This article explains why introducing a message queue like Kafka helps decouple systems and control traffic, then dives into the three key questions of detecting, locating, and preventing message loss, offering concrete monitoring methods, configuration settings, and troubleshooting steps for producers, brokers, and consumers.
This article provides a comprehensive overview of Alluxio's metadata and data synchronization mechanisms, covering its unified namespace, mounting strategies, consistency models, various write modes, read workflows, metadata sync techniques, performance optimizations, and recommended configurations for different deployment scenarios.
Shopee’s Real‑time Checking System provides configurable, non‑intrusive data consistency verification for micro‑services by capturing change events via CDC, streaming them through Kafka, applying flexible rules and expressions, and instantly alerting mismatches, delivering second‑level detection while scaling to tens of thousands of checks per second.
This article explains why caching is essential for high‑concurrency systems, analyzes the challenges of keeping database and cache data consistent, and compares five practical cache‑invalidation strategies—including write‑after‑DB, delete‑before‑write, delayed double delete, queue‑based deletion, and binlog subscription—highlighting their trade‑offs and suitable scenarios.
This article explains how message queues help decouple services, enable asynchronous processing, smooth traffic spikes, and improve system resilience, while also detailing common challenges such as reduced availability, increased complexity, duplicate consumption, data consistency, message loss, ordering, and backlog, along with practical mitigation strategies.
After a decade of rapid growth, JD.com’s information system now handles procurement, sales, and inventory across a horizontally and vertically split architecture, tackling challenges of service decoupling, efficient communication, data consistency, and governance, while outlining its vision, principles, implementation details, and stability‑enhancing governance practices.
The paper describes iQIYI’s Content Platform Data Center (OLTP), a unified data hub that solves microservice data‑island problems by providing hundred‑billion‑scale storage, high‑QPS reads/writes, field‑level change notifications, final consistency between MongoDB and Elasticsearch, active‑active high availability, and a generic SDK, now serving 26 business lines with thousands of QPS and hundreds of millions of rows.
This article explains the "account conflict" problem caused by massive concurrent transactions, proposes splitting read/write operations and using lazy or timed data integration to improve performance and consistency, and discusses how to ensure atomicity and isolation with transactions or distributed locks.
This article introduces a custom MySQL data consistency verification tool inspired by pt-table-checksum, explains the challenges of handling complex primary keys, details the implementation logic with shell scripts and SQL, and provides installation, configuration, and usage instructions including parallelism, network monitoring, and scheduling.
The article examines how banks can select and implement cloud‑native persistent storage on Kubernetes‑based container platforms, covering storage types, CSI integration, DBaaS options, data consistency challenges, and strategies for high‑concurrency fault recovery in the financial sector.
This article explores the challenges of data consistency in distributed systems, explains fundamental concepts such as CAP and BASE, compares industry solutions like 2PC, Saga, TCC, and local message tables, and shares practical insights from payment service refactoring and Alipay's DTS framework.
The article introduces the freight BCS system, explains its business background, describes multiple validation modes for data consistency and business logic correctness, compares implementation approaches, and outlines the architecture, task flow, and future enhancements to improve system reliability and operational monitoring.
From its 2017 v2.0 upgrade to a service‑oriented design, vivo Mall’s product system was split from the monolithic mall, progressively adding independent activity, flash‑sale, consignment, and inventory services while employing rate limiting, multi‑level caching, circuit breaking, and distributed transactions to ensure stability, high performance, and data consistency.
Data Consistencybackend developmentcaching strategies
This article explains what data consistency means for MySQL and Redis, analyzes inconsistency cases in both single‑threaded and concurrent environments, and proposes practical strategies—including read‑only and read‑write cache handling, message‑queue retries, binlog subscription, delayed double‑delete, and distributed locking—to achieve eventual or strong consistency.
This article explains when and how to use caching with Redis, compares recursive and table‑lookup algorithms for factorial calculation, discusses cache design, expiration policies, and presents several write‑through and write‑behind strategies to maintain data consistency between Redis and MySQL under various concurrency scenarios.
To keep MySQL and Redis synchronized, the article defines consistency, examines read‑only and read‑write cache inconsistency cases, and recommends practical strategies such as update‑DB‑then‑delete‑cache with message‑queue retries, binlog‑driven updates, delayed double‑delete, and Redis distributed locks, while outlining stronger protocols and key design considerations.
