Skip to content

Widhian Bramantya

coding is an art form

Menu
  • About Me
Menu
kafka

Delivery Semantics in Kafka: At Most Once, At Least Once, Exactly Once

Posted on September 22, 2025September 22, 2025 by admin

3. Exactly Once

  • Meaning: Each message is delivered once and only once.
  • No loss, no duplicates.
  • Hardest to achieve in distributed systems.

How Kafka Does It:

  • Kafka uses idempotent producers + transactional writes.
  • This ensures:
    • No duplicate writes from producers.
    • Consumers commit offsets as part of the same transaction as processing.

Example:

  • A payment system processes a message and commits offset in one atomic step.
  • Even if the system crashes, Kafka guarantees no double charging.

Benefits

  • No data loss and no duplicates.
  • Strongest guarantee possible.
  • Ensures perfect correctness in critical workflows.

Drawbacks

  • Slower performance (transaction overhead).
  • Higher complexity in configuration and operations.
  • Not always necessary (often “at least once + idempotent consumer” is good enough).

Use Case:

  • Financial transactions, inventory updates, or any workflow where duplication is unacceptable.

Diagram: Delivery Semantics

flowchart TD
    A[Message Produced] --> B[Delivered to Consumer]

    B -->|At Most Once| C[Processed 0-1 times<br/>Fast but risk of loss]
    B -->|At Least Once| D[Processed >=1 times<br/>No loss but duplicates]
    B -->|Exactly Once| E[Processed 1 time<br/>No loss, no duplicates]

Quick Summary

SemanticsData LossDuplicatesComplexityUse Cases
At Most Once✅ possible❌ noneLowLogs, metrics
At Least Once❌ none✅ possibleMediumPayments (idempotent), alerts
Exactly Once❌ none❌ noneHighBanking, inventory, critical data

Conclusion

  • At Most Once: simplest, but risky (data may be lost).
  • At Least Once: safest default (no data loss, but handle duplicates).
  • Exactly Once: strongest guarantee, but comes with extra complexity and cost.

In real-world systems, most teams use At Least Once with idempotent consumers (so duplicates don’t cause harm). Kafka’s Exactly Once is used only when absolutely necessary.

See also  Ensuring Message Ordering in NATS: A Kafka-like Approach

Related posts:

Getting Started with Apache Kafka: Core Concepts and Use Cases

Kafka Architecture Explained: Brokers, Topics, Partitions, and Offsets

RabbitMQ vs Kafka: Choosing the Right Messaging System for Your Project

Pages: 1 2
Category: Kafka

Leave a Reply Cancel reply

Your email address will not be published. Required fields are marked *

Linkedin

Widhian Bramantya

Recent Posts

  • Smart Automation in PostgreSQL: Managing Time-Based Data with pg_partman and pg_cron
  • Understanding PostgreSQL WAL, Slot, Publication, LSN, and Replication Lag
  • PostgreSQL Write-Ahead Log (WAL): Durability, Performance Tuning, and Recovery Explained
  • PostgreSQL Replication Deep Dive: From High Availability to Multi-Master Clusters
  • Finding Nearby Merchants in a Ride-Hailing App Using Elasticsearch Polygon Search
  • Advanced Text Search in Elasticsearch: N-Gram, Reverse, Fuzzy, and Search-as-you-type
  • Understanding and Customizing Analyzers in Elasticsearch
  • Log Management at Scale: Integrating Elasticsearch with Beats, Logstash, and Kibana
  • Index Lifecycle Management (ILM) in Elasticsearch: Automatic Data Control Made Simple
  • Blue-Green Deployment in Elasticsearch: Safe Reindexing and Zero-Downtime Upgrades
  • Maintaining Super Large Datasets in Elasticsearch
  • Elasticsearch Best Practices for Beginners
  • Implementing the Outbox Pattern with Debezium
  • Production-Grade Debezium Connector with Kafka (Postgres Outbox Example – E-Commerce Orders)
  • Connecting Debezium with Kafka for Real-Time Streaming
  • Debezium Architecture – How It Works and Core Components
  • What is Debezium? – An Introduction to Change Data Capture
  • Offset Management and Consumer Groups in Kafka
  • Partitions, Replication, and Fault Tolerance in Kafka
  • Delivery Semantics in Kafka: At Most Once, At Least Once, Exactly Once

Recent Comments

No comments to show.

Archives

  • October 2025
  • September 2025
  • August 2025
  • November 2021
  • October 2021
  • August 2021
  • July 2021
  • June 2021
  • March 2021
  • January 2021

Categories

  • Debezium
  • Devops
  • ElasticSearch
  • Golang
  • Kafka
  • Lua
  • NATS
  • PostgreSQL
  • Programming
  • RabbitMQ
  • Redis
  • VPC
© 2026 Widhian Bramantya | Powered by Minimalist Blog WordPress Theme