2 Streams Capture Process

Row LCRs

A row LCR describes a change to the data in a single row or a change to a single LONG, LONG RAW, or LOB column in a row. The change results from a data manipulation language (DML) statement or a piecewise update to a LOB. For example, a single DML statement can insert or merge multiple rows into a table, can update multiple rows in a table, or can delete multiple rows from a table.

Therefore, a single DML statement can produce multiple row LCRs. That is, a capture process creates an LCR for each row that is changed by the DML statement. In addition, an update to a LONG, LONG RAW, or LOB column in a single row can result in more than one row LCR.

Each row LCR is encapsulated in an object of LCR$_ROW_RECORD type and contains the following attributes:

• source_database_name: The name of the source database where the row change occurred.

• command_type: The type of DML statement that produced the change, either INSERT, UPDATE, DELETE, LOB ERASE, LOB WRITE, or LOB TRIM.

• object_owner: The schema name that contains the table with the changed row.

• object_name: The name of the table that contains the changed row.

• tag: A raw tag that can be used to track the LCR.

• transaction_id: The identifier of the transaction in which the DML statement was run.

• scn: The system change number (SCN) at the time when the change record was written to the redo log.

• old_values: The old column values related to the change. These are the column values for the row before the DML change. If the type of the DML statement is UPDATE or DELETE, then these old values include some or all of the columns in the changed row before the DML statement. If the type of the DML statement is INSERT, then there are no old values.

• new_values: The new column values related to the change. These are the column values for the row after the DML change. If the type of the DML statement is UPDATE or INSERT, then these new values include some or all of the columns in the changed row after the DML statement. If the type of the DML statement is DELETE, then there are no new values.

A captured row LCR can also contain transaction control statements. These row LCRs contain directives such as COMMIT and ROLLBACK. Such row LCRs are internal and are used by an apply process to maintain transaction consistency between a source database and a destination database.

DDL LCRs

A DDL LCR describes a data definition language (DDL) change. A DDL statement changes the structure of the database. For example, a DDL statement can create, alter, or drop a database object.

Each DDL LCR contains the following information:

• source_database_name: The name of the source database where the DDL change occurred.

• command_type: The type of DDL statement that produced the change, for example ALTER TABLE or CREATE INDEX.

• object_owner: The schema name of the user who owns the database object on which the DDL statement was run.

• object_name: The name of the database object on which the DDL statement was run.

• object_type: The type of database object on which the DDL statement was run, for example TABLE or PACKAGE.

• ddl_text: The text of the DDL statement.

• logon_user: The logon user, which is the user whose session executed the DDL statement.

• current_schema: The schema that is used if no schema is specified for an object in the DDL text.

• base_table_owner: The base table owner. If the DDL statement is dependent on a table, then the base table owner is the owner of the table on which it is dependent.

• base_table_name: The base table name. If the DDL statement is dependent on a table, then the base table name is the name of the table on which it is dependent.

• tag: A raw tag that can be used to track the LCR.

• transaction_id: The identifier of the transaction in which the DDL statement was run.

• scn: The SCN when the change was written to the redo log.

  
  

  Note: Both row LCRs and DDL LCRs contain the source database name of the database where a change originated. If captured messages will be propagated by a propagation or applied by an apply process, then, to avoid propagation and apply problems, Oracle recommends that you do not rename the source database after a capture process has started capturing changes.

  
  

Extra Information in LCRs

In addition to the information discussed in the previous sections, row LCRs and DDL LCRs optionally can include the following extra information (or LCR attributes):

• row_id: The rowid of the row changed in a row LCR. This attribute is not included in DDL LCRs or row LCRs for index-organized tables.

• serial#: The serial number of the session that performed the change captured in the LCR.

• session#: The identifier of the session that performed the change captured in the LCR.

• thread#: The thread number of the instance in which the change captured in the LCR was performed. Typically, the thread number is relevant only in a Real Application Clusters environment.

• tx_name: The name of the transaction that includes the LCR.

• username: The name of the current user who performed the change captured in the LCR.

You can use the INCLUDE_EXTRA_ATTRIBUTE procedure in the DBMS_CAPTURE_ADM package to instruct a capture process to capture one or more extra attributes.

Types of DML Changes Captured

When you specify that DML changes made to certain tables should be captured, a capture process captures the following types of DML changes made to these tables:

• INSERT

• UPDATE

• DELETE

• MERGE

• Piecewise updates to LOBs

The following are considerations for capturing DML changes:

• A capture process converts each MERGE change into an INSERT or UPDATE change. MERGE is not a valid command type in a row LCR.

• A capture process can capture changes made to an index-organized table only if the index-organized table does not contain any columns of the following datatypes:

  • ROWID

  • UROWID

  • User-defined types (including object types, REFs, varrays, and nested tables)

  If an index-organized table contains a column of one of these datatypes, then a capture process raises an error when a user makes a change to the index-organized table and the change satisfies the capture process rule sets.

• A capture process ignores CALL, EXPLAIN PLAN, or LOCK TABLE statements.

• A capture process cannot capture DML changes made to temporary tables or object tables. A capture process raises an error if it attempts to capture such changes.