The article examines the 2012 GitHub data‑leak caused by MySQL master‑slave failover and auto_increment misuse, explains why developers' expectations of uniqueness, monotonicity and continuity are unrealistic, and proposes half‑sync replication and distributed ID algorithms like Snowflake as more reliable solutions.
The article outlines five essential dimensions—data consistency, isolation, extensibility, business sequencing, and high‑concurrency handling—explaining why each matters, illustrating them with concrete examples, and showing how to evaluate and improve a technical design accordingly.
The article explains common microservice transaction patterns—including blocking retry, asynchronous queues, TCC compensation transactions, and local message tables—detailing their implementations, advantages, drawbacks, and practical code examples for ensuring data consistency in distributed systems.
The article examines common strategies for handling service call failures and maintaining data consistency in microservice architectures, comparing blocking retries, asynchronous queues, TCC compensation transactions, local message tables, and MQ transactions, highlighting their trade‑offs, implementation details, and practical considerations.
To prevent duplicate OpenID records caused by concurrent synchronization requests in the Fast App Center, this article analyzes the root cause, evaluates database‑level unique indexes versus application‑level distributed locks, and presents a Redis‑based lock implementation with cleanup procedures to ensure data consistency.
The article examines how to prevent data inconsistency during cross‑service money transfers by using local transactions, two‑phase commit, and message‑queue based eventual consistency, providing detailed code examples, performance considerations, and practical solutions for large‑scale backend systems.
The article describes Xianyu’s local services architecture, tackling rapid supplier onboarding, heterogeneous quality, and stability by reusing core platform capabilities, defining merchant, audit, and independent business domains, employing high‑concurrency rate limiting, idempotent retries, unified exception handling, status‑change logging, and proactive monitoring with alerts and reporting.
This article examines common techniques for handling service call failures in micro‑service architectures—blocking retries, asynchronous queues, TCC compensation transactions, local message tables, and MQ transactions—detailing their implementations, pitfalls, and trade‑offs to achieve reliable data consistency.
DCheck is a real‑time asset‑loss detection platform that monitors MySQL binlog and MQ streams, matches upstream and downstream states, and triggers alerts via configurable topics, events, sub‑events and Groovy scripts implementing filter and check logic, while offering rule‑based controls, debugging tips, and proposed enhancements for reliability.
Asset Loss PreventionData ConsistencyGroovy Scripts
This article examines the core concepts, requirements, architectural models, persistence strategies, scalability, high‑availability mechanisms, performance optimizations, security models, and practical considerations of distributed file systems such as HDFS, GFS, and Ceph, offering a comprehensive guide for engineers and researchers.
Data ConsistencyHigh Availabilitydistributed file system
The article proposes a comprehensive microservice‑based caching solution for low‑frequency static data such as vehicle models and user profiles, detailing why caching is needed, why Redis and persistent queues are chosen, how consistency checks work, and the trade‑offs compared with simple expiration strategies.
The article explains how lack of native transactions in MongoDB 3.0 can cause data inconsistency during order processing, compares it with SQL transaction mechanisms, and proposes optimization and compensation strategies such as retry queues, asynchronous tasks, and refund handling to ensure eventual consistency.
This article explains a reusable caching architecture for low‑frequency static data in microservice systems, covering why caching is needed, the role of Redis, persistent queues, consistency checks, and trade‑offs such as cache expiration and operational complexity.
This article examines the challenges of synchronizing enterprise data across systems when region identifiers differ, explores several technical approaches—including persisting local tables, name‑based and code‑based lookups, snapshot storage, manual updates, and flexible API fields—and concludes with a pragmatic, business‑centric solution for reliable integration.
This article outlines the background, design choices, and implementation details of a data‑consistency engine for an advertising billing platform, comparing TCC and saga‑style approaches, describing the state‑machine architecture, configuration, initialization, and asynchronous execution patterns.
This article provides a comprehensive overview of Zookeeper, covering its purpose, cluster roles, Znode types, session handling, watcher mechanism, ACL permissions, common use cases, data consistency via the ZAB protocol, leader election, synchronization methods, potential inconsistency scenarios, and a comparison with other service‑registry solutions.
