SharePlex 9.2.6 - Administration Guide

Configure horizontally partitioned replication

Use horizontally partitioned replication to divide the rows of a table into separate processing streams. You can use horizontally partitioned replication to:

• Replicate a subset of rows to a target, while retaining the rest of the rows in the source.
• Replicate different subsets of rows to different targets.
• Divide the replication of a source table into parallel Post queues for faster posting to the target table.

Supported sources

Oracle

Supported targets

All

Overview of horizontally partitioned replication

To configure horizontally partitioned replication for a table, the steps are:

1. Define row partitions and link them to a partition scheme.

   • A row partition is a subset of rows in a source table that you want to replicate as a group.

   • A partition scheme is a logical container for row partitions.

2. Specify the name of the partition scheme in the SharePlex configuration file to include the partitions in replication.

Partition types

The row partitions in a partition scheme can be based on one of the following:

• Columns: A column-based partition scheme contains row partitions defined by a column condition. A column condition is a WHERE clause that defines a subset of the rows in the table.
• Hash: A hash-based partition scheme contains row partitions defined by a hash value that directs SharePlex to distribute rows evenly across multiple queues.

About column-based partition schemes

You can use row partitions based on column conditions for the following purposes:

• Use a single row partition to replicate only a subset of the rows of a table. For example, you can replicate only those rows where the value of the YEAR column is greater than 2014. The partition scheme in this case could be named "Since2014" or "Recent."

• Use multiple row partitions to divide the rows of a table so that each set of rows replicates to a different target. For example, a table named CORPORATE.SALES could have two row partitions named "East" and "West." The column conditions are defined accordingly, where the rows that satisfy REGION = EAST replicate to one location and the rows that satisfy REGION = WEST replicate to a different location. The partition scheme could be named "Sales_by_region."

  

• Use multiple row partitions to divide the rows of a table into parallel processing streams (parallel Export-Import-Post streams) for faster posting to a target table. For example, you can improve the flow of replication to a heavily updated target table. The use of column conditions for this purpose is appropriate only if the table contains a column that enables you to split the processing evenly among parallel Post processes.

  

About hash-based partition schemes

You can use row partitions based on a hash value to divide the rows of a table into parallel processing streams (parallel Export-Import-Post) for faster posting to a target table. The advantage of using a hash value over column conditions to create partitions is that the rows are divided evenly and automatically by SharePlex, without the need to reference table columns in WHERE clauses. However, unlike column-based partition schemes, you cannot use the SharePlex compare or repair commands for hash-based partition schemes.

Combine partitioned replication with full-table replication

You can combine horizontally partitioned and vertically partitioned replication for maximum control over how information is distributed.

For example:

• A corporate headquarters maintains a master corporate database.
• Each of the corporation's four regional offices maintains its own database.
• Corporate headquarters uses vertically partitioned replication to share some column data of a table with the regional offices, but does not share any columns that contain sensitive data.
• The rows of the table are horizontally partitioned into four groups (East, West, North, South) for replication, so that each region receives only the record changes that apply to it.

Horizontally partitioned replication can be used in conjunction with full-table replication for the same table, for example to route groups of rows to different reporting systems and all rows to a backup system.

Limitations of use

Hash-based partitioning does not support the following:

• The compare and repair commands.
• Index-organized tables (IOT) and tables that contain LOBs or LONGs.
• Operations that delete or update a key value and then reinsert the same key value. This can cause unique constraint violations because of different rowids.
• Column-based partitioning on the same table.

Hash-based partitioning also does not support operations that cause rows to migrate into a different partition. Examples of such operations are:

• Update a value so that it moves to a new row partition
• Table reorganization
• Split a table partition or combine two partitions
• Export or import the table
• ALTER TABLE with the MOVE option
• ALTER TABLE with the SHRINK SPACE option
• FLASHBACK TABLE
• Redefine a table by using dbms_redefinition
• UPDATE to a non-partitioned table that changes row size so that the data does not fit into the current block
• DELETE of a row in a non-partitioned table and then re-insert.

Note: If Post returns the error message "shs_SEMERR: an error occurred with the semaphore" on a Windows system, the number of semaphores may need to be increased to accommodate the queues that you created. For more information, see Post stopped .

Define partition schemes and row partitions

Use the add partition command to create row partitions and assign them to a partition scheme.

To partition rows based on column conditions

Issue add partition for each row partition that you want to create in a given partition scheme. When you create the first row partition, SharePlex creates the partition scheme as well.

sp_ctrl> add partition to scheme_name set condition = column_condition and route = routing_map [and name = name] [and tablename = owner.table] [and description = description]

To partition rows based on a hash value

Issue add partition once to specify the number of hash partitions to create.

sp_ctrl> add partition to scheme_name set hash = value and route = value

add partition command syntax

Note: After you specify add partition with to scheme_name and the set keyword, all other components can be in any order.

