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 OCI's facility for working with objects in an Oracle database server. It also discusses OCI's object navigational function calls. This chapter includes the following sections:
The object cache is a client-side memory buffer that provides lookup and memory management support for objects. It stores and tracks object instances that have been fetched by an OCI application. The object cache provides memory management.
When objects are fetched by the application through a SQL SELECT statement, or through an OCI pin operation, a copy of the object is stored in the object cache. Objects that are fetched directly through a SELECT statement are fetched by value, and they are non-referenceable objects which cannot be pinned. Only referenceable objects may be pinned.
If an object is being pinned, and an appropriate version already exists in the cache, it does not need to be fetched from the server.
Every client program that uses OCI to dereference REFs to retrieve objects utilizes the object cache. A client-side object cache is allocated for every OCI environment handle initialized in object mode. Multiple threads of a process can share the same client-side cache by sharing the same OCI environment handle.
Exactly one copy of each referenceable object exists in the cache for each connection. Dereferencing a REF many times or dereferencing several equivalent REFs returns the same copy of the object.
If you modify a copy of an object in the cache, you must flush the changes to the server before they are visible to other processes. Objects that are no longer needed can be unpinned or freed; they can then be swapped out of the cache, freeing the memory space they occupied.
When database objects are loaded into the cache, they are transparently mapped into the C language structures. The object cache maintains the association between all object copies in the cache and their corresponding objects in the database. When the transaction is committed, changes made to the object copy in the cache are automatically propagated to the database.
The cache does not manage the contents of object copies; it does not automatically refresh object copies. The application must ensure the correctness and consistency of the contents of object copies. For example, if the application marks an object copy for insert, update, or delete, then aborts the transaction, the cache simply unmarks the object copy but does not purge or invalidate the copy. The application must pin recent or latest, or refresh the object copy in the next transaction. If it pins any, it may get the same object copy with its uncommitted changes from the previous aborted transaction.
| See Also:
For more information about pin options, see "Pinning an Object Copy" |
The object cache is created when the OCI environment is initialized using OCIEnvCreate() with mode set to OCI_OBJECT.
The object cache maintains a fast look-up table for mapping REFs to objects. When an application de-references a REF and the corresponding object is not yet cached in the object cache, the object cache automatically sends a request to the server to fetch the object from the database and load it into the object cache.
Subsequent de-references of the same REF will be faster since they become local cache access and do not incur network round trips. To notify the object cache that an application is accessing an object in the cache, the application pins the object; when it is done with the object, it should unpin it. The object cache maintains a pin count for each object in the cache, the count is incremented upon a pin call and unpin call decrements it. When the pin count goes to zero, that means the object is no longer needed by the application.
The object cache uses an least-recently used (LRU) algorithm to manage the size of the cache. The LRU algorithm frees candidate objects when the cache reaches the maximum size. The candidate objects are objects with a pin count of zero.
Each application processes running against the same server has its own object cache, as shown in Figure 13-1, "The Object Cache".
The object cache tracks the objects that are currently in memory, maintains references to the objects, manages automatic object swapping, and tracks object meta-attributes.
The object cache does not automatically maintain value coherency or consistency between object copies and their corresponding objects in the database. In other words, if an application makes changes to an object copy, the changes are not automatically applied to the corresponding object in the database, and vice versa. The cache provides operations such as flushing a modified object copy to the database and refreshing a stale object copy with the latest value from the database to enable the program to maintain some coherency.
The object cache has two important parameters associated with it, which are attributes of the environment handle:
These parameters refer to levels of cache memory usage, and they help to determine when the cache automatically ages out eligible objects to free up memory.
If the memory occupied by the objects currently in the cache reaches or exceeds the maximum cache size, the cache automatically begins to free (or ages out) unmarked objects which have a pin count of zero. The cache continues freeing such objects until memory usage in the cache reaches the optimal size, or until it runs out of objects eligible for freeing. Note that the cache can grow beyond the specified maximum cache size.