The article reviews common cache‑side data‑sync patterns, highlights their inconsistency and data‑loss risks under high load, and presents JD's solution that combines Cache‑Aside, Change Data Capture, message queues, delayed consumption, versioning, and persistence to ensure eventual consistency between cache and relational databases.
This article explains the technical details of data cleaning, dirty‑data handling, and three methods—database read‑only, application termination, and network ACL isolation—to ensure data consistency during the data‑regulation and cutover phases of cross‑cloud migration, illustrated with real‑world case studies.
This article explains practical techniques for handling dirty data, guaranteeing data consistency, and performing validation during the data cleaning and cut‑over phases of a cross‑cloud migration, including read‑only database accounts, process termination, and network ACL isolation.
This article explains why cache‑database inconsistencies occur in large systems, details the cache‑aside, read‑through, write‑through and write‑behind strategies, highlights three common pitfalls with concrete examples, and offers practical recommendations such as proper update ordering and cache expiration to ensure data freshness.
The article compares three approaches—synchronous dual write, asynchronous dual write with a message queue, and CDC via Canal—to keep Elasticsearch and a relational database consistent for cross‑database aggregated full‑text search, outlining their steps, advantages, and drawbacks.
This article explains the importance of data consistency in distributed microservice systems, introduces the Saga pattern as a solution, compares it with two‑phase commit and TCC, and details its architecture, recovery mechanisms, and implementation considerations within ServiceComb.
This article explains why data consistency is critical in microservice architectures, introduces the Saga pattern and its execution and recovery mechanisms, compares it with two‑phase commit and TCC, and presents a centralized Saga design using ServiceComb for reliable distributed transactions.
This article explains how to use Percona Toolkit's pt-table-checksum tool to detect and verify data consistency between MySQL master and replica servers, covering its principles, suitable scenarios, safety mechanisms, and step‑by‑step commands for setup, data generation, and checksum execution.
This article reviews the evolution from traditional local and distributed transactions to BASE theory and presents four microservice data‑consistency patterns—reliable event notification, maximum‑effort notification, business compensation, and TCC—detailing their principles, advantages, drawbacks, and implementation examples.
Cache is a temporary storage layer that speeds data access by keeping frequently used items close to the processor, spanning hardware (CPU registers, multi‑level caches) to software (database, Redis, Memcached), and requires careful design to avoid penetration, breakdown, and consistency issues through techniques such as empty‑result caching, Bloom filters, pre‑warming, jittered expirations, logical refreshes, and multi‑level strategies.
Journal file systems use write‑ahead logging to record each write operation as a transaction, ensuring that after power loss or crashes the system can replay logs and maintain metadata and user‑data consistency, avoiding corruption and space waste through techniques like data, ordered, and metadata journaling.
This article introduces the limitations of traditional local and distributed transactions for microservices, explains the BASE theory, and details four practical patterns—reliable event notification, maximum‑effort notification, business compensation, and TCC—providing code examples, diagrams, and a comparative table to guide developers in achieving eventual consistency across microservice architectures.
Uber engineers rebuilt their payment system, Gulfstream, into a SOX‑compliant, job‑based double‑entry ledger that migrates billions of transactions across asynchronous services while maintaining zero‑downtime, high availability, idempotency, and strong data consistency.
Data ConsistencyDistributed SystemsHigh Availability
The article reviews microservice data consistency challenges, explains why traditional distributed transactions like 2PC/3PC are unsuitable, introduces the BASE theory, and details four implementation patterns—reliable event notification, maximum effort notification, business compensation, and TCC—to achieve eventual consistency.
The article explains how MySQL stores persistent data while Redis serves read‑only data, outlines read/write request handling, discusses consistency issues in low and high concurrency scenarios, and provides practical solutions such as cache invalidation, queue‑based updates, and deployment considerations.
This article explains the classic Cache‑Aside pattern, analyzes common cache‑database consistency problems in high‑traffic systems, and presents practical solutions—including delete‑first updates, internal JVM queues, lazy recomputation, and routing considerations—to ensure reliable data synchronization under heavy concurrency.
This article explains how to restore a single table on a MySQL replica after GTID‑based replication errors by using backup snapshots, replication filters, and controlled start‑up points to keep data consistent across master and slave.
This article provides a comprehensive overview of distributed file systems, covering their historical evolution, essential design requirements, centralized and decentralized architecture models, persistence, scalability, high availability, performance optimization, security, and additional practical aspects such as space allocation, file deletion, small‑file handling, and deduplication.
