principle-distributed-systems

Use CAP and PACELC frameworks to explicitly choose consistency versus availability during network partitions, and name the latency versus consistency trade-off during normal operation to match your application's correctness requirements.

Exactly-once delivery is a myth, while exactly-once effects are achievable by implementing idempotent operations and atomicity under retries and partitions, distinguishing between at-most-once, at-least-once, and exactly-once effect semantics.

Prevent split-brain in distributed systems by implementing quorum-based writes, leader election, and fencing mechanisms, while framing protocol limits like FLP and two-generals to maintain correctness during partial network failures.

You need logical clocks instead of wall-clock timestamps when ordering cross-node events in distributed systems, replacing unsafe wall-clock ordering with Lamport clocks, vector clocks, or Hybrid Logical Clocks to guarantee correct causal ordering.

Select a consistency model for your distributed database by comparing linearizable, causal, read-your-writes, monotonic reads, and eventual consistency to match your actual correctness needs and latency constraints across your system.