OCI_ATTR_CACHE_MAX_SIZE is specified as a percentage of OCI_ATTR_CACHE_OPT_SIZE. The maximum object cache size (in bytes) is computed by incrementing OCI_ATTR_CACHE_OPT_SIZE by OCI_ATTR_CACHE_MAX_SIZE percentage, as follows:
maximum_cache_size = optimal_size + optimal_size * max_size_percentage / 100
or
maximum_cache_size = OCI_ATTR_CACHE_OPT_SIZE + OCI_ATTR_CACHE_OPT_SIZE * OCI_ATTR_CACHE_MAX_SIZE / 100
Set the value of OCI_ATTR_CACHE_MAX_SIZE at 110% of the OCI_ATTR_CACHE_OPT_SIZE. The default value for OCI_ATTR_CACHE_OPT_SIZE is 8M bytes.
The cache size attributes of the environment handle can be set with the OCIAttrSet() call and retrieved with the OCIAttrGet() function.
| See Also:
See the section "Environment Handle Attributes" for more information. |
This section describes the most important functions the object cache provides to operate on object copies.
| See Also:
All of the OCI's navigational and cache/object management functions are listed in the section "OCI Navigational Functions". |
Pinning an object copy enables the application to access it in the cache by dereferencing the REF to it.
Unpinning an object indicates to the cache that the object currently is not being used. Objects should be unpinned when they are no longer needed to make them eligible for implicit freeing by the cache, thus freeing up memory.
Freeing an object copy removes it from the cache and frees its memory.
Marking an object notifies the cache that an object copy has been updated in the cache and the corresponding object must be updated in the server when the object copy is flushed.
Unmarking an object removes the indication that the object has been updated.
Flushing an object writes local changes made to marked object copies in the cache to the corresponding objects in the server. When this happens, the copies in the object cache are unmarked.
Refreshing an object copy in the cache replaces it with the latest value of the corresponding object in the server.
|
Note: Pointers to top-level object memory are valid after a refresh. Pointers to secondary-level memory (for example, string text pointers, collections, etc.) may become invalid after a refresh. |
Pin, unpin, and free functions are discussed in this section.
When an application needs to dereference a REF in the object cache, it calls OCIObjectPin(). This call dereferences the REF and pins the object copy in the cache. As long as the object copy is pinned, it is guaranteed to be accessible by the application. Another variation of OCIObjectPin() is OCIObjectArrayPin() which takes an array of REFs, dereferences the REFs, and pins the object copies. Both OCIObjectPin() and OCIObjectArrayPin() take a pin option, any, recent, or latest. The datatype of the pin option is OCIPinOpt.
When the program pins an object, the program also specifies one of two possible values for the pin duration: session or transaction. The datatype of the duration is OCIDuration.
OCIObjectUnpin()).When loading an object copy into the cache from the database, the cache effectively executes
SELECT VALUE(t) FROM t WHERE REF(t) = :r
where t is the object table storing the object, and r is the REF, and the fetched value becomes the value of the object copy in the cache.
Since the object cache effectively executes a separate SELECT statement to load each object copy into the cache, in a read-committed transaction, object copies are not guaranteed to be read-consistent with each other.
In a serializable transaction, object copies pinned recent or latest are read-consistent with each other because the SELECT statements to load these object copies are executed based on the same database snapshot.
The object cache model is orthogonal to or independent of the Oracle transaction model. The behavior of the object cache does not change based on the transaction model, even though the objects that are retrieved from the server through the object cache can be different when running the same program under different transaction models (for example, read committed versus serializable).
An object copy can be unpinned when it is no longer used by the program. It then becomes available to be freed. An object copy must be both completely unpinned and unmarked in order to become eligible to be implicitly freed by the cache when the cache begins to run out of memory. To be completely unpinned, an object copy that has been pinned N times must be unpinned N times.
An unpinned but marked object copy is not eligible for implicit freeing until the object copy is flushed or explicitly unmarked by the user. However, the object cache implicitly frees object copies only when it begins to run out of memory, so an unpinned object copy need not necessarily be freed. If it has not been implicitly freed and is pinned again (with the any or recent options), the program gets the same object copy.
An application calls OCIObjectUnpin() or OCIObjectPinCountReset() to unpin an object copy. In addition, a program can call OCICacheUnpin() to completely unpin all object copies in the cache for a specific connection.
Freeing an object copy removes it from the object cache and frees up its memory. The cache supports two methods for freeing up memory:
OCIObjectFree() which takes an option to (forcefully) free either a marked or pinned object copy. The program can also call OCICacheFree() to free all object copies in the cache.| See Also:
For more information, see the section "Object Cache Parameters". |
For memory management reasons, it is important that applications unpin objects when they are no longer needed. This makes these objects available for aging out of the cache, and makes it easier for the cache to free memory when necessary.