• If you share a sequence at multiple databases, then sequence values used for individual rows at these databases might vary. Also, changes to actual sequence values are not captured. For example, if a user references a NEXTVAL or sets the sequence, then a capture process does not capture changes resulting from these operations.

  
  

  See Also: "Datatypes Captured" for information about the datatypes supported by a capture process Chapter 6, "How Rules Are Used in Streams" for more information about rule sets for Streams clients and for information about how messages satisfy rule sets Oracle Streams Replication Administrator's Guide for information about applying DML changes with an apply process and for information about strategies to avoid having the same sequence-generated value for two different rows at different databases Oracle XML DB Developer's Guide for information about SQL functions that update XML data

  
  

DDL Changes and Capture Processes

A capture process captures the DDL changes that satisfy its rule sets, except for the following types of DDL changes:

• ALTER DATABASE

• CREATE CONTROLFILE

• CREATE DATABASE

• CREATE PFILE

• CREATE SPFILE

• FLASHBACK DATABASE

A capture process can capture DDL statements, but not the results of DDL statements, unless the DDL statement is a CREATE TABLE AS SELECT statement. For example, when a capture process captures an ANALYZE statement, it does not capture the statistics generated by the ANALYZE statement. However, when a capture process captures a CREATE TABLE AS SELECT statement, it captures the statement itself and all of the rows selected (as INSERT row LCRs).

Some types of DDL changes that are captured by a capture process cannot be applied by an apply process. If an apply process receives a DDL LCR that specifies an operation that cannot be applied, then the apply process ignores the DDL LCR and records information about it in the trace file for the apply process.

When a capture process captures a DDL change that specifies timestamps or system change number (SCN) values in its syntax, configure a DDL handler for any apply processes that will dequeue the change. The DDL handler must process timestamp or SCN values properly. For example, although a capture process always ignores FLASHBACK DATABASE statements, a capture process captures FLASHBACK TABLE statements when its rule sets instruct it to capture DDL changes to the specified table. FLASHBACK TABLE statements include timestamps or SCN values in its syntax.

Other Types of Changes Ignored by a Capture Process

The following types of changes are ignored by a capture process:

• The session control statements ALTER SESSION and SET ROLE.

• The system control statement ALTER SYSTEM.

• Invocations of PL/SQL procedures, which means that a call to a PL/SQL procedure is not captured. However, if a call to a PL/SQL procedure causes changes to database objects, then these changes can be captured by a capture process if the changes satisfy the capture process rule sets.

• Changes made to a table or schema by online redefinition using the DBMS_REDEFINITION package. Online table redefinition is supported on a table for which a capture process captures changes, but the logical structure of the table before online redefinition must be the same as the logical structure after online redefinition.

NOLOGGING and UNRECOVERABLE Keywords for SQL Operations

If you use the NOLOGGING or UNRECOVERABLE keyword for a SQL operation, then the changes resulting from the SQL operation cannot be captured by a capture process. Therefore, do not use these keywords if you want to capture the changes that result from a SQL operation.

If the object for which you are specifying the logging attributes resides in a database or tablespace in FORCE LOGGING mode, then Oracle ignores any NOLOGGING or UNRECOVERABLE setting until the database or tablespace is taken out of FORCE LOGGING mode. You can determine the current logging mode for a database by querying the FORCE_LOGGING column in the V$DATABASE dynamic performance view. You can determine the current logging mode for a tablespace by querying the FORCE_LOGGING column in the DBA_TABLESPACES static data dictionary view.

UNRECOVERABLE Clause for Direct Path Loads

If you use the UNRECOVERABLE clause in the SQL*Loader control file for a direct path load, then the changes resulting from the direct path load cannot be captured by a capture process. Therefore, if the changes resulting from a direct path load should be captured by a capture process, then do not use the UNRECOVERABLE clause.

If you perform a direct path load without logging changes at a source database, but you do not perform a similar direct path load at the destination databases of the source database, then apply errors can result at these destination databases when changes are made to the loaded objects at the source database. In this case, a capture process at the source database can capture changes to these objects, and one or more propagations can propagate the changes to the destination databases. When an apply process tries to apply these changes, errors result unless both the changed object and the changed rows in the object exist on the destination database.

Therefore, if you use the UNRECOVERABLE clause for a direct path load and a capture process is configured to capture changes to the loaded objects, then make sure any destination databases contain the loaded objects and the loaded data to avoid apply errors. One way to make sure that these objects exist at the destination databases is to perform a direct path load at each of these destination databases that is similar to the direct path load performed at the source database.

If you load objects into a database or tablespace that is in FORCE LOGGING mode, then Oracle ignores any UNRECOVERABLE clause during a direct path load, and the loaded changes are logged. You can determine the current logging mode for a database by querying the FORCE_LOGGING column in the V$DATABASE dynamic performance view. You can determine the current logging mode for a tablespace by querying the FORCE_LOGGING column in the DBA_TABLESPACES static data dictionary view.

Local Capture

Local capture means that a capture process runs on the source database. Figure 2-1 shows a database using local capture.

Downstream Capture

Downstream capture means that a capture process runs on a database other than the source database. The following types of downstream capture configurations are possible: real-time downstream capture and archived-log downstream capture. The DOWNSTREAM_REAL_TIME_MINE capture process parameter controls whether a downstream capture process performs real-time downstream capture or archived-log downstream capture. A real-time downstream capture process and one or more archived-log downstream capture processes can coexist at a downstream database.

