Which approach best supports data consistency in a distributed system when strict distributed transactions are not feasible?

• A. Use idempotent operations, appropriate isolation models, eventual consistency with conflict resolution, and patterns like sagas.
• B. Rely solely on dry-run mode and client-side retries.
• C. Disable caching to avoid stale reads.
• D. Use immediate global locking for every operation.

Answer: (C)

When you can’t use strict distributed transactions, keeping data consistent across services means designing for eventual consistency with safe retry behavior and clear reconciliation paths. Idempotent operations are crucial here because they let the system handle retries without causing duplicate effects or corrupted state—the same request can be applied multiple times and yield the same result. Pair that with appropriate isolation practices within and across services to minimize anomalies and race conditions, so operations don’t see conflicting states more often than acceptable.

Beyond that, plan for eventual consistency by allowing updates to propagate asynchronously and define deterministic conflict resolution rules so competing updates settle in a known way. Patterns like sagas further help by decomposing a long-running transaction into a sequence of local actions, each with a compensating operation if later steps fail; this avoids locking the entire distributed system while still preserving overall consistency.

Why this beats the others: relying only on dry-run modes and client-side retries doesn’t provide a concrete mechanism to reconcile divergent states or prevent duplicates across services. Disabling caching might reduce stale reads, but it hurts performance and doesn’t address cross-service consistency. Trying to lock everything globally is typically infeasible in distributed systems—latency and partition tolerance make immediate global locks impractical or impossible.

So the best approach combines idempotence, sensible isolation, eventual consistency with conflict resolution, and saga-like patterns to maintain data correctness without heavy distributed locking.