This chapter contains instructions for preparing an Oracle source or target database environment for replication. The tasks outlined in this chapter must be performed before you perform initial synchronization to start replication in your environment. Perform all tasks that apply to your data and to your replication objectives.
SharePlex captures from the online and archived Oracle redo logs. SharePlex supports redo logs and datafiles that are stored on raw devices, on filesystem devices, and ASM instances.
Minimal supplemental logging must be set before you activate the SharePlex replication configuration.
In addition, SharePlex recommends that both primary key and unique key supplemental logging be set, or that a supplemental log group on unique columns be defined for every table in replication. When the key column values for a row update are in the redo log, SharePlex does not have to fetch them from the database. On a busy system, this greatly improves the performance of the Read process. Some SharePlex features require primary key and unique key logging to be enabled.
ALTER TABLE DDL commands that change the rowid of a table can affect subsequent DML operations if the primary or unique keys of the tables in replication are not being logged. When the keys are not logged, SharePlex fetches their values based on the rowid. Any operation that changes the rowid, such as ALTER TABLE...MOVE, can cause the wrong key values to be used for subsequent DML operations.
See Set up Oracle database objects for replication for more information about defining key values.
When primary and unique key supplemental logging is enabled and a table does not have a primary key, Oracle has to decide what type of unique key to log. If the table has multiple unique keys, Oracle determines the best key to use and logs those column values for every UPDATE. If a table does not have any type of key, Oracle logs all columns that are not a LONG or a LOB.
SharePlex must also identify a key to use to replicate data. Like Oracle, SharePlex chooses a key in this order:
If a table being replicated by SharePlex does not have a primary key, but does have multiple unique keys, it is possible that the key columns that Oracle logs are not those that SharePlex requires.
See Set up Oracle database objects for replication for more information about defining key values.
If the Capture process stops (or is stopped by a SharePlex user) while replication is active, Capture records its location in the redo log and continues from that point when started again. If the redo logs wrap before Capture starts again, Capture reads through the archive logs to locate the records it missed.
Ideally, the redo logs should be configured so that SharePlex can avoid reading the archive logs. In most cases, reading the online logs is faster than reading the archives. Ensure that the online redo logs are large and numerous enough to minimize processing from the archive logs. At minimum, there should be enough redo log capacity to hold several hours worth of data without wrapping.
Note: On Exadata systems, you might get faster Capture speed by multiplexing the logs to a different system. See Tune Capture on Exadata
To test an appropriate online log configuration
In pre-production testing, you can determine if Capture is reading the archive logs by doing the following:
Determine the log that SharePlex is processing by querying the SHAREPLEX_ACTID table.
SQL> select seqno from splex.shareplex_actid
Determine the log that Oracle is writing to by querying Oracle's V$LOG table.
SQL> select sequence# from v$log where status='CURRENT'
If there is a long delay between when Capture stops and then starts again, Capture can lose pace with Oracle activity and there will be latency between the source and target data. Capture usually needs to read the archives in this case because the required log is no longer online. To avoid Capture issues, configure archive logging to support faster and uninterrupted replication as follows:
|Enable archive logging on the source system||Archive logging must be enabled on the source system. Otherwise, if the online logs wrap before Capture is finished processing them, a resynchronization of the source and target data will be necessary.|
|Time compression properly||Do not compress the archive logs until SharePlex is finished processing them. Otherwise, SharePlex returns a "log wrap detected" message and will stop because it cannot process the data. To determine the current log for SharePlex, issue the show capture command with the detail option in sp_ctrl on the source system. You can compress any logs that were generated before the current one.|
|Specify a non-default archive location||If you are storing the archive logs in a location other than the Oracle default, set the SP_OCT_ARCH_LOC parameter to the full path name of the directory where the archive logs are located. If the redo logs wrap, SharePlex searches for the archive logs in Oracle’s archive log list. If SharePlex does not find the archive log there, it searches the directory or directories specified by the SP_OCT_ARCH_LOC parameter. To have Capture go directly to the SP_OCT_ARCH_LOC location and skip the read of the Oracle log list, set the SP_OCT_CK_LOC_FIRST to 1.|
|Configure Capture to wait for log-management processes||If using SP_OCT_ARCH_LOC and you are using an automated method to move the logs to that location, you can configure Capture to wait a certain amount of time for the move to be completed. This prevents Capture from stopping because a log it needs is not yet available. Capture waits, checks for the logs, stops if they are not yet available, and continues checking and stopping until the logs are available. To configure Capture to wait, set the SP_OCT_LOGWRAP_RESTART parameter to the number of seconds that you want Capture to wait. Monitor these processes regularly to prevent replication latency.|
|Disable archive logging on the target||You can disable archive logging on the target systems to eliminate unnecessary Oracle activity on that system, except for high availability or peer-to-peer strategies.|
|Do not place logs at the root ASM location||
If the database uses ASM, the Oracle redo logs (online and archive) cannot be located under the ASM root directory. SharePlex will not be able to read them at that location.
|Read from the archive logs on Exadata||
Normally, SharePlex provides the least latency when reading from the online redo logs. On Exadata, however, SharePlex can process higher volumes of data when reading from a multiplexed archive location outside of the Exadata ASM file system. For more information, see Tune Capture on Exadata.