Real-Time Downstream Capture

A real-time downstream capture configuration works in the following way:

• Redo transport services use the log writer process (LGWR) at the source database to send redo data to the downstream database either synchronously or asynchronously. At the same time, the LGWR records redo data in the online redo log at the source database.

• A remote file server process (RFS) at the downstream database receives the redo data over the network and stores the redo data in the standby redo log.

• A log switch at the source database causes a log switch at the downstream database, and the ARCHn process at the downstream database archives the current standby redo log file.

• The real-time downstream capture process captures changes from the standby redo log whenever possible and from the archived standby redo log files whenever necessary. A capture process can capture changes in the archived standby redo log files if it falls behind. When it catches up, it resumes capturing changes from the standby redo log.

Figure 2-2 Real-Time Downstream Capture

Description of the illustration strms047.gif

The advantage of real-time downstream capture over archived-log downstream capture is that real-time downstream capture reduces the amount of time required to capture changes made at the source database. The time is reduced because the real-time capture process does not need to wait for the redo log file to be archived before it can capture data from it.

Note: Only one real-time downstream capture process can exist at a downstream database.

Archived-Log Downstream Capture

A archived-log downstream capture process configuration means that archived redo log files from the source database are copied to the downstream database, and the capture process captures changes in these archived redo log files. You can copy the archived redo log files to the downstream database using redo transport services, the DBMS_FILE_TRANSFER package, file transfer protocol (FTP), or some other mechanism.

Figure 2-3 Archived-Log Downstream Capture

Description of the illustration strms037.gif

Note: As illustrated in Figure 2-3, the source database for a change captured by a downstream capture process is the database where the change was recorded in the redo log, not the database running the downstream capture process.

The advantage of archived-log downstream capture over real-time downstream capture is that archived-log downstream capture allows multiple downstream capture processes at a downstream database. You can copy redo log files from multiple source databases to a single downstream database and configure multiple archived-log downstream capture processes to capture changes in these redo log files.

See Also: Oracle Data Guard Concepts and Administration for more information about redo transport services

Captured SCN and Applied SCN

The captured SCN is the SCN that corresponds to the most recent change scanned in the redo log by a capture process. The applied SCN for a capture process is the SCN of the most recent message dequeued by the relevant apply processes. All messages lower than this SCN have been dequeued by all apply processes that apply changes captured by the capture process. The applied SCN for a capture process is equivalent to the low-watermark SCN for an apply process that applies changes captured by the capture process.

First SCN and Start SCN

This section describes the first SCN and start SCN for a capture process.

COLUMN FIRST_CHANGE# HEADING 'First SCN' FORMAT 999999999
COLUMN NAME HEADING 'Log File Name' FORMAT A50

SELECT DISTINCT FIRST_CHANGE#, NAME FROM V$ARCHIVED_LOG
  WHERE DICTIONARY_BEGIN = 'YES';

Capture Process Components

A capture process consists of the following components:

• One reader server that reads the redo log and divides the redo log into regions.

• One or more preparer servers that scan the regions defined by the reader server in parallel and perform prefiltering of changes found in the redo log. Prefiltering involves sending partial information about changes, such as schema and object name for a change, to the rules engine for evaluation, and receiving the results of the evaluation.

• One builder server that merges redo records from the preparer servers. These redo records either evaluated to TRUE during partial evaluation or partial evaluation was inconclusive for them. The builder server preserves the SCN order of these redo records and passes the merged redo records to the capture process.

• The capture process (cnnn) performs the following actions for each change when it receives merged redo records from the builder server:

  • Formats the change into an LCR

  • If the partial evaluation performed by a preparer server was inconclusive for the change in the LCR, then sends the LCR to the rules engine for full evaluation

  • Receives the results of the full evaluation of the LCR if it was performed

  • Enqueues the LCR into the queue associated with the capture process if the LCR satisfies the rules in the positive rule set for the capture process, or discards the LCR if it satisfies the rules in the negative rule set for the capture process or if it does not satisfy the rules in the positive rule set

Each reader server, preparer server, and builder server is a parallel execution server. A capture process (cnnn) is an Oracle background process.

Capture Process States

The state of a capture process describes what the capture process is doing currently. You can view the state of a capture process by querying the STATE column in the V$STREAMS_CAPTURE dynamic performance view. The following capture process states are possible:

• INITIALIZING - Starting up.

• WAITING FOR DICTIONARY REDO - Waiting for redo log files containing the dictionary build related to the first SCN to be added to the capture process session. A capture process cannot begin to scan the redo log files until all of the log files containing the dictionary build have been added.

• DICTIONARY INITIALIZATION - Processing a dictionary build.

• MINING (PROCESSED SCN = scn_value) - Mining a dictionary build at the SCN scn_value.

• LOADING (step X of Y) - Processing information from a dictionary build and currently at step X in a process that involves Y steps, where X and Y are numbers.

• CAPTURING CHANGES - Scanning the redo log for changes that evaluate to TRUE against the capture process rule sets.