Data ConsistencyHigh Availabilitydistributed file system
Microservice architectures struggle with traditional ACID transactions, so this article reviews local and distributed transaction basics, explains why 2PC/3PC are unsuitable, introduces the BASE model, and details four practical consistency patterns—reliable event, async event, business compensation, and TCC—highlighting their mechanisms, advantages, and drawbacks.
This article explains the data‑delay problem inherent in asynchronous MySQL master‑slave replication, demonstrates how Sharding‑JDBC provides read‑write splitting and forced‑master routing via HintManager, and offers configuration and code examples as well as alternative async‑query strategies for real‑time consistency.
This article shares practical guidance on microservice benefits, common pitfalls such as stability and data consistency issues, and detailed solutions including circuit breakers, service degradation tactics, TCC distributed transactions, transactional messaging with RocketMQ, seamless data migration, and full‑stack APM monitoring.
This article systematically introduces the advertising transmission process, outlines challenges of scale and stability, and presents a comprehensive set of architectural, operational, and performance optimizations—including primary‑backup high availability, stateless incremental processing, message reliability, flow tiering, indexing sharding, and schedule management—to achieve high‑throughput, low‑latency, and eventually consistent ad delivery.
This article explains how to design database schemas and choose between MySQL and Redis for handling likes, comments, and bookmark features in mobile applications, covering requirements, schema examples, query patterns, scaling challenges, and data consistency considerations.
This article presents a comprehensive case study of migrating a 200‑plus node Hadoop offline platform across data centers, covering background, architecture, solution evaluation, detailed implementation steps, consistency checks, operational safeguards, encountered issues, and future recommendations.
This article details Ctrip's practical design and implementation of a bidirectional Redis synchronization system across data centers, covering replication loop breaking, conflict resolution with Last Write Wins and Vector Clocks, tombstone handling, garbage collection strategies, and expiration policies to ensure data consistency.
Static business data in microservice systems, such as vehicle models and user profiles, change rarely yet demand high accuracy and real‑time access; this article proposes a universal caching architecture using a business service, persistent queue, Redis cache, and consistency checks to achieve scalable, reliable reads.
This article proposes a universal cache processing solution for low‑frequency static business data in microservice systems, detailing why caching is needed, the role of services, queues, Redis, consistency checks, and trade‑offs such as cache expiration, aiming to achieve high‑throughput, real‑time queries while ensuring data reliability.
This article introduces the limitations of traditional local and distributed transactions for microservices, explains the BASE theory, and details four major patterns—reliable event notification, max‑effort notification, business compensation, and TCC—to achieve eventual consistency in microservice architectures.
Data Consistencydistributed transactionseventual consistency
The article examines how financial platforms can acquire user accounts, design fast‑perceived business experiences through account mapping, capital advances, and cross‑account interoperability, and build stable, high‑performance backend and transaction systems that balance compliance, user experience, and operational costs.
The article explains why achieving data consistency across multiple microservices is challenging, reviews the limitations of two‑phase commit, and presents practical techniques such as the Saga pattern, reconciliation, event logs, orchestration vs. choreography, and change‑data‑capture to keep distributed systems eventually consistent.
The article proposes a universal caching solution for low‑frequency static business data in microservice systems, detailing why caching is required, the role of services, queues, Redis, consistency checks, and trade‑offs such as cache expiration and operational complexity.
The article explains three approaches to ensuring data consistency in big‑data systems—basic multi‑node replication, HDFS pipeline replication, and Elasticsearch primary‑replica replication—detailing their workflows, advantages, and drawbacks.
The article explains the pitfalls of updating databases and caches separately, illustrates race conditions that cause data inconsistency, and introduces the Cache Aside strategy as a safer approach that reduces inconsistency by deleting stale cache entries before reads.
This article explains cache read and write mechanisms, the impact of operation order on data consistency, and presents optimization techniques such as delayed double deletion and binlog subscription to mitigate inconsistency in high‑concurrency backend systems.
This article explains core caching concepts, common pitfalls such as cache penetration, breakdown and avalanche, presents classic update patterns like Cache‑Aside, Write‑Through and Write‑Behind, analyzes consistency challenges, and offers practical guidelines for designing robust multi‑level cache architectures.
