Patterns for Fault Tolerant Software

Chapter 6. Error Recovery Patterns

Error detection, discussed in the previous chapter, does not 'fix' the error. The error is still present in the system and can still cause a failure. The patterns in this chapter address error recovery, one of the two means of processing the error and resuming error-free operation. Error recovery continues execution even with a detected error by placing the system in a state that does not contain the error. This means resuming execution at a known place, allowing it to continue processing at least as well as it did before the error was detected. The other means of error processing is error mitigation to mask the error. The next chapter discusses error mitigation.

Error recovery consists of two main parts. The first part involves undoing the bad effects of the error. The second part involves creating an error free state in the system that can resume execution. Both of these must use a minimum amount of time in order to maximize availability. Redundancy (3) is used when possible to produce the fastest possible recovery.

Many of these patterns involve preserving system state through checkpoints. A Checkpoint (37) is an incrementally saved state that facilitates rapidly restoring processing to a point at which the state was saved. Restoring from the saved state decreases the time required to return to the same state that existed at the time of the error. Instead of having to replay the entire sequence of events from the beginning, processing can resume ...