• WAITING FOR REDO - Waiting for new redo log files to be added to the capture process session. The capture process has finished processing all of the redo log files added to its session. This state is possible if there is no activity at a source database. For a downstream capture process, this state is possible if the capture process is waiting for new log files to be added to its session.

• EVALUATING RULE - Evaluating a change against a capture process rule set.

• CREATING LCR - Converting a change into an LCR.

• ENQUEUING MESSAGE - Enqueuing an LCR that satisfies the capture process rule sets into the capture process queue.

• PAUSED FOR FLOW CONTROL - Unable to enqueue LCRs either because of low memory or because propagations and apply processes are consuming messages slower than the capture process is creating them. This state indicates flow control that is used to reduce spilling of captured messages when propagation or apply has fallen behind or is unavailable.

• SHUTTING DOWN - Stopping.

Multiple Capture Processes in a Single Database

If you run multiple capture processes in a single database, consider increasing the size of the System Global Area (SGA) for each instance. Use the SGA_MAX_SIZE initialization parameter to increase the SGA size. Also, if the size of the Streams pool is not managed automatically in the database, then you should increase the size of the Streams pool by 10 MB for each capture process parallelism. For example, if you have two capture processes running in a database, and the parallelism parameter is set to 4 for one of them and 1 for the other, then increase the Streams pool by 50 MB (4 + 1 = 5 parallelism).

Also, Oracle recommends that each ANYDATA queue used by a capture process, propagation, or apply process have captured messages from at most one capture process from a particular source database. Therefore, a separate queue should be used for each capture process that captures changes originating at a particular source database.

Capture Process Checkpoints

A checkpoint is information about the current state of a capture process that is stored persistently in the data dictionary of the database running the capture process. A capture process tries to record a checkpoint at regular intervals called checkpoint intervals.

Checkpoint Retention Time

The checkpoint retention time is the amount of time, in number of days, that a capture process retains checkpoints before purging them automatically. A capture process periodically computes the age of a checkpoint by subtracting the NEXT_TIME of the archived redo log that corresponds to the checkpoint from FIRST_TIME of the archived redo log file containing the required checkpoint SCN for the capture process. If the resulting value is greater than the checkpoint retention time, then the capture process automatically purges the checkpoint by advancing its first SCN value. Otherwise, the checkpoint is retained. The DBA_REGISTERED_ARCHIVED_LOG view displays the FIRST_TIME and NEXT_TIME for archived redo log files, and the REQUIRED_CHECKPOINT_SCN column in the DBA_CAPTURE view displays the required checkpoint SCN for a capture process. Figure 2-4 shows an example of a checkpoint being purged when the checkpoint retention time is set to 20 days.

Figure 2-4 Checkpoint Retention Time Set to 20 Days

Description of the illustration strms051.gif

In Figure 2-4, with the checkpoint retention time set to 20 days, the checkpoint at SCN 435250 is purged because it is 21 days old, while the checkpoint at SCN 479315 is retained because it is 8 days old.

Whenever the first SCN is reset for a capture process, the capture process purges information about archived redo log files prior to the new first SCN from its LogMiner data dictionary. After this information is purged, the archived redo log files remain on the hard disk, but the files are not needed by the capture process. The PURGEABLE column in the DBA_REGISTERED_ARCHIVED_LOG view displays YES for the archived redo log files that are no longer needed. These files can be removed from disk or moved to another location without affecting the capture process.

If you create a capture process using the CREATE_CAPTURE procedure in the DBMS_CAPTURE_ADM package, then you can specify the checkpoint retention time, in days, using the checkpoint_retention_time parameter. The default checkpoint retention time is 60 days if the checkpoint_retention_time parameter is not specified in the CREATE_CAPTURE procedure, or if you use the DBMS_STREAMS_ADM package to create the capture process. The CHECKPOINT_RETENTION_TIME column in the DBA_CAPTURE view displays the current checkpoint retention time for a capture process.

You can change the checkpoint retention time for a capture process by specifying a new time in the ALTER_CAPTURE procedure in the DBMS_CAPTURE_ADM package. If you do not want checkpoints for a capture process to be purged automatically, then specify DBMS_CAPTURE_ADM.INFINITE for the checkpoint_retention_time parameter in CREATE_CAPTURE or ALTER_CAPTURE.

Note: To specify a checkpoint retention time for a capture process, the compatibility level of the database running the capture process must be 10.2.0 or higher. If the compatibility level is lower than 10.2.0 for a database, then the checkpoint retention time for all capture processes running on the database is infinite.

See Also: "The LogMiner Data Dictionary for a Capture Process" "First SCN and Start SCN Specifications During Capture Process Creation" "A New First SCN Value and Purged LogMiner Data Dictionary Information" "Managing the Checkpoint Retention Time for a Capture Process" Oracle Database PL/SQL Packages and Types Reference for more information about the CREATE_CAPTURE and ALTER_CAPTURE procedures

Capture Process Creation

You can create a capture process using the DBMS_STREAMS_ADM package or the DBMS_CAPTURE_ADM package. Using the DBMS_STREAMS_ADM package to create a capture process is simpler because defaults are used automatically for some configuration options. In addition, when you use the DBMS_STREAMS_ADM package, a rule set is created for the capture process and rules can be added to the rule set automatically. The rule set is a positive rule set if the inclusion_rule parameter is set to true (the default), or it is a negative rule set if the inclusion_rule parameter is set to false.