Examples

Partitions based on a column condition

Route different sets of rows through different post queues:

sp_ctrl> add partition to scheme1 set name = q1 and condition = "C1 >= 200" and route = sysb:q1@o.orasid

sp_ctrl> add partition to scheme1 set name = q2 and condition = "C1 < 200" and route = sysb:q2@o.orasid

Route different sets of rows to different target systems and different table names from the source:

sp_ctrl> add partition to scheme1 set name = east and condition = "area = east" and route = sys1e@o.orasid and tablename = ora1.targ

sp_ctrl> add partition to scheme1 set name = west and condition = "area = west" and route = sys2w@o.orasid and tablename = ora2.targ

Partitions based on a hash

Divide rows into four partitions, each processing through a different post queue:

sp_ctrl> add partition to scheme1 set hash = 4 and route = sysb:hash|#@o.ora112

How to create a valid column condition

The following are guidelines for creating column conditions. These guidelines do not apply to row partitions that are created with a hash value.

Choose appropriate columns

The types of columns on which you base your column conditions vary per datasource:

Base column conditions on columns whose values will not change, such as PRIMARY or UNIQUE key columns. The objective is to avoid a partition shift, where changes made to the conditional columns of a partition can cause the underlying data to satisfy the conditions of a different (or no) partition.

Partition shift case 1: The column value is updated so that the new value no longer satisfies any column condition:

• SharePlex performs the operation, but future operations on that row are not replicated. The reason: the row no longer satisfies a column condition.
• The source and target tables of the original partition are now out of synchronization, but Post returns no errors.

Partition shift case 2: A row that satisfies one column condition gets updated to meet a different condition:

• Post cannot find a matching target row. The reason: the original change was not replicated because it did not meet the column condition.
• Post returns an out-of-sync error.

You can use the following method to repair the out-of-sync rows that are caused by changes to the values of column conditions:

• Use the compare command to repair the out-of-sync rows. For more information about this command, see the SharePlex Reference Guide.

Additionally, you can ensure that data is replicated properly by setting the following parameter on the source prior to activating the configuration file.

• Set the SP_ORD_HP_IN_SYNC parameter to a value of 1. When this parameter is enabled, if an UPDATE changes a column (conditional column) value resulting in a row that no longer satisfies the correct condition, SharePlex corrects the row. Enabling this parameter causes some performance degradation, depending on how many tables are configured for horizontally partitioned replication. See the SharePlex Reference Guide for more information and a list of conditions corrected by this parameter.

Note: If you are using a column other than a key to base the column condition on, and you notice reduced performance with horizontally partitioned replication enabled, add a log group for that column.

Use supported data types

SharePlex supports the following data types in column conditions:

• NUMBER
• DATE
• CHAR
• VARCHAR
• VARCHAR2
• LONG VARCHAR

Notes:

• For the dates, SharePlex uses MMDDSYYYYHH24MISS. For example:

  hiredate<‘1111 2011000000’

• Horizontally partitioned replication does not support the following:

  • data types other than the ones listed in this section. This also excludes large types like LOBs and object types such as VARRAYs and abstract data types.
  • Oracle TO_DATE function
  • UPDATEs or INSERTs on LONG columns larger than 100k
  • Sequences
  • TRUNCATEs of an Oracle partition

Use standard conditional syntax

The following list shows the conditional syntax that SharePlex supports in a column condition, where:

• value can be a string or a number. Enclose strings and dates within single quote marks (‘west’). Do not use quote marks for numbers .
• column is the name of a column in the table that you are configuring to use horizontally partitioned replication.

Conditions can be combined into nested expressions with parentheses and the AND, OR, and NOT logical connectives.

Example column conditions

Additional guidelines

• NULLs are replicated by SharePlex in cases such as this one: not (department_id = 90). If department_id is NULL, it is replicated. To avoid replicating records with NULLs, include the column is not null syntax as part of the condition, for example: not (department_id = 90) and department_id is not null.
• If parentheses are not used to indicate operator precedence, SharePlex supports operator precedence in the same order that SQL does. For example, a condition not x and y behaves the same way as (not x) and y. A condition x and y or z behaves the same as (x and y) or z. When a condition includes parentheses, the explicit precedence is respected.
• If the condition column is a VARCHAR column and the values used to define the partitions are string literals, the entire condition must be enclosed in double quotes, as in the following example: add partition to scheme set route=route and condition="C2 = 'Fred'"
• If the column name must be enclosed in quotes, then the entire condition must be enclosed in quotes, as in the following example: add partition to scheme set route=route and condition="\"c2\" > 0"

• Do not:

  • include references to other tables in the column condition.
  • exceed the 1024 bytes maximum defined storage.
• During the activation of a configuration that refers to partition schemes, SharePlex verifies the syntax in the column conditions of those schemes. If any syntax is incorrect, the activation fails. SharePlex prints an error to the Event Log that indicates where the error occurred.

