Cancellation in Systems: An Empirical Study of Task Cancellation Patterns and Failures

Modern software applications rely on the execution and coordination of many different kinds of tasks. Often overlooked is the need to sometimes prematurely terminate or cancel a task, either to accommodate a conflicting task, to manage system resources, or in response to system or user events that make the task irrelevant. In this paper, we studied 62 cancel-feature requests and 156 cancel-related bugs across 13 popular distributed and concurrent systems written in Java, C#, and Go to understand why task cancel is needed, what are the challenges in implementing task cancel, and how severe are cancel-related failures. Guided by the study, we generalized a few cancel-related anti-patterns, and implemented static checkers that found many code snippets matching these anti-patterns in the latest versions of these popular systems. We hope this study will help guide better and more systematic approaches to task cancellation.