As most developers have experienced, code frequently calls into unreliable agents: databases, remote services and highly inefficient algorithms. The calls can timeout, throw exceptions or return invalid results. The industry has invested much time and money in devising software patterns that attempt to minimise the impact of these routine errors. We propose to record the execution history of such calls for the purpose of assessing the likelihood of a task completing successfully. Our system can decide to cancel a call that is likely to fail in order to save the resource of both the caller and the callee, generally improving system resilience. Additionally, the system validates the results returned from successful calls and provides a recourse mechanism for cancelled calls. We achieve this by introducing instrumentation that is simple to add to C++ code that provides cancelling logic, validation, and result normalization.
Advertisement
