Chapter 7 - System Recovery

Abstract

Transaction processing (TP) systems often are expected to be available 24 hours per day, 7 days per week, to support around-the-clock business operations. Two factors affect their availability: the mean time between failures (MTBF) and the mean time to repair (MTTR). Improving availability requires increasing MTBF, decreasing MTTR, or both. Computer failures occur because of environmental factors, system management, hardware, and software. If the operating system fails, then just reboot it. For other types of software failure, the transactional middleware or database system must detect the failure of a process and re-create it. The re-created process must then run a recovery procedure to reconstruct its state. Transactions simplify recovery by allowing a server to focus on restoring its state to contain only the results of committed transactions, rather than recovering to a state that is consistent with the last operations it ran. All of today's recovery mechanisms require every transaction to hold its write locks until it commits, to avoid cascading aborts, and to ensure that undo can be implemented simply by restoring an update's before image. The recovery manager uses a cache manager to fetch pages from disk and later flush them. In addition to processing commit and abort operations, it implements a recovery algorithm to recover from system failures. The recovery manager tells the cache manager about dependencies between dirty database pages and log pages so the cache manager can enforce the write-ahead log protocol.