Highly Available, Fault-Tolerant Pull Queries in ksqlDB

This post introduces significant enhancements to ksqlDB's pull query availability and consistency. It details a new failure detection mechanism using heartbeating between ksqlDB servers, reducing detection time to seconds. It also introduces standby routing, allowing pull queries to be served by standby replicas during active server failures, and lag-aware routing, which selects the most up-to-date standby replica based on changelog topic offsets. These changes allow ksqlDB to offer tunable consistency/availability trade-offs, moving from a CP system to an AP system during partitions loss.