Important: If Capture falls behind the pace at which Oracle generates redo volume, the following may apply:
This topic provides information about how to configure certain characteristics of Oracle database objects that you will be replicating with SharePlex.
SharePlex must have a way to ensure that the row it is changing on the target is the correct one that matches the source row. This is accomplished by ensuring a one-to-one relationship through the use of keys and indexes.
SharePlex works fastest when there is a primary key or a unique key on all source and target tables being replicated, especially large tables and tables containing LONG columns. When choosing a key to use, SharePlex uses the best available key column in the following order of priority.
Unique key with the fewest columns.
For best performance, it is recommended that you enable primary and unique key supplemental logging.
If a table does not have a primary or unique key, or if Oracle logs the wrong unique key for SharePlex, you can specify columns for SharePlex to use as a key when you create the configuration file. This is known as a key definition and is specified in the configuration file. For more information, see Define a unique key.
An alternative to a key definition is to create or use a unique index based on one or more columns that establish uniqueness.
If SharePlex cannot detect a key or unique index on a table, it constructs a key using all of the columns except LONGs and LOBs. This key is maintained internally and is not created on the table itself.
This is not a desirable option, because the resultant WHERE clause causes Oracle to perform a full table scan on the target table to find the row, and that significantly slows replication. Additionally, row uniqueness cannot be enforced.
For example, if there is a possibility that non-LONG columns in different rows could contain identical values, but the LONG columns could have different values, the table can be out of synchronization without being detected by you or SharePlex. The following example illustrates the problem. The rows in the table are identical except for the LONG columns, and there is no primary or unique key.
|COLUMN A||COLUMN B||COLUMN C (LONG)|
Suppose a user on the source system changes Column A to 15 in the first row. When constructing the SQL statement to apply the change to the target table, SharePlex constructs a key using Columns A and B (UPDATE tablename SET Column A to 15 WHERE Column A = 10 and Column B = 20) to locate the row to change. There are three rows that meet this criteria, so SharePlex could post the change to the wrong row.
If a key permits NULLs, SharePlex cannot ensure the uniqueness of a row for UPDATEs and DELETES, so there is a possibility it could change the wrong row on the target system. To control how SharePlex addresses keys that permit NULLs, set the SP_SYS_IN_SYNC parameter. See the SharePlex Reference Guide for more information.
SharePlex handles changes to the value of a key column without any special settings. However, if sequences are used for keys and if there is a possibility that those values will be updated, create the sequences so that the updates cannot result in a duplication of keys on the target system. Otherwise, if a new value is used to apply an operation, and that value already exists as a key in another row in the target table, SharePlex will return unique-key constraint violations and out-of-sync errors. This type of error can happen when you update values using an “x +n” formula, where n is an incremental increase. It is possible for one of the “x +n” values to equal an existing value.
The following is an example, where the values in the key column are increased by 1:
SQL> update table X set a=a+1; commit
The new values become the following, and they are replicated to the target system:
SharePlex performs the update in the order that the operations enter the redo log:
update x set a=2 where a=1; (Succeeds)
update x set a=5 where a=4; (Fails because a value of 5 already exists.)
update x set a=6 where a=5; (Succeeds)
update x set a=8 where a=7; (Succeeds)
The pre-image value that Post uses for the target sequence is the same as the increased value replicated from the source. Oracle rejects the operation as a unique constraint violation. Another example would be a transaction that updates A to B and then updates B to C.
Important! If you plan to use peer-to-peer replication, there are additional requirements for keys. For more information, see Configure replication to maintain multiple peer databases.
The correct use of indexes is important in a replication environment. An index maintains the integrity of the target data.
If tables have foreign keys, make sure the appropriate columns are indexed so that modifications to the foreign keys do not cause full table scans.
If there are too many indexes on a target table, Oracle must update them all as rows are added and deleted. This slows the entire system, including replication. Consider limiting the number of indexes to the ones that have the most utility.
