracle is the predominant database accessed, applications can take advantage of the OCI interface to access non-Oracle data by using the user callbacks to access them. This allows an application written in OCI to access Oracle data without any performance penalty. To access non-Oracle data sources, drivers can be written that access the non-Oracle data in user callbacks. Because OCI provides a very rich interface, there is usually a straightforward mapping of OCI calls to most data sources. This solution is better than writing applications for other middle layers such as ODBC that introduce performance penalties for all data sources. Using OCI does not incur any penalty for the common case of accessing Oracle data sources, and incurs the same penalty that ODBC does for non-Oracle data sources.
There are certain restrictions on the usage of callback functions, including OCIEnvCallback():
OCIUserCallbackRegister(), OCIUserCallbackGet(), OCIHandleAlloc(), OCIHandleFree(). Even for these functions, if they are called in a user callback, then callbacks on them are not called to avoid recursion. For example, if OCIHandleFree() is called in the callback for OCILogoff(), then the callback for OCIHandleFree() is disabled during the execution of the callback for OCILogoff().OCIUserCallbackRegister() call itself, or for any of the following:
For examp