Specify partition schemes in the configuration file

Use one configuration file for all of the data that you want to replicate from a given datasource, including tables that will have full-table replication and those that will use partitioned replication. For more information about how to create a configuration file, see Configure data replication. To configure entries for horizontally partitioned replication, use the following syntax.

Examples

To specify a partition scheme:

To specify multiple partition schemes for the same source table:

To specify a placeholder routing map:

This placeholder is only required for partitions based on column conditions.

View the partitions and schemes

Use the view partitions command to view the row partitions in one partition scheme or all partition schemes in a horizontally partitioned replication configuration.

To view row partitions

1. Run sp_ctrl on the source system.

2. Issue the following command with either option, depending on whether you want to view all partitions or just those for a particular partition scheme.

   sp_ctrl> view partitions for {scheme_name | all}

The following example shows both a hash-based partition scheme and a column-based partition scheme.

sp_ctrl> view partitions all

Scheme        Name           Route                            Hash     Condition
-----------   -------------  ------------------------------   ------   ---------------
HASH4         hash           sys02:hash|#@o.ora112            4        ROWID
TEST_CT       highvalues     sys02:highvalues@o.ora112                 sales>=10000
TEST_CT       lowvalues      sys02:lowvalues@o.ora112                  sales<10000

Hash4 hash-based partition scheme

• The Scheme column shows a partition scheme named HASH4.
• The Name column shows that the name for the partition definition is hash.
• The Route column shows that the partitions are created automatically and that the target is o.ora112.
• The Hash column has a value of 4, which indicates that this is a hash-based partition scheme with four partitions.
• The Condition column shows that the type of hash algorithm that is being used is the default of rowid, rather than block.

TEST_CT column-based partition scheme

• The Scheme column shows a partition scheme named TEST_CT. There are two entries for this name, indicating that it contains two partitions.
• The Name column shows the name of each partition, which by default is the name of the post queue or the value set with the Name option of the add partition command.
• The Route column shows that the names of the post queues are based on the parition name and that the target is o.ora112.
• The Hash column is empty for a column-based partition scheme.
• The Condition column shows the column condition that creates the row partition.

To view partition post queues

The qstatus command on the target shows the post queues that are associated with horizontally partitioned replication.

Queues for TEST_CT column-based partition scheme

sp_ctrl sys02> qstatus
 
Queues Statistics for sys02
  
  Name:  highvalues (o.ora11-o.ora112) (Post queue)
    Number of messages:          0 (Age         0 min; Size          1 mb)
    Backlog (messages):          0 (Age         0 min)
 
  Name:  lowvalues (o.ora11-o.ora112) (Post queue)
    Number of messages:          0 (Age         0 min; Size          1 mb)
    Backlog (messages):          0 (Age         0 min)

Queues for HASH4 hash-based partition scheme:

Queues Statistics for sys02
  
  Name:  hash1 (o.ora11-o.ora112) (Post queue)
    Number of messages:          0 (Age         0 min; Size          1 mb)
    Backlog (messages):          0 (Age         0 min)
 
  Name:  hash2 (o.ora11-o.ora112) (Post queue)
    Number of messages:          0 (Age         0 min; Size          1 mb)
    Backlog (messages):          0 (Age         0 min)

 Name:  hash3 (o.ora11-o.ora112) (Post queue)
    Number of messages:          0 (Age         0 min; Size          1 mb)
    Backlog (messages):          0 (Age         0 min)

 Name:  hash4 (o.ora11-o.ora112) (Post queue)
    Number of messages:          0 (Age         0 min; Size          1 mb)
    Backlog (messages):          0 (Age         0 min)

Make changes to partition schemes

The following commands or parameters are available to manage partition schemes. For more information, see the SharePlex Reference Guide.

Configure vertically partitioned replication

Use vertically partitioned replication to replicate a subset of the columns of a table. For example, you can replicate data changes made to C1, C2, C3, and C4, but not changes made to C5 and C6, as shown in the diagram.

Supported sources

Oracle

Supported targets

All

Guidelines for using vertically partitioned replication

Follow these guidelines when creating a configuration file that includes vertically partitioned replication.

• Vertically partitioned replication is appropriate for reporting and other data sharing strategies, but it is not appropriate for high availability environments. Once you configure a table for vertically partitioned replication, SharePlex does not recognize the other columns, so data in those columns is not replicated.

• You can combine horizontally partitioned and vertically partitioned replication for maximum control over which information is distributed, and to where.