Alternatively, using the DBMS_CAPTURE_ADM package to create a capture process is more flexible, and you create one or more rule sets and rules for the capture process either before or after it is created. You can use the procedures in the DBMS_STREAMS_ADM package or the DBMS_RULE_ADM package to add rules to a rule set for the capture process. To create a capture process at a downstream database, you must use the DBMS_CAPTURE_ADM package.

When you create a capture process using a procedure in the DBMS_STREAMS_ADM package and generate one or more rules in the positive rule set for the capture process, the objects for which changes are captured are prepared for instantiation automatically, unless it is a downstream capture process and there is no database link from the downstream database to the source database.

When you create a capture process using the CREATE_CAPTURE procedure in the DBMS_CAPTURE_ADM package, you should prepare for instantiation any objects for which you plan to capture changes as soon as possible after capture process creation. You can prepare objects for instantiation using one of the following procedures in the DBMS_CAPTURE_ADM package:

• PREPARE_TABLE_INSTANTIATION prepares a single table for instantiation.

• PREPARE_SCHEMA_INSTANTIATION prepares for instantiation all of the objects in a schema and all objects added to the schema in the future.

• PREPARE_GLOBAL_INSTANTIATION prepares for instantiation all of the objects in a database and all objects added to the database in the future.

These procedures can also enable supplemental logging for the key columns or for all columns in the table or tables prepared for instantiation.

The LogMiner Data Dictionary for a Capture Process

A capture process requires a data dictionary that is separate from the primary data dictionary for the source database. This separate data dictionary is called a LogMiner data dictionary. There can be more than one LogMiner data dictionary for a particular source database. If there are multiple capture processes capturing changes from the source database, then two or more capture processes can share a LogMiner data dictionary, or each capture process can have its own LogMiner data dictionary. If the LogMiner data dictionary needed by a capture process does not exist, then the capture process populates it using information in the redo log when the capture process is started for the first time.

The DBMS_CAPTURE_ADM.BUILD procedure extracts data dictionary information to the redo log, and this procedure must be run at least once on the source database before any capture process capturing changes originating at the source database is started. The extracted data dictionary information in the redo log is consistent with the primary data dictionary at the time when the DBMS_CAPTURE_ADM.BUILD procedure is run. This procedure also identifies a valid first SCN value that can be used to create a capture process.

You can perform a build of data dictionary information in the redo log multiple times, and a particular build might or might not be used by a capture process to create a LogMiner data dictionary. The amount of information extracted to a redo log when you run the BUILD procedure depends on the number of database objects in the database. Typically, the BUILD procedure generates a large amount of redo data that a capture process must scan subsequently. Therefore, you should run the BUILD procedure only when necessary.

In most cases, if a build is required when a capture process is created using a procedure in the DBMS_STREAMS_ADM or DBMS_CAPTURE_ADM package, then the procedure runs the BUILD procedure automatically. However, the BUILD procedure is not run automatically during capture process creation in the following cases:

• You use CREATE_CAPTURE and specify a non-NULL value for the first_scn parameter. In this case, the specified first SCN must correspond to a previous build.

• You create a downstream capture process that does not use a database link. In this case, the command at the downstream database cannot communicate with the source database to run the BUILD procedure automatically. Therefore, you must run it manually on the source database and specify the first SCN that corresponds to the build during capture process creation.

A capture process requires a LogMiner data dictionary because the information in the primary data dictionary might not apply to the changes being captured from the redo log. These changes might have occurred minutes, hours, or even days before they are captured by a capture process. For example, consider the following scenario:

1. A capture process is configured to capture changes to tables.

2. A database administrator stops the capture process. When the capture process is stopped, it records the SCN of the change it was currently capturing.

3. User applications continue to make changes to the tables while the capture process is stopped.

4. The capture process is restarted three hours after it was stopped.

In this case, to ensure data consistency, the capture process must begin capturing changes in the redo log at the time when it was stopped. The capture process starts capturing changes at the SCN that it recorded when it was stopped.

The redo log contains raw data. It does not contain database object names and column names in tables. Instead, it uses object numbers and internal column numbers for database objects and columns, respectively. Therefore, when a change is captured, a capture process must reference a data dictionary to determine the details of the change.

Because a LogMiner data dictionary might be populated when a capture process is started for the first time, it might take some time to start capturing changes. The amount of time required depends on the number of database objects in the database. You can query the STATE column in the V$STREAMS_CAPTURE dynamic performance view to monitor the progress while a capture process is processing a data dictionary build.

See Also: "Capture Process Rule Evaluation" "First SCN and Start SCN" "Capture Process States" Oracle Streams Replication Administrator's Guide for more information about preparing database objects for instantiation

Scenario Illustrating Why a Capture Process Needs a LogMiner Data Dictionary

Consider a scenario in which a capture process has been configured to capture changes to table t1, which has columns a and b, and the following changes are made to this table at three different points in time:

Time 1: Insert values a=7 and b=15.

Time 2: Add column c.

Time 3: Drop column b.

