| 摘要 |
Data declaration language (DDL) replication can be performed without stopping the system. In other words, DDL changes can be replicated seamlessly. Within a RDBMS (relational database management system), the execution of a DDL statement involves many internal steps which are not visible to an end user. These steps can be logically divided into three stages, "pre-stage" which refers to the steps before the server makes a physical change to the system catalog for the specific DDL, "middle-stage" which refers to the action of making a change to the system catalog, and "post-stage" which refers to the steps after changes to the system catalog were made. Schema changes are intercepted in both pre-stage and post-stage. Schema changes are intercepted in pre-stage so an old image of schema information can be saved. This helps to support schema versioning logic. Schema changes are intercepted in post-stage so sets of commands can be posted to propagate the schema change, as well as to refresh any dependencies. This is desirably performed before the DDL command commits so all updates are protected by the same transaction. Moreover, the schema change and related updates are propagated in the same connection as DML changes, honoring the original sequence of action from the publishing side.
|