  For example:

  A company has a headquarters and regional divisions. The headquarters maintains the corporate database, and each region maintains a regional database. The headquarters uses vertically partitioned replication to share some of the column data of a table to those locations, while retaining other sensitive data at headquarters. Row changes made to the shared columns are further partitioned horizontally, for replication to the appropriate regional database.
• A table cannot be configured to replicate some columns to one target system and all columns to another (combination of vertically partitioned replication and full-table replication). You can, however, configure full-table replication to an identical table on one target, and then configure vertically partitioned replication from that target to a second target that contains the table that requires only the partition columns.
• A target table can, but does not have to, contain all of the same columns as its source table. The target can contain just the columns being replicated from the source table. The names of corresponding source and target columns do not need to be the same. Corresponding columns must contain the same data types (same type, size, precision).

To configure vertically partitioned replication, you specify either a column partition or an exclusion column partition in the configuration file:

• A column partition specifies the columns that you want to include in replication. Only data changes that are made to the specified columns get sent to the target.
• An exclusion column partition specifies columns to be excluded from replication. No data from those columns is replicated to the target.

Follow these rules to specify either type of column partition:

• There can be one partition per source table. A column partition and an exclusion partition are mutually exclusive.
• A column list must be enclosed within parentheses.
• Separate each column name with a comma. A space after the comma is optional.
• The maximum length of a partition is 174820 bytes (the maximum line length allowed in a configuration file). Therefore, the actual number of columns that you can list depends on the length of each name.
• The columns can be contiguous or non-contiguous in the source table. For example, you can replicate the first, third and seventh columns of a table.
• Key columns are not required to be included in the partition.
• If using horizontally partitioned and vertically partitioned replication together for this table, all of the columns in the partition scheme must be part of the column condition.
• Use one configuration file for all of the data that you want to replicate from a given datasource, including tables that will have full-table replication and those that will use partitioned replication.

To configure entries for vertically partitioned replication, use the following syntax. For more information about how to create a configuration file, see Configure data replication.

Configuration examples

The following is a vertically partitioned replication configuration replicating to multiple targets by using a compound routing map. To use a compound routing map for this source table, all targets must be named scott.sal.

The following is a vertically partitioned replication configuration replicating to a single target where the target columns have different names from those of the source.

The following configuration file is not valid because it repeats the same column partition of scott.emp (c1, c2) twice in the configuration file.

Configure replication to a change history target

This chapter contains instructions for how to configure SharePlex to maintain a change-history target. SharePlex enables you to maintain this history, while also replicating the same data set to maintain up-to-date targets.

Contents

Overview of the change-history target

A change history target differs from a replication target in that a change history target maintains a record of every change that occurs to a source object or objects, rather than simply maintaining a mirror of the current state of the source data. While regular replication overlays current target data with change data, change history inserts the change data to the target as a new record. The old data is preserved as a step-by-step record of change. The historical data can be queried and analyzed for such purposes as data mining or resolving customer disputes.

By using SharePlex to maintain change history on a secondary server, you can offload the overhead from the production database. Such overhead includes the SQL work of adding the history rows, the extra storage of those rows, and the query activity against the historical data.

Note: File, JMS, and Kafka targets support change history by default, because every source change is written as a separate XML record. There is no overlaying of old data with new. Metadata that is supported for these targets is included automatically when Post writes the XML. For a list of supported metadata, see the target command in the SharePlex Reference Guide.

Capabilities

This replication strategy supports the following:

• Identical or different source and target names
• Use of vertically partitioned replication
• Use of horizontally partitioned replication
• Use of named export and post queues
• Combination of regular replication and change-history replication of the same source object(s)

Supported sources

Oracle

Supported targets

Oracle target

Operations supported

SharePlex supports adding a change history row for these operations:

• INSERT
• UPDATE
• DELETE
• TRUNCATE
• ALTER TABLE to DROP COLUMN (Note: for ADD COLUMN, Post adds a column to the table, but does not create a change history row.)
• ALTER TABLE to MODIFY the data type of a column

Operations not supported

• Changes made to UDT or VARRAY columns
• DBMS_LOB operation that is used to change a part of a LOB column (The value stored for that column on the target will not be the complete LOB column.)

How SharePlex maintains change history

In a change history configuration, each target table serve as a history table that records every change made to the source data as a continuous series of rows. Each new change row that SharePlex inserts includes the following:

• the values of the key columns
• the after image of the changed columns. For inserts and updates, the after image consists of the new values of the columns that were changed (or added in the case of an insert). For deletes, the after image consists of the key values plus the other columns set to null.
• (optionally) a set of metadata values that provide context for the change. For example, there is metadata that captures the userid of the user who made the change and the source system where the change was made (useful when change data is tracked from multiple databases).

SharePlex can be configured to include the before image of update operations in the history or to control which operation types are included in the history. For example, you could include only updates and deletes.