If for some reason the capture process is capturing changes from an earlier time, then the primary data dictionary and the relevant version in the LogMiner data dictionary contain different information. Table 2-1 illustrates how the information in the LogMiner data dictionary is used when the current time is different than the change capturing time.

Table 2-1 Information About Table t1 in the Primary and LogMiner Data Dictionaries

Current Time	Change Capturing Time	Primary Data Dictionary	LogMiner Data Dictionary
1	1	Table t1 has columns a and b.	Table t1 has columns a and b at time 1.
2	1	Table t1 has columns a, b, and c.	Table t1 has columns a and b at time 1.
3	1	Table t1 has columns a and c.	Table t1 has columns a and b at time 1.

Assume that the capture process captures the change resulting from the insert at time 1 when the actual time is time 3. If the capture process used the primary data dictionary, then it might assume that a value of 7 was inserted into column a and a value of 15 was inserted into column c, because those are the two columns for table t1 at time 3 in the primary data dictionary. However, a value of 15 actually was inserted into column b, not column c.

Because the capture process uses the LogMiner data dictionary, the error is avoided. The LogMiner data dictionary is synchronized with the capture process and continues to record that table t1 has columns a and b at time 1. So, the captured change specifies that a value of 15 was inserted into column b.

Multiple Capture Processes for the Same Source Database

If one or more capture processes are capturing changes made to a source database, and you want to create a new capture process that captures changes to the same source database, then the new capture process can either create a new LogMiner data dictionary or share one of the existing LogMiner data dictionaries with one or more other capture processes. Whether a new LogMiner data dictionary is created for a new capture process depends on the setting for the first_scn parameter when you run CREATE_CAPTURE to create a capture process:

• If you specify NULL for the first_scn parameter, then the new capture process attempts to share a LogMiner data dictionary with one or more existing capture processes that capture changes from the same source database. NULL is the default for the first_scn parameter.

• If you specify a non-NULL value for the first_scn parameter, then the new capture process uses a new LogMiner data dictionary that is created when the new capture process is started for the first time.

  
  

  Note: When you create a capture process and specify a non-NULL first_scn parameter value, this value should correspond to a data dictionary build in the redo log obtained by running the DBMS_CAPTURE_ADM.BUILD procedure. During capture process creation, if the first_scn parameter is NULL and the start_scn parameter is non-NULL, then an error is raised if the start_scn parameter setting is lower than all of the first SCN values for all existing capture processes.

  
  

If multiple LogMiner data dictionaries exist, and you specify NULL for the first_scn parameter during capture process creation, then the new capture process automatically attempts to share the LogMiner data dictionary of one of the existing capture processes that has taken at least one checkpoint. You can view the maximum checkpoint SCN for all existing capture processes by querying the MAX_CHECKPOINT_SCN column in the DBA_CAPTURE data dictionary view.

If multiple LogMiner data dictionaries exist, and you specify a non-NULL value for the first_scn parameter during capture process creation, then the new capture process creates a new LogMiner data dictionary the first time it is started. In this case, before you create the new capture process, you must run the BUILD procedure in the DBMS_CAPTURE_ADM package on the source database. The BUILD procedure generates a corresponding valid first scn value that you can specify when you create the new capture process. You can find a first SCN generated by the BUILD procedure by running the following query:

COLUMN FIRST_CHANGE# HEADING 'First SCN' FORMAT 999999999
COLUMN NAME HEADING 'Log File Name' FORMAT A50

SELECT DISTINCT FIRST_CHANGE#, NAME FROM V$ARCHIVED_LOG
  WHERE DICTIONARY_BEGIN = 'YES';

This query can return more than one row if the BUILD procedure was run more than once.

The most important factor to consider when deciding whether a new capture process should share an existing LogMiner data dictionary or create a new one is the difference between the maximum checkpoint SCN values of the existing capture processes and the start SCN of the new capture process. If the new capture process shares a LogMiner data dictionary, then it must scan the redo log from the point of the maximum checkpoint SCN of the shared LogMiner data dictionary onward, even though the new capture process cannot capture changes prior to its first SCN. If the start SCN of the new capture process is much higher than the maximum checkpoint SCN of the existing capture process, then the new capture process must scan a large amount of redo data before it reaches its start SCN.

A capture process creates a new LogMiner data dictionary when the first_scn parameter is non-NULL during capture process creation. Follow these guidelines when you decide whether a new capture process should share an existing LogMiner data dictionary or create a new one:

• If one or more maximum checkpoint SCN values is greater than the start SCN you want to specify, and if this start SCN is greater than the first SCN of one or more existing capture processes, then it might be better to share the LogMiner data dictionary of an existing capture process. In this case, you can assume there is a checkpoint SCN that is less than the start SCN and that the difference between this checkpoint SCN and the start SCN is small. The new capture process will begin scanning the redo log from this checkpoint SCN and will catch up to the start SCN quickly.

• If no maximum checkpoint SCN is greater than the start SCN, and if the difference between the maximum checkpoint SCN and the start SCN is small, then it might be better to share the LogMiner data dictionary of an existing capture process. The new capture process will begin scanning the redo log from the maximum checkpoint SCN, but it will catch up to the start SCN quickly.

