Replicating Only Changed Columns

By default, Enterprise Replication replicates the entire row, even if only one column changed. (For more information on how Enterprise Replication evaluates data for replication, see Evaluating Data for Replication.)

You can change the default behavior to replicate only the columns that changed. To replicate only changed columns, include the --fullrow n option in the replicate definition.

Replicating only the columns that changed has the following advantages:

• Sends less data

  This is particularly useful when replicating large objects, such as data stored as smart large objects and opaque user-defined types (UDTs). For example, when you replicate a row with a large column that changes infrequently and a small column that changes frequently, Enterprise Replication sends significantly less data each time it updates the row if it only replicates the changed columns.

• Uses fewer Enterprise Replication resources, such as memory

If Enterprise Replication replicates an entire row from the source, and the corresponding row does not exist on the target, Enterprise Replication applies the update as an insert, also known as an upsert, on the target. By replicating the entire row, Enterprise Replication corrects any errors during replication. If any errors occur in an update of the target database server (for example, a large object is deleted before Enterprise Replication can send the data), the next update from the source database server (a complete row image) corrects the data on the target server.

Replicating only the columns that changed has the following disadvantages:

• Enterprise Replication does not apply upserts.

  If the row to replicate does not exist on the target, Enterprise Replication does not apply it. If you set up error logging (Setting Up Error Logging), Enterprise Replication logs this information as a failed operation.

• You cannot use the SPL routine or time stamp with SPL routine conflict-resolution rules. For more information, see Conflict Resolution.
• You cannot use update-anywhere replication; doing so can result in inconsistent conflict resolution.
• All database servers in the enterprise must be Version 9.3 or later.

  Enterprise Replication does not enforce this restriction. If you attempt to replicate only changed columns to a pre-Version 9.3 database server, you will corrupt the data on that database server.

Enterprise Replication logs bitmap information about the updated columns in the logical-log file. For more information, see the CDR record type in the logical-logs chapter in the IBM Informix Administrator's Reference.

For more information on the --fullrow option, see Special Options.