Failure Handling in an Optimized Two-Safe Approach to Maintaining Remote Backup Systems

In a primary-backup system, transaction processing takes place at the primary and the log records generated are propagated to the backup which uses them to reconstruct a recent state of the database. If the primary fails, the backup takes over transaction processing without causing users to observe a breach in service. Existing approaches can be broadly classiied as 2-safe or 1-safe based on whether or not log propagation from primary to backup is synchronized with the transaction commitment at the primary. While the 2-safe (synchronous) approaches guarantee ACID properties of transactions even in case of primary failure and backup takeover, they incur a higher overhead and reduced primary system performance. Recently, in 9], we introduced an optimized 2-safe approach which attempts to achieve the desirable properties of both 1-safe and 2-safe approaches by careful addition of partial synchrony in the log transfer protocol. Many approaches to maintaining primary-backup systems have been developed for a variety of system architectures including environments where both primary and backup consist of multiple processors connected via multiple communication lines. A limitation of existing algorithms for such environments is that they ignore the issue of partial primary and backup failures in which one or more primary/backup processors fail but the system as a whole survives. This paper describes a design of a primary-backup system which uses the backup to provide continued availability not only during disasters but also during partial failures as long as both the primary and its corresponding backup do not fail at the same time. Furthermore, the developed approach does not compromise disaster protection of the database in the presence of partial system failures. The approach is developed in the context of the optimized 2-safe approach to log propagation.