• If no maximum checkpoint SCN is greater than the start SCN, and if the difference between the highest maximum checkpoint SCN and the start SCN is large, then it might take a long time for the capture process to catch up to the start SCN. In this case, it might be better for the new capture process to create a new LogMiner data dictionary. It will take some time to create the new LogMiner data dictionary when the new capture process is first started, but the capture process can specify the same value for its first SCN and start SCN, and thereby avoid scanning a large amount of redo data unnecessarily.

Figure 2-5 illustrates these guidelines.

Figure 2-5 Deciding Whether to Share a LogMiner Data Dictionary

Description of the illustration strms042.gif

Note: If you create a capture process using one of the procedures in the DBMS_STREAMS_ADM package, then it is the same as specifying NULL for the first_scn and start_scn parameters in the CREATE_CAPTURE procedure. You must prepare database objects for instantiation if a new capture process will capture changes made to these database objects. This requirement holds even if the new capture process shares a LogMiner data dictionary with one or more other capture processes for which these database objects have been prepared for instantiation.

See Also: "First SCN and Start SCN" "Capture Process Checkpoints"

A New First SCN Value and Purged LogMiner Data Dictionary Information

When you reset the first SCN value for an existing capture process, Oracle automatically purges LogMiner data dictionary information prior to the new first SCN setting. If the start SCN for a capture process corresponds to information that has been purged, then Oracle automatically resets the start SCN to the same value as the first SCN. However, if the start SCN is higher than the new first SCN setting, then the start SCN remains unchanged.

Figure 2-6 shows how Oracle automatically purges LogMiner data dictionary information prior to a new first SCN setting, and how the start SCN is not changed if it is higher than the new first SCN setting.

Figure 2-6 Start SCN Higher than Reset First SCN

Description of the illustration strms038.gif

Given this example, if the first SCN is reset again to a value higher than the start SCN value for a capture process, then the start SCN no longer corresponds to existing information in the LogMiner data dictionary. Figure 2-7 shows how Oracle resets the start SCN automatically if it is lower than a new first SCN setting.

Figure 2-7 Start SCN Lower than Reset First SCN

Description of the illustration strms039.gif

As you can see, the first SCN and start SCN for a capture process can continually increase over time, and, as the first SCN moves forward, it might no longer correspond to an SCN established by the DBMS_CAPTURE_ADM.BUILD procedure.

The Streams Data Dictionary

Propagations and apply processes use a Streams data dictionary to keep track of the database objects from a particular source database. A Streams data dictionary is populated whenever one or more database objects are prepared for instantiation at a source database. Specifically, when a database object is prepared for instantiation, it is recorded in the redo log. When a capture process scans the redo log, it uses this information to populate the local Streams data dictionary for the source database. In the case of local capture, this Streams data dictionary is at the source database. In the case of downstream capture, this Streams data dictionary is at the downstream database.

When you prepare a database object for instantiation, you are informing Streams that information about the database object is needed by propagations that propagate changes to the database object and apply processes that apply changes to the database object. Any database that propagates or applies these changes requires a Streams data dictionary for the source database where the changes originated.

After an object has been prepared for instantiation, the local Streams data dictionary is updated when a DDL statement on the object is processed by a capture process. In addition, an internal message containing information about this DDL statement is captured and placed in the queue for the capture process. Propagations can then propagate these internal messages to destination queues at databases.

A Streams data dictionary is multiversioned. If a database has multiple propagations and apply processes, then all of them use the same Streams data dictionary for a particular source database. A database can contain only one Streams data dictionary for a particular source database, but it can contain multiple Streams data dictionaries if it propagates or applies changes from multiple source databases.

ARCHIVELOG Mode and a Capture Process

The following list describes how different types of capture processes read the redo data:

• A local capture process reads online redo logs whenever possible and archived redo log files otherwise. Therefore, the source database must be running in ARCHIVELOG mode when a local capture process is configured to capture changes.

• A real-time downstream capture process reads online redo data from its source database whenever possible and archived redo log files that contain redo data from the source database otherwise. In this case, the redo data from the source database is stored in the standby redo log at the downstream database, and the archiver at the downstream database archives the redo data in the standby redo log. Therefore, both the source database and the downstream database must be running in ARCHIVELOG mode when a real-time downstream capture process is configured to capture changes.

• An archived-log downstream capture process always reads archived redo log files from its source database. Therefore, the source database must be running in ARCHIVELOG mode when an archived-log downstream capture process is configured to capture changes.

You can query the REQUIRED_CHECKPOINT_SCN column in the DBA_CAPTURE data dictionary view to determine the required checkpoint SCN for a capture process. When the capture process is restarted, it scans the redo log from the required checkpoint SCN forward. Therefore, the redo log file that includes the required checkpoint SCN, and all subsequent redo log files, must be available to the capture process.

You must keep an archived redo log file available until you are certain that no capture process will need that file. The first SCN for a capture process can be reset to a higher value, but it cannot be reset to a lower value. Therefore, a capture process will never need the redo log files that contain information prior to its first SCN. Query the DBA_LOGMNR_PURGED_LOG data dictionary view to determine which archived redo log files will never be needed by any capture process.

