ArrowLeft Icon

Distributed Transactions: Navigating the Complex World of Data Consistency

📆 · ⏳ 6 min read · · 👀


Hey there! Today, let’s dive into the intriguing world of distributed transactions. Imagine you’re orchestrating a symphony with musicians spread across different cities. Ensuring they play in perfect harmony, without any notes out of sync, is quite a challenge. Similarly, in the world of distributed systems, maintaining data consistency across multiple services can be a complex task.

When we talk about transactions in a single database, it’s like conducting a musical performance in one room, where every musician follows the same conductor’s baton. But when it comes to distributed transactions, it’s like conducting the symphony with musicians in different cities, each following their own conductor. Coordinating their performance and ensuring a harmonious outcome becomes a delicate dance.

The Complexity of Distributed Transactions

Distributed transactions offer the promise of seamless data operations across different systems. However,

With great power comes great complexity.

— Uncle Ben from Tech Multiverse

In a distributed environment, ensuring that multiple operations are either all successful or all rolled back is not as straightforward as a single-database transaction.

Two-Phase Commit Protocol - The Maestro’s Baton

Imagine you have multiple conductors leading each section of the symphony, and they all communicate with the main conductor before performing their parts.

The Two-Phase Commit Protocol works in a similar manner, ensuring that all participating systems are ready to commit a transaction before the final decision is made.

Phase One: The Prepare Phase

The 2PC Protocol divides the transaction into two crucial phases. The first phase, known as the “Prepare Phase,” is all about coordination and agreement.

Let’s dive deeper into what happens during this phase:

  • Coordinator’s Query: In a distributed system, one node takes on the role of the coordinator, while others are participants.

    The coordinator initiates the transaction and sends a “prepare” request to all participants involved in the transaction.

  • Participant’s Response: Upon receiving the “prepare” request, each participant assesses its ability to commit the transaction successfully. It checks whether all the necessary resources are available and the operation can proceed without errors.

    If everything looks good, the participant responds with a “YES” vote. Otherwise, it replies with a “NO” vote, signaling that it’s unable to commit.

  • Coordinator’s Decision: The coordinator patiently awaits responses from all participants. It collects their votes and makes a decision based on the received responses.

    If all participants respond with “YES”, indicating their readiness to commit, the coordinator moves to the second phase.

    However, if even one participant responds with a “NO”, the coordinator decides to abort the transaction, ensuring data consistency by preventing a partial commit.

Phase Two: The Commit (or Abort) Phase

With a unanimous “YES” vote from all participants, the coordinator enters the second phase, the “Commit Phase”. Here’s how it unfolds:

  • Coordinator’s Commit Request: The coordinator sends a “commit” request to all participants, signifying their commitment to executing the transaction.

  • Participant’s Action: Upon receiving the “commit” request, each participant carries out the transaction’s operation. This typically involves making changes to the local data or resource as dictated by the transaction.

  • Final Acknowledgment: Once the participant successfully completes its part of the transaction, it sends an acknowledgment back to the coordinator. This acknowledgment confirms that the operation was executed as intended.

  • Coordinator’s Transaction Confirmation: The coordinator waits for acknowledgments from all participants. When it receives all acknowledgments, it considers the transaction officially committed.

    It sends a final “commit” message to all participants, and at this point, the transaction is deemed successful.

Eventual Consistency - Embracing Reality

Sometimes, perfection is an illusion. Just like we may have slight delays in a global symphony, distributed systems may encounter eventual consistency.

This means that, after a transaction, it might take a little time for all systems to converge to the same state. It’s a trade-off between immediate consistency and performance in a distributed world.

At its core, eventual consistency acknowledges that in a distributed system, achieving instant consistency across all nodes, especially in the presence of network partitions or failures, can be impractical.

Instead, it adopts a more lenient stance, accepting that there may be temporary discrepancies in data between nodes, but these will eventually resolve.

Key Characteristics of Eventual Consistency

  • Asynchronous Updates: In systems designed for eventual consistency, updates to data are typically asynchronous.

    When a change occurs, it is propagated to other nodes over time. During this propagation, some nodes may have more up-to-date data than others.

  • Lack of Strict Guarantees: Eventual consistency doesn’t provide the strict guarantees of immediate consistency.

    In other words, if you read data from one node and then immediately from another, you might observe different values, which can be disconcerting in traditional database systems.

  • Resolution Mechanisms: To achieve eventual consistency, systems implement mechanisms to reconcile conflicting updates.

    This may involve techniques like conflict resolution algorithms, version vectors, or last-write-wins policies. These mechanisms ensure that over time, all nodes converge to a consistent state.

Saga Pattern - Choreographing the Dance

In a ballet performance, individual dancers have their moves, but together they create a captivating saga.

The Saga Pattern applies the same concept to distributed transactions. It breaks a large transaction into smaller, manageable steps, ensuring that each step can be either committed or compensated for in case of failure.

Instead of enforcing strong consistency across all services involved in a transaction, the Saga Pattern embraces the concept of “compensating actions” or “saga steps” to ensure eventual consistency.

Key Concepts of the Saga Pattern

  • Transactional Steps: A saga is composed of multiple transactional steps, each representing an atomic action within a distributed transaction. These steps can be thought of as dance moves in our choreography.

  • Local Transactions: Each step within a saga performs a local transaction on its own data. It’s responsible for maintaining the consistency of its local state.

  • Compensating Actions: If a step in the saga fails or encounters an error, the Saga Pattern relies on compensating actions to revert or undo the effects of previous steps.

    These compensating actions mirror the original steps but are executed in reverse.


In conclusion, distributed transactions are like orchestrating a symphony across multiple stages. It requires coordination, communication, and handling failures gracefully. From the Two-Phase Commit Protocol to embracing eventual consistency and adopting the Saga Pattern, there are multiple techniques to tackle the challenges of data consistency in distributed systems.

So, the next time you venture into the world of distributed systems and face the complexities of maintaining data integrity, remember these patterns and strategies. Embrace the art of conducting distributed transactions, and let your applications resonate with seamless harmony in a distributed world. Happy orchestrating!

EnvelopeOpen IconStay up to date

Get notified when I publish something new, and unsubscribe at any time.

You may also like

  • # system design# database

    Choosing the Right Data Storage Solution: SQL vs. NoSQL Databases

    Navigating the world of data storage solutions can be like choosing the perfect tool for a job. Join me as we dive into the dynamic debate of SQL and NoSQL databases, understanding their strengths, limitations, and where they best fit in real-world scenarios.

  • # system design

    Raft and Paxos: Distributed Consensus Algorithms

    Dive into the world of distributed systems and unravel the mysteries of consensus algorithms with Raft and Paxos. In this blog, we'll embark on a human-to-human exploration, discussing the inner workings of these two popular consensus algorithms. If you have a solid grasp of technical concepts and a curious mind eager to understand how distributed systems achieve consensus, this guide is your ticket to clarity!

  • # system design

    Understanding Load Balancing Algorithms: Round-robin and Consistent Hashing

    Welcome to the world of load balancing algorithms, where we unravel the magic behind Round-robin and Consistent Hashing. If you have a solid grasp of technical concepts and are eager to understand how these algorithms efficiently distribute traffic across servers, this blog is your ultimate guide. We'll embark on a human-to-human conversation, exploring the inner workings of Round-robin and Consistent Hashing, and how they keep our systems scalable and performant.