To mitigate the inherent single‑point‑failure risk of the open‑source InfluxDB community edition, the article proposes deploying multiple InfluxDB instances with concurrent client writes, tracking failed writes, temporarily storing them, and using custom workers to replay data, while addressing timeout, data consistency, and storage considerations.
This article describes how an e‑commerce platform built a big‑data pipeline—including TTLog, Lindorm, BCP, MetaQ, Blink and Xflush—to detect and verify search advertising placement errors in real time, covering background, implementation details, core challenges, and future optimization directions.
This article examines a universal caching solution for business static data in microservice systems, detailing the definition of static data, the need for caching, a six‑component architecture involving Redis, queues, and consistency checks, and the trade‑offs of various design choices.
This article explains what constitutes data inconsistency in Redis, outlines key and value mismatch scenarios, and introduces the open‑source redis‑full‑check tool with usage instructions for efficiently comparing two Redis instances across various deployment modes.
The article outlines essential techniques for building highly available internet services, covering system availability metrics, multi‑level caching, database and service isolation, concurrency control, gray‑release deployment, comprehensive monitoring, graceful degradation, asynchronous design, and data‑consistency scenarios for both real‑time and offline big‑data workloads.
In this interview, Qunar.com’s database director Wang Zhufeng shares his six‑year journey, discusses the significance of the DTCC conference, outlines the evolution of Qunar’s MySQL architecture toward clustering, cloud‑native solutions, and data consistency, and offers advice for aspiring database engineers.
This article provides an in‑depth overview of why Redis is used, its drawbacks, the reasons behind its single‑threaded speed, data types and use‑cases, expiration policies, memory eviction strategies, consistency challenges with databases, and practical solutions for cache penetration, cache avalanche, and concurrent key competition.
The article analyzes QQGame’s room‑entry failures caused by massive concurrent users, explores the limits of client‑side data synchronization, and proposes a scalable backend architecture using region‑based partitioning, autonomous server processing, and distributed database sharding to achieve high availability and data consistency.
This article analyzes data‑inconsistency problems caused by concurrent multithreaded updates in a CMDB system and presents a step‑by‑step exploration of synchronization solutions, ultimately implementing a robust Redis‑backed distributed lock in Python to guarantee atomic reads and writes while handling edge cases such as crashes and lock expiration.
The article argues that cold backup and active‑passive setups provide false security and outlines how true active‑active disaster‑recovery requires local‑datacenter request handling, business‑driven data sharding, and low‑latency cross‑site synchronization, recommending a staged rollout from city‑level to cross‑region architectures while weighing ROI.
Data Consistencyactive-active-architecturecloud-native
Xianyu migrated its tens‑billion‑row product database from a shared MySQL cluster to an isolated instance by refactoring with dual‑write, employing TDDL for sharding and sequence management, and using Alibaba Cloud DTS for full‑load, incremental sync, consistency verification, and a no‑data‑loss rollback, achieving seamless, user‑transparent migration and improved stability.
Journal file systems protect against data corruption caused by power loss or crashes by recording each write operation as a transaction in a dedicated log, then committing the changes only after the log is safely stored, enabling replay to restore consistency.
The article describes YouZan's business reconciliation platform for distributed systems, which detects and quantifies data inconsistencies by offering easy plug‑in integration, a four‑step orchestrated workflow, high‑throughput offline processing with Spark, second‑level real‑time event handling, a three‑layer architecture, and health monitoring for transaction chains.
This article explains the fundamental reasons for building distributed systems, examines the inevitable side‑effects—especially data consistency challenges—analyzes the root causes of inconsistency, and walks through various consistency models from eventual to linearizability with clear examples and illustrations.
Data ConsistencyDistributed Systemsconsistency models
Microsoft's Volume Shadow Service (VSS) is a backup and recovery framework that creates consistent point-in-time snapshots by coordinating requestors, writers, and providers, supporting fast data backup, file-level restores, and various storage scenarios through full and copy-on-write snapshot methods.
This article explains the consequences of high‑traffic spikes, presents practical database and code‑level strategies to keep data consistent, and outlines server‑side architectures—including load balancing, caching, and Redis queues—to sustain massive concurrent requests without crashing.
