Thoughts on Armstrong thesis ch. 5 - Programming Fault-Tolerant Systems

This chapter of Armstrong's thesis presents a specific method for programming fault-tolerant systems. In my experience, it is often easier to envision how a system should work then how it may fail. Unfortunately, the later is critical for error handling and that in turn is a precursor for having a fault-tolerant system.

Armstrong's idea of having a hierarchy of tasks and supervisors is interesting. I have worked on systems that have supervisors monitoring worker processes, but the arranging of supervisors in a tree here is different and it is great as it also sets up monitoring hierarchy among the supervisors. I suspect this hierarchy eventually translates to the different task levels that the author has mentioned multiple time, although I'll still need to read more into the thesis to understand how it is actually set up.

At the end of the chapter, Armstrong describes the four rules of a Well-Behaved Function (WBF). Rule 2 is about "if the specification doesn't say what to do raise an exception" which is particularly interesting to me. In my experience, when a specification doesn't say something, it usually does not mean that something is not supposed to happen but rather there is not a required behavior. Thus, most of the time it is handled by a system-specific behavior. The author obviously has a different take on it and it seems he would like to eliminate such ambiguity. I am not sure whether this is a realistic expectation for the specification.