Chapter 8: Distributed Systems Fundamentals for Databases

The material distinguishes between concurrency, involving interleaved task sequences that may not execute simultaneously, and parallelism, where operations genuinely occur at the same time across multiple processing units. A significant portion addresses the prevalent misconceptions in distributed computing, particularly the erroneous assumptions that networks maintain perfect reliability, operate with zero latency, or provide unlimited bandwidth capacity. These fallacies create vulnerabilities that lead to system breakdowns, requiring developers to understand processing delays, implement effective message queuing strategies, and establish backpressure mechanisms to prevent resource exhaustion. The chapter emphasizes temporal coordination challenges, explaining how local clock drift makes timestamp-based event ordering unreliable without specialized synchronization hardware. Communication reliability is addressed through various link abstractions, from fair-loss links that may drop messages to perfect links that guarantee reliable delivery using sequence numbers and acknowledgment protocols. Theoretical limitations on distributed agreement are explored through foundational problems like the Two Generals' Problem, which demonstrates the impossibility of achieving perfect coordination over unreliable channels, and the FLP Impossibility theorem, proving that consensus cannot be guaranteed in asynchronous systems where even a single process might fail. The discussion categorizes different failure scenarios, from straightforward crash-stop failures and message omissions to complex Byzantine faults where nodes exhibit arbitrary or malicious behavior. Practical resilience strategies are presented, including circuit breaker patterns, exponential backoff retry mechanisms, and randomized jitter techniques, all essential for building robust systems that can withstand the inevitable partial failures inherent in distributed database environments.