When a local capture process falls behind, there is a seamless transition from reading an online redo log to reading an archived redo log, and, when a local capture process catches up, there is a seamless transition from reading an archived redo log to reading an online redo log. Similarly, when a real-time downstream capture process falls behind, there is a seamless transition from reading the standby redo log to reading an archived redo log, and, when a real-time downstream capture process catches up, there is a seamless transition from reading an archived redo log to reading the standby redo log.

Capture Process Parameters

After creation, a capture process is disabled so that you can set the capture process parameters for your environment before starting it for the first time. Capture process parameters control the way a capture process operates. For example, the time_limit capture process parameter specifies the amount of time a capture process runs before it is shut down automatically.

Capture Process Rule Evaluation

A capture process evaluates changes it finds in the redo log against its positive and negative rule sets. The capture process evaluates a change against the negative rule set first. If one or more rules in the negative rule set evaluate to TRUE for the change, then the change is discarded, but if no rule in the negative rule set evaluates to TRUE for the change, then the change satisfies the negative rule set. When a change satisfies the negative rule set for a capture process, the capture process evaluates the change against its positive rule set. If one or more rules in the positive rule set evaluate to TRUE for the change, then the change satisfies the positive rule set, but if no rule in the positive rule set evaluates to TRUE for the change, then the change is discarded. If a capture process only has one rule set, then it evaluates changes against this one rule set only.

A running capture process completes the following series of actions to capture changes:

1. Finds changes in the redo log.

2. Performs prefiltering of the changes in the redo log. During this step, a capture process evaluates rules in its rule sets at a basic level to place changes found in the redo log into two categories: changes that should be converted into LCRs and changes that should not be converted into LCRs. Prefiltering is done in two phases. In the first phase, information that can be evaluated during prefiltering includes schema name, object name, and command type. If more information is needed to determine whether a change should be converted into an LCR, then information that can be evaluated during the second phase of prefiltering includes tag values and column values when appropriate.

   Prefiltering is a safe optimization done with incomplete information. This step identifies relevant changes to be processed subsequently, such that:

   • A capture process converts a change into an LCR if the change satisfies the capture process rule sets. In this case, proceed to Step 3.

   • A capture process does not convert a change into an LCR if the change does not satisfy the capture process rule sets.

   • Regarding MAYBE evaluations, the rule evaluation proceeds as follows:

     • If a change evaluates to MAYBE against both the positive and negative rule set for a capture process, then the capture process might not have enough information to determine whether the change will definitely satisfy both of its rule sets. In this case, further evaluation is necessary. Proceed to Step 3.

     • If the change evaluates to FALSE against the negative rule set and MAYBE against the positive rule set for the capture process, then the capture process might not have enough information to determine whether the change will definitely satisfy both of its rule sets. In this case, further evaluation is necessary. Proceed to Step 3.

     • If the change evaluates to MAYBE against the negative rule set and TRUE against the positive rule set for the capture process, then the capture process might not have enough information to determine whether the change will definitely satisfy both of its rule sets. In this case, further evaluation is necessary. Proceed to Step 3.

     • If the change evaluates to TRUE against the negative rule set and MAYBE against the positive rule set for the capture process, then the capture process discards the change.

     • If the change evaluates to MAYBE against the negative rule set and FALSE against the positive rule set for the capture process, then the capture process discards the change.

3. Converts changes that satisfy, or might satisfy, the capture process rule sets into LCRs based on prefiltering.

4. Performs LCR filtering. During this step, a capture process evaluates rules regarding information in each LCR to separate the LCRs into two categories: LCRs that should be enqueued and LCRs that should be discarded.

5. Discards the LCRs that should not be enqueued because they did not satisfy the capture process rule sets.

6. Enqueues the remaining captured messages into the queue associated with the capture process.

For example, suppose the following rule is defined in the positive rule set for a capture process: Capture changes to the hr.employees table where the department_id is 50. No other rules are defined for the capture process, and the parallelism parameter for the capture process is set to 1.

Given this rule, suppose an UPDATE statement on the hr.employees table changes 50 rows in the table. The capture process performs the following series of actions for each row change:

1. Finds the next change resulting from the UPDATE statement in the redo log.

2. Determines that the change resulted from an UPDATE statement to the hr.employees table and must be captured. If the change was made to a different table, then the capture process ignores the change.

3. Captures the change and converts it into an LCR.

4. Filters the LCR to determine whether it involves a row where the department_id is 50.

5. Either enqueues the LCR into the queue associated with the capture process if it involves a row where the department_id is 50, or discards the LCR if it involves a row where the department_id is not 50 or is missing.

   
   

   See Also: "Capture Process Components" Chapter 6, "How Rules Are Used in Streams" for more information about rule sets for Streams clients and for information about how messages satisfy rule sets

   
   

Figure 2-8 illustrates capture process rule evaluation in a flowchart.

Figure 2-8 Flowchart Showing Capture Process Rule Evaluation

Description of the illustration strms026.gif

Persistent Capture Process Status Upon Database Restart

A capture process maintains a persistent status when the database running the capture process is shut down and restarted. For example, if a capture process is enabled when the database is shut down, then the capture process automatically starts when the database is restarted. Similarly, if a capture process is disabled or aborted when a database is shut down, then the capture process is not started and retains the disabled or aborted status when the database is restarted.