Phi Accrual Failure Detection is an adaptive algorithm for detecting failures in distributed systems. Unlike conventional heartbeat-based methods that output a binary up/down result, it produces a continuous suspicion value (φ) that reflects current network conditions. The φ value is computed using -log10(probability of receiving a heartbeat), derived from a sliding window of past heartbeat arrival times that follow a normal distribution. Higher φ values indicate greater confidence that a system has crashed, and multiple thresholds can be defined to trigger progressively more drastic responses, decoupling monitoring logic from application requirements.
Table of contents
Conventional Failure DetectionPhi Accrual Failure DetectionBenefits of using Accrual Failure DetectorsReferences9 Impressions