Software Diagnostics and Observability Institute

A process may terminate while its service continues. A service may stop doing useful work while all its processes remain present. A computation may be recoverable after a crash, and its history may become unrecoverable after an apparently successful restart. We consider these situations as different forms of computational death. Starting from the distinction between crashes and hangs, we extend the analysis to progress, deadlines, state, identity, dependencies, and diagnostic evidence. We propose a Computational Death Profile and a candidate Death Boundary analysis pattern. Both connect termination and recovery decisions to memory dump analysis patterns and trace and log analysis patterns. We also revisit the four causes of crash dumps to distinguish the original failure, the terminating action, and the purpose of collecting diagnostic evidence. The contribution is a diagnostic synthesis with explicit conditions for its use and subsequent validation.

PDF: https://www.dumpanalysis.org/files/Forms_of_Computational_Death-Version4...