OCI does not provide a function to free unreferenced objects in the client-side cache.
Functions for marking and unmarking object copies are discussed in this section.
An object copy can be created, updated, and deleted locally in the cache. If the object copy is created in the cache (by calling OCIObjectNew()), the object copy is marked for insert by the object cache, so that the object will be inserted in the server when the object copy is flushed.
If the object copy is updated in the cache, the user has to notify the object cache by marking the object copy for update (by calling OCIObjectMarkUpdate()). When the object copy is flushed, the corresponding object in the server is updated with the value in the object copy.
If the object copy is deleted, the object copy is marked for delete in the object cache (by calling OCIObjectMarkDelete()). When the object copy is flushed, the corresponding object in the server is deleted. The memory of the marked object copy is not freed until it is flushed and unpinned. When pinning an object marked for delete, the program receives an error, as if the program is dereferencing a dangling reference.
When a user makes multiple changes to an object copy, it is the final results of these changes which are applied to the object in the server when the copy is flushed. For example, if the user updates and deletes an object copy, the object in the server is simply deleted when the object copy is flushed. Similarly, if an attribute of an object copy is updated multiple times, it is the final value of this attribute which is updated in the server when the object copy is flushed.
The program can mark an object copy as updated or deleted only if the object copy has been loaded into the object cache.
A marked object copy can be unmarked in the object cache. By unmarking a marked object copy, the changes that are made to the object copy are not flushed to the server. The object cache does not undo the local changes that are already made to the object copy.
A program calls OCIObjectUnmark() to unmark an object. In addition, a program can call OCICacheUnmark() to unmark all object copies in the cache for a specific connection.
Cache/server synchronization operations (flushing, refreshing) are discussed in this section.
The local changes made to a marked object copy in the cache are written to the server when the object copy is flushed. The program can call OCIObjectFlush() to flush a single object copy or OCICacheFlush() to flush all marked object copies in the cache or a list of selected marked object copies. OCICacheFlush() flushes objects associated with a specific service context. See OCICacheFlush().
After flushing an object copy, the object copy is unmarked. (Note that the object is locked in the server after it is flushed; the object copy is therefore marked as locked in the cache.)
|
Note: The |
If an application wishes to flush only dirty objects of a certain type, this functionality is available through the callback function which is an optional argument to the OCICacheFlush() call. The application can define a callback which returns only the desired objects. In this case the operation still incurs only a single server round-trip for the flush.
In the default mode during OCICacheFlush(), the objects are flushed in the order that they are marked dirty. The performance of this flush operation can be considerably improved by setting the OCI_ATTR_CACHE_ARRAYFLUSH attribute in the environment handle.
However, OCI_ATTR_CACHE_ARRAYFLUSH mode should be used only if the order in which the objects are flushed is not important. During this mode, the dirty objects are grouped together and sent to the server in a manner that enables the server to efficiently update its tables. When this mode is enabled, it is not guaranteed that the order in which the objects are marked dirty is preserved.
When refreshed, an object copy is reloaded with the latest value of the corresponding object in the server. The latest value may contain changes made by other committed transactions and changes made directly (not through the object cache) in the server by the transaction. The program can change objects directly in the server using SQL DML, triggers, or stored procedures.
To refresh a marked object copy, the program must first unmark the object copy. An unpinned object copy is simply freed when it is refreshed (that is, when the whole cache is refreshed).
The program can call OCIObjectRefresh() to refresh a single object copy or OCICacheRefresh() to refresh all object copies in the cache, all object copies that are loaded in a transaction (that is, object copies that are pinned recent or pinned latest), or a list of selected object copies.
When an object is flushed to the server, triggers can be fired to modify more objects in the server. The same objects (modified by the triggers) in the object cache become out-of-date, and must be refreshed before they can be locked or flushed.
The various meta-attribute flags and durations of an object are modified as described in Table 13-1 after being refreshed:
| Object Attribute | Status After Refresh |
|---|---|
|
existent |
set to appropriate value |
|
pinned |
unchanged |
|
flushed |
reset |
|
allocation duration |
unchanged |
|
pin duration |
unchanged |
During refresh, the object cache loads the new data into the top-level memory of an object copy, thus reusing the top level memory. The top-level memory of an object copy contains the in-line attributes of the object. On the other hand, the memory for the out-of-line attributes of an object copy may be freed and relocated, since the out-of-line attributes can vary in size.
| See Also:
See the section "Memory Layout of an Instance" for more information about object memory |
OCI functions related to object locking are discussed in this section.
