Derby: Replication and Availability

Abstract

This paper describes the work done to add hot standby replication functionality to the Apache Derby Database Management System. The Apache Derby project is a relational database implemented entirely in Java. Its key advantages are that it has a small footprint and it is based on the standard Java JDBC and SQL standards. It is also easy to install, deploy and use as well as it can be embedded in almost any light-weight Java application. By implementing a hot standby scheme in Apache Derby several features are added. The contents of the database is replicated at run time to another site providing online runtime backup. As the hot standby takes over on faults availability is added in that a client can connect to the hot standby after a crash. Thus the crash is masked from the clients. In addition to this, online upgrades of software and hardware can be done by taking down one database at the time. Then when the upgrade is completed the upgraded server is synchronized and back online with no downtime. A fully functional prototype of the Apache Derby hot standby scheme has been created in this project using logical logs, fail-fast takeovers and logical catchups after an internal up-to-crash recovery and reconnection. This project builds on the ideas that are presented in Derby: Write to Neighbor Mode.