Time to build a CRDT! Specifically a grow-only counter that works across distributed nodes without coordination. This is where distributed systems theory gets practical.
What it Does
- Implements a grow-only (increment-only) counter
- Each node maintains its own counter value
- Nodes periodically sync to get the global view
- No coordination needed for increments
The CRDT Approach
Used a G-Counter (Grow-only Counter) CRDT:
- Each node tracks increments per node (vector of counters)
- To increment: just bump your own counter
- To read: sum all nodes' counters
- To merge: take the max for each node position
Key Properties
- Convergence: All nodes eventually see the same total
- No Coordination: Increments are pure local operations
- Partition Tolerant: Works even during network splits
- Monotonic: Counter never decreases
Implementation Details
- Vector of counters (one per node)
- Periodic state exchange between nodes
- Max-based merging for conflict resolution
- Clean separation of local vs global operations
Why CRDTs Are Cool
- Mathematical properties guarantee correctness
- No need for locks, consensus, or coordination
- Naturally partition tolerant
- Simple to reason about
Trade-offs
Pros:
- Zero coordination overhead
- Partition tolerant by design
- Eventually consistent
- Simple local operations
Cons:
- Space grows with number of nodes
- Only supports increment (no decrement)
- Network overhead for state sync
This pattern shows up everywhere in distributed systems - any time you need shared state without coordination, CRDTs are often the answer.