When pinning an object you can specify whether the object should be locked or not through lock options. When an object is locked a server side lock is acquired and this prevents any other user from modifying the object. The lock is released when the transaction commits or rollbacks. The different lock options are:
SELECT FOR UPDATE statement.SELECT FOR UPDATE WITH NOWAIT statement.The program can optionally call OCIObjectLock() to lock an object for update. This call instructs the object cache to get a row lock on the object in the database. This is similar to executing
SELECT NULL FROM t WHERE REF(t) = :r FOR UPDATE
where t is the object table storing the object to be locked and r is the REF identifying the object. The object copy is marked locked in the object cache after OCIObjectLock() is called.
To lock a graph or set of objects, several OCIObjectLock() calls are required, one for each object, or the array pin OCIObjectArrayPin() call can be used for better performance.
By locking an object, the application is guaranteed that the object in the cache is up-to-date. No other transaction can modify the object while the application has it locked.
At the end of a transaction, all locks are released automatically by the server. The locked indicator in the object copy is reset.
In some cases, an application may attempt to lock an object which is currently locked by another user. In this case the application is blocked.
In order to avoid blocking when trying to lock an object, an application can use the OCIObjectLockNoWait() call instead of OCIObjectLock(). This function returns an error if it is unable to lock an object immediately because it is locked by another user.
The NOWAIT option is also available to pin calls by passing a value of OCI_LOCK_X_NOWAIT as the lock option parameter.
There are two options available for implementing optimistic locking in an OCI application.
The first optimistic locking option is for OCI applications that run transactions at the serializable level.
OCI supports calls that allow you to dereference and pin objects in the object cache without locking them, modify them in the cache (again without locking them), and then flush them (the dirtied objects) to the database.
During the flush, if a dirty object has been modified by another committed transaction since the beginning of your transaction, a non-serializable transaction error is returned. If none of the dirty objects has been modified by any other any other transaction since the beginning of your transaction, then the changes are written to the database successfully.
|
Note:
|
The above mechanism effectively implements an optimistic locking model.
Alternately, an application can enable object change detection mode. To do this, set the OCI_ATTR_OBJECT_DETECTCHANGE attribute of the environment handle to a value of TRUE.
When this mode has been activated, the application receives an ORA-08179 error ("concurrency check failed") when attempting to flush an object that has been changed in the server by another committed transaction. The application can then handle this error in an appropriate manner.
When a transaction is committed (OCITransCommit()), all marked objects are flushed to the server. If an object copy is pinned with a transaction duration, the object copy is unpinned.
When a transaction is rolled back, all marked objects are unmarked. If an object copy is pinned with a transaction duration, the object copy is unpinned.
In order to maintain free space in memory, the object cache attempts to reuse objects' memory whenever possible. The object cache reuses an object's memory when the object's lifetime (allocation duration) expires or when the object's pin duration expires. The allocation duration is set when an object is created with OCIObjectNew(), and the pin duration is set when an object is pinned with OCIObjectPin(). The datatype of the duration value is OCIDuration.
When an object reaches the end of its allocation duration, it is automatically deleted and its memory can be reused. The pin duration indicates when an object's memory can be reused, and memory is reused when the cache is full.
OCI supports two predefined durations:
The transaction duration expires when the containing transaction ends (commits or aborts). The session duration expires when the containing session/connection ends.
The application can explicitly unpin an object using OCIObjectUnpin(). To minimize explicit unpinning of individual objects, the application can unpin all objects currently pinned in the object cache using the function OCICacheUnpin(). By default, all objects are unpinned at the end of the pin duration.
Table 13-2 illustrates the use of the different durations in an application. Four objects are created or pinned in this application over the course of one connection and three transactions. The first column indicates the action performed by the database, and the second column indicates the function which performs the action. The remaining columns indicate the states of the various objects at each point in the application.
For example, Object 1 comes into existence at T2 when it is created with a connection duration, and it exists until T19 when the connection is terminated. Object 2 is pinned at T7 with a transaction duration, after being fetched at T6, and it remains pinned until T9 when the transaction is committed.