For applications that perform mostly one type of DML, consider the following:
If you run large batch jobs that do millions of SQL operations, remove unnecessary indexes before the batch job, then rebuild them at the end. This makes SharePlex run faster, and you will have more organized indexes afterward.
For performance purposes, avoid the use of bitmap indexes while the Post process is applying the data. These indexes can adversely affect the performance of the Post process.
If you must use bitmap indexes on target tables, weigh their benefit for queries against the their impact on the transactions applied by Post:
Note: Replicating bitmap indexes is not recommended. Every time you change a table with a bitmap index, the index is rebuilt. This associated cost (Oracle time and resources) to rebuild is added to your SQL UPDATE statement.
DML changes resulting from a trigger firing on the source system enter the redo log and, therefore, are replicated and posted to the target database by SharePlex. Consequently, if the same trigger fires on the target system and initiates the same DML changes (which already were made through replication) there will be out-of-sync errors.
For example, if an INSERT to TableA on the source system triggers an INSERT into TableB, SharePlex replicates both INSERTs to the target system. The Post process applies the first INSERT to TableA on the target system, triggering an INSERT into TableB. Thus, when Post attempts to post the replicated INSERT to TableB, there will be a unique-key violation. The row already exists because the trigger fired for TableA.
Triggers can be handled as follows, depending on your replication strategy:
|Replication strategy||How to handle triggers on the target|
|Reporting, data sharing, other basic one-way replication||
For important information about how to use the trigger scripts, see the SharePlex Reference Guide.
Integrity constraints have a effect on replication. Follow these guidelines to ensure that they are handled.
Foreign key constraints must be disabled on the target tables. SharePlex replicates the results of the source foreign key constraints. Tables with foreign keys to one another must all be included in the replication configuration for accurate replication of the source foreign key results. All tables with referential constraints must exist in the target database. If you leave one or more out, the referential integrity could become corrupted.
Note: If constraints are DEFERRED on the target tables, the Post transaction can fail on the constraint validation. To work around this problem, enable the SP_OPO_DISABLE_OBJNUM parameter to allow the transaction to succeed. The underlying target table still remains out-of-sync until it is resynchronized.
SharePlex allows ON DELETE CASCADE constraints to remain enabled on the target tables, but this feature must be enabled. Post detects the ON DELETE CASCADE dependencies and suppresses the posting of any replicated cascaded deletes to the child tables.
If you do not enable this support through SharePlex you must disable these constraints manually on the target. Otherwise SharePlex replicates both the primary delete and the cascaded deletes, which results in conflicts and errors when the deletes cascade on the target.
To enable ON DELETE CASCADE support
Set the following SharePlex parameters:
Disable check constraints on the target system. They add unnecessary overhead. These checks are redundant in a well maintained, synchronized replication environment because they are satisfied on the source system. For high-availability purposes, you can build a script that re-enables the constraints as part of the failover procedure.
In all scenarios except peer-to-peer replication, the SharePlex database user must be the only user allowed to perform DML or DDL on the target objects. If DML or DDL changes are made to target objects by other individuals, jobs or applications, the target data may no longer reflect the state of the data on the source system. For more information, see Understand the concept of synchronization.
SharePlex replicates changes to Oracle sequences that are made in an ALTER SEQUENCE and DROP SEQUENCE commands and during DML transactions. It may not be necessary to replicate sequences in certain replication strategies.
High availability: Yes
The way that SharePlex replicates sequences allows users to begin using the failover database immediately without worrying about incrementing the sequences or reusing one.
Reporting, data sharing, other basic one-way replication: No
If sequences are unnecessary on the target system, do not replicate them. It can slow down replication. Even if a sequence is used to generate keys in a source table, the sequence values are part of the key columns when the replicated rows are inserted on the target system. The sequence itself does not have to be replicated.
SharePlex does not support peer-to-peer replication of identical sequences. For more information, see Configure replication to maintain multiple peer databases.
To configure sequences for replication
To ensure the uniqueness of sequences on the target system, the start value of the target sequence must be larger than the start value of the source sequence. Use the following formula to determine the target START_WITH value:
source_current_value+ (source_INCREMENT_BY_value x source_CACHE_value) =target_START_WITH_value
Important! (source_INCREMENT_BY_value x source_CACHE_value) should not exceed 2 GB, or replication of sequences will fail.
SharePlex uses the ALTER SEQUENCE command to update sequences in a target database as follows:
Change the increment value to:
ALTER the sequence(s) again by setting the following values:
SharePlex treats ALTER SEQUENCE operations like a simple SELECT (UPDATE) to a sequence because the redo log records make no distinctions between the two operations.
Certain Oracle database settings have an affect on replication and must be set appropriately.
