Eventual Consistency and Conflict Resolution with CRDTs in Distributed Databases

The latency challenge at the speed of light

When building systems that serve users across the planet, physics imposes an insurmountable limit: the speed of light in fiber optic cables. Sending data from New York to Tokyo takes dozens of precious milliseconds. If your application requires every server worldwide to agree on a change before responding to the user, the entire system slows down and becomes vulnerable to internet outages.

To bypass this bottleneck, software engineering adopts multi-master replication. In practice, this means each region of the globe has its own database capable of accepting reads and writes independently and immediately. The user writes data to the closest server and receives an instant response, while synchronization with the rest of the world happens asynchronously in the background.

The divergence dilemma and eventual consistency

Allowing multiple servers to accept writes simultaneously creates a classic problem: data conflict. Imagine a user changing their email address in London while another user updates the phone number of the same profile in Sydney, all within the exact same fraction of a second. When the two servers talk to each other, the information collides, and the system must decide which version should prevail.

This is where eventual consistency comes in, a promise that all copies of the database will eventually align, provided they stop receiving new changes for a brief moment. In practice, this means accepting a temporary period of disorder where different data coexists in different parts of the world, trusting that an intelligent mechanism will sort out the confusion later without losing anyone's work.

How conflict-free replicated data types work

For years, conflict resolution depended on pessimistic locks that froze data access or complex consensus algorithms like Paxos and Raft that require real-time voting. CRDTs, or Conflict-Free Replicated Data Types, change this logic by introducing smart mathematical structures that make destructive conflicts impossible to happen.

In practice, a CRDT is like a shared document where anyone can edit any part at any time, but whose internal rules ensure the merge of all edits results in the exact same final text for everyone. This works because the mathematical operations applied to the data have special properties, such as commutativity, where the order in which changes arrive does not alter the final outcome.

State-based and operation-based structures

There are two major families of CRDTs in distributed systems architecture: state-based and operation-based. State-based models send the entire content of the data structure to other nodes periodically, where a special mathematical function merges the remote state with the local one, always keeping the highest value or the union of sets.

Operation-based models transmit only the executed command, such as 'add item X to the cart'. In practice, the challenge of the operation-based model is ensuring no message gets lost on the network, as missing a single command can break convergence. Therefore, reliable messaging systems or robust transport layers are fundamental to maintaining the integrity of these architectures.

Real-world use cases and practical application

Major tech companies use these approaches in products we use daily, such as collaborative text editors, offline-first note-taking apps, and e-commerce shopping carts. When you edit a document offline on an airplane and it syncs seamlessly upon reconnecting to the internet, there is a high probability that a CRDT-inspired structure is operating behind the scenes.

In practice, adopting this technology requires rethinking your application's data modeling. Instead of traditional mutable fields that overwrite values, you model the domain using grow-only counters, sets that accept additions and removals tracked by timestamps, or registers where the latest valid timestamp modification deterministically prevails. To complement this implementation in production environments, see our related technical analyses on Transaction Consistency and Conflict Resolution in Multi-Master Distributed Databases and Eventual Consistency and Conflict Resolution in Distributed Systems Based on CRDTs.

Final considerations on distributed resilience

Eventual consistency combined with CRDTs represents one of the most fascinating evolutions in modern systems engineering. It frees us from reliance on flawless connections and low latencies, allowing us to build truly global applications resilient to network failures. Although it requires a profound shift in how we think about data storage, the gain in availability and user experience amply justifies the implementation effort.