See Also:
|
An instance in memory is composed of a top-level memory chunk of the instance, a top-level memory of the null indicator structure and optionally, a number of secondary memory chunks. Consider a DEPARTMENT row type,
CREATE TYPE department AS OBJECT ( dep_name varchar2(20), budget number, manager person, /* person is an object type */ employees person_array ); /* varray of person objects */
and its C representation
struct department { OCIString * dep_name; OCINumber budget; struct person manager; OCIArray * employees; ); typedef struct department department;
Each instance of DEPARTMENT has a top-level memory chunk which contains the top-level attributes such as dep_name, budget, manager and employees. The attributes dep_name and employees are themselves actually pointers to the additional memory (the secondary memory chunks). The secondary memory is used to contain the actual data for the embedded instances (for example, employees varray and dep_name string).
The top-level memory of the null indicator structure contains the null statuses of the attributes in the top level memory chunk of the instance. From the above example, the top level memory of the null structure contains the null statuses of the attributes dep_name, budget, manager and the atomic nullity of employees.
This section discusses how OCI applications can navigate through graphs of objects in the object cache.
In the example in the previous sections, the object retrieved by the application was a simple object, whose attributes were all scalar values. If an application retrieves an object with an attribute which is a REF to another object, the application can use OCI calls to traverse the object graph and access the referenced instance.
As an example, consider the following declaration for a new type in the database:
CREATE TYPE person_t AS OBJECT ( name VARCHAR2(30), mother REF person_t, father REF person_t);
An object table of person_t objects is created with the following statement:
CREATE TABLE person_table OF person_t;
Instances of the person_t type can now be stored in the typed table. Each instance of person_t includes references to two other objects, which would also be stored in the table. A NULL reference could represent a parent about whom information is not available.
An object graph is a graphical representation of the REF links between object instances. For example, Figure 13-2, "Object Graph of person_t Instances" on the following page depicts an object graph of person_t instances, showing the links from one object to another. The circles represent objects, and the arrows represent references to other objects.
In this case, each object has links to two other instances of the same object. This need not always be the case. Objects may have links to other object types. Other types of graphs are also possible. For example, if a set of objects is implemented as a linked list, the object graph could be viewed as a simple chain, where each object references the previous and/or next objects in the linked list.
You can use the methods described earlier in this chapter to retrieve a reference to a person_t instance and then pin that instance. OCI provides functionality which enables you to traverse the object graph by following a reference from one object to another.
As an example, assume that an application fetches the person1 instance in the above graph and pins it as pers_1. Once that has been done, the application can access the mother instance of person1 and pin it into pers_2 through a second pin operation:
OCIObjectPin(env, err, pers_1->mother, OCI_PIN_ANY, OCI_DURATION_TRANS, OCI_LOCK_X, (OCIComplexObject *) 0, &pers_2);
In this case, an OCI fetch operation is not required to retrieve the second instance.
The application could then pin the father instance of person1, or it could operate on the reference links of person2.
This section provides a brief summary of the available OCI navigational functions. The functions are grouped according to their general functionality.
| See Also:
More detailed descriptions of each of these functions can be found in Chapter 17, "OCI Navigational and Type Functions" |
The use of these functions is described in the earlier sections of this chapter.
The navigational functions follow a naming scheme which uses different prefixes for different types of functionality:
OCICache*() - these functions are Cache operations
OCIObject*() - these functions are individual Object operations
The following functions are available to pin, unpin, or free objects:
The following functions are available to flush modified objects to the server:
The following functions allow an application to mark or unmark an object by modifying one of its meta-attributes:
The following functions allow an application to access the meta-attributes of an object:
The following functions provide additional object functionality for OCI applications:
When type information is requested based on the type name, OCI returns the type descriptor object (TDO) corresponding to the latest version of the type. Since there is no synchronization between the server and the object cache, the TDO in the object cache may not be current.
It is possible that when pinning an object, the version of the image differs from the TDO's version. Then, an error will be issued. It is up to you to abort the application or refresh the TDO and re-pin the object. Continuing with the application may cause the application to fail because even if the image and the TDO are at the same version, there is no guarantee that the object structure (that is, C struct) defined in the application is compatible with the new type version, especially for the case when an attribute has been dropped from the type in the server.
Thus, when the structure of a type is altered, you must regenerate the header files of the changed type, modify their application, re-compile and re-link before executing the program again.
|
 Copyright © 1996, 2002 Oracle Corporation. All Rights Reserved. |
|