CRDTs as Architecture Practice

CRDTs are a distributed-systems practice for designing replicated state that can accept concurrent updates and converge without synchronous coordination for every update.

The technical semantics are captured by CRDTs.

Problem

CRDTs address the problem of coordination under partitions, replication, offline updates, and low-latency local writes.

Cohesive Formulation

The practice moves part of the coordination problem into the algebra of the data type:

State-based CRDTs converge by merge.
Operation-based CRDTs converge by commutative operations under delivery assumptions.
Persistence and recovery must preserve causal metadata needed for convergence.
Invariants must be compatible with the merge or operation algebra, or else require additional coordination.

In the Model

CRDTs are suitable when the domain can tolerate temporary divergence and the intended state evolution can be modeled with monotonic merge or commutative updates.

The CALM theorem generalizes this practice beyond CRDT implementations: monotone application logic is the part of the system that can usually remain coordination-free.