SharePlex requires the value of the Oracle OPEN_CURSORS parameter to be set correctly on the target database. To view the OPEN_CURSORS value, query the database using the following SQL statement:
select value from V$PARAMETER where name = 'open_cursors';
The Post process reserves 10 cursors for routine calls that are closed once they finish, plus a minimum of 50 cursors per transaction when the SQL Cache feature is enabled (this is the default). For more information, see Tune SQL Caching.
If you intend to disable SQL caching, then estimate the peak number of concurrent updating transactions (sessions) that your applications generate, and follow this formula:
10 + (peak number of concurrent transactions x 2) = minimum open cursors needed
The OPEN_CURSORS value can be modified or added if absent. Consult the Oracle documentation prior to making changes to Oracle parameters.
The PROCESSES parameter in the init.ora file must be set to accommodate connections created by SharePlex and the database users. Its value depends on whether a database is a source database, a target database, or one that serves as both a source and target database.
If a database serves only as a source, the following formula takes into account logins made by the Read process.
(peak number of source database sessions) + (background Oracle processes) + (value of SP_ORD_LDA_ARRAY_SIZE parameter +3) = setting for PROCESSES
The Post process creates as many connections on the target system as there are sessions on the source system in order to maintain transaction consistency. The PROCESSES parameter on the target system must be set high enough to accommodate all of those connections, plus:
Use the following formula as a guide:
(peak number of source database sessions) + (peak number of target database sessions) + (background Oracle processes) = setting for PROCESSES
If a database serves as both a source and a target, the following formula takes into account connections made by:
(peak number of source database sessions) + (peak number of target database sessions) + (background Oracle processes) + (value of SP_ORD_LDA_ARRAY_SIZE parameter +3) = setting for PROCESSES
The number of database writers has an impact on replication, especially when there are numerous concurrent transactions. Whenever a transaction is committed, its buffered data is flushed to disk. If most transactions are small, but the buffer is large, this can cause slow posting. When a large transaction is committed while another, more normal-sized transaction is committed, the second COMMIT must wait while the entire buffer is flushed to disk.
Reducing the size of the buffer that is flushed to disk can speed the Post process. Try decreasing the size of the log buffer to 1024 KB, or even 512 KB if possible.
SharePlex updates the SHAREPLEX_TRANS table to maintain read consistency for the target database. You might need to adjust the initrans setting of this table to improve replication performance and reduce contention on that table:
This topic explains how SharePlex handles character set conversion between an Oracle source and target, and between an Oracle source and a non-Oracle target.
For SharePlex to replicate all characters within the Oracle character sets that you are using, one of the following must be true:
The following character sets are tested and supported for SharePlex:
By default, SharePlex allows an Oracle target database to perform character conversion. Post notifies Oracle of the character encoding of the source data and Oracle performs any required conversion.
Depending on the character sets involved, the Oracle conversion might lead to data loss. For example:
Example 1: The Japanese character for 'rice' in the JA16SJIS character set has no corresponding symbol in the US7ASCII character set. If you attempt to replicate this symbol into a US7ASCII database, Oracle converts it to a '?' character.
Example 2: According to Oracle, the WE8ISO8859P1 character set is a superset of the US7ASCII character set, so it is logical to assume that any character in US7ASCII is posted unconverted into a WE8ISO8859P1 target database. This is true for characters in the range 0x00 to 0x7F. However, Oracle strips off the top bit of characters in the range 0x80 to 0xFF. This "conversion" may result in data loss while replicating to a character set that is a superset of the source.
Note: Oracle does not convert characters if the character sets are identical. Thus, posting WE8ISO8859P1 data to a database with a character set of WE8ISO8859P1 bypasses the Oracle conversion process.
To apply data without conversion
Set the SP_OPO_NLS_CONVERSION parameter to 1 to apply the data without conversion. For more information, see the SharePlex SharePlex Reference Guide.
Note: SharePlex will always convert NVARCHAR and NCLOB data if the NLS_NCHAR_CHARACTERSET of the source database is not the same as that of the target database.
When replicating to an Open Target target (non-Oracle target), SharePlex supports replication from any Oracle Unicode character set and the US7ASCII character set. SharePlex posts data to Open Target in the Unicode character set, and therefore if the source data is Unicode or US7ASCII, no conversion on the target is required.
However, if the following are true, conversion is required on the target:
To perform conversion with an Oracle client on Linux
To perform conversion with an Oracle client on Windows
To apply Unicode and US7ASCII data without conversion
If the source data is Unicode or US7ASCII and you are not replicating LOB data, no conversion or Oracle client is needed. Set the SP_OPX_NLS_CONVERSION parameter to 0 to disable conversion, and then restart Post if it is running.