The article describes Meituan’s reconciliation system for delivery settlement, which ingests diverse external data, normalizes it, performs multi‑stage offline and online verification to detect missing, duplicate, or incorrect settlements, and automatically records and resolves errors through a unified balancing center, ensuring fund safety at tens‑million‑order scale.
This article details the background, design considerations, migration workflows, tooling choices, data consistency verification, rollback mechanisms, and practical experiences of moving a large‑scale production environment from Microsoft SQL Server to MySQL, covering both offline and online migration scenarios.
This article details JD.com's multi-active distributed architecture, covering its evolution from single‑data‑center to multi‑region deployments, network design, leaf‑spine topology, data consistency mechanisms, application scheduling, monitoring, and disaster recovery strategies that enhance high availability and user experience.
This article explains microservice architecture fundamentals, typical design patterns such as chain, aggregator, data‑sharing and asynchronous messaging, and presents practical solutions for service splitting, data consistency, and distributed transaction management based on real‑world experience at the ZhaiZhai platform.
The article explains how to break large transactions into small atomic operations combined with asynchronous messaging, describes ACID properties, presents banking and flow‑recharge scenarios, compares local and distributed (flexible) transactions, and details micro‑service architecture, compensation and async strategies to achieve eventual consistency.
Data Consistencyasynchronous notificationcompensation
This article explains the principles of distributed transaction design, covering ACID fundamentals, typical banking and e‑commerce scenarios, various transaction models such as two‑phase, compensation, asynchronous and best‑effort notifications, and presents a micro‑service architecture with concrete flow diagrams and scaling strategies for a high‑throughput traffic‑recharge platform.
Data Consistencydistributed transactionseventual consistency
The article explains how achieving high availability through data redundancy introduces consistency challenges that in turn affect performance, and it reviews partitioning, mirroring, consistency models, replication architectures, and two/three‑phase commit protocols in distributed systems.
Data ConsistencyDistributed SystemsHigh Availability
Ensuring final data consistency in modern applications—whether enterprise or internet—requires understanding ACID, CAP, BASE theories and applying practical techniques such as single-database transactions, two-phase commit, transactional message queues, compensation tasks, async callbacks, and double-check mechanisms to build robust distributed systems.
The article chronicles the Qinglong logistics platform from its 2012 MVP launch through successive versions to a smart‑logistics system, detailing architectural evolution, high‑availability, performance, data‑consistency strategies, and user‑experience practices that underpin large‑scale backend development.
To keep its rapidly expanding hotel‑direct platform consistent despite unstable supplier interfaces, Meituan evolved from full‑batch pulls to segmented fetching, predictive trigger‑based updates, and finally supplier‑initiated pushes, creating a hybrid pull‑push architecture that ensures low‑latency, reliable product and inventory data.
Data ConsistencyMySQL replicationbackend development
The live‑streaming gift system is built with a producer‑consumer message queue, KV store, cache, SQL and SSD layers, employing asynchronous leaderboard updates, unique transaction IDs with redo handling, end‑to‑end encryption and replay protection to ensure high‑consistency, real‑time, secure processing of monetary gifts.
This article examines common misconceptions in designing geo-distributed active-active systems, explains why striving for perfect real-time data sync is unrealistic, and offers practical strategies—such as prioritizing core services, reducing distance, limiting data replication, and combining storage sync with messaging—to achieve reliable high-availability.
This article examines MySQL's semi‑synchronous replication, analyzes consistency scenarios across master‑slave failures and switches, and highlights the difficulties of data rollback and multi‑master conflicts that affect high‑availability clusters.
Data ConsistencyHigh Availabilitydisaster recovery
This article explores typical cache challenges in distributed systems—including data consistency, high availability, cache avalanche, and cache penetration—explaining their causes, real‑world scenarios, and practical mitigation strategies to ensure reliable and efficient caching.
This guide explains why MySQL replication can fail with errors 1452, 1032, and 1062, analyzes each root cause, and provides both quick‑fix procedures and a permanent solution using pt‑table‑checksum and pt‑table‑sync to keep master‑slave data consistent.
The article examines the challenges of maintaining data consistency across distributed services in e‑commerce, explains strong, weak and eventual consistency models, and presents six practical solutions—including business integration, the eBay BASE pattern, Qunar, Mogujie, Alipay DTS, and Nongxin—highlighting how local transactions, idempotent messaging, and eventual consistency can replace heavyweight distributed‑transaction frameworks.