Depends on #34 (Multi-Node-Spec) for convergence testing. This is the largest single roadmap item — recommended last.
Two related concepts in two phases:
Phase 5a: CRDT library (standalone, no persistence integration)
Conflict-free replicated data types — distributed structures that converge under replication without coordination. Akka has ddata (GCounter, PNCounter, GSet, ORSet, LWWMap, ORMap, …); we ship a starter set:
GCounter (grow-only counter)PNCounter (positive/negative counter)GSet (grow-only set)ORSet (observed-remove set)LWWRegister (last-writer-wins register)
Each implements merge(a, b) → c (idempotent, commutative, associative) and value(): T. A DistributedData extension on the cluster wraps them in a gossip-replicated key-value store — update(key, fn), subscribe(key), delete(key).
Phase 5b: Replicated Event Sourcing
On the persistence layer. Multiple nodes can write events for the same persistenceId in parallel; a conflict-resolution mechanism (vector clocks + pluggable resolver: LWW or custom merge) merges divergent histories deterministically.
Components:
| File |
Task |
| src/crdt/{GCounter,PNCounter,GSet,ORSet,LWWRegister}.ts (new) |
pure CRDT data types |
| src/crdt/DistributedData.ts (new) |
extension + gossip-replicated key-value store |
| src/persistence/ReplicatedEventSourcedActor.ts (new) |
phase 5b |
| src/persistence/replicated/VectorClock.ts (new) |
logical timestamps |
| src/persistence/replicated/ConflictResolver.ts (new) |
LWW + custom merge plugin point |
| tests/unit/crdt/** (new) |
property tests for idempotency / commutativity / associativity |
| tests/multi-node/replicated-es.test.ts (new) |
requires #34 |
| examples/crdt/shopping-cart-orset.ts (new) |
classic OR-Set demo |
| examples/persistence/replicated-counter.ts (new) |
replicated event-sourced counter |
Estimate: 8-12 days. This estimate is optimistic if we want provably-correct garbage collection of old vector-clock entries — which we explicitly defer to v2.
Out of scope (for first iteration):
- Cross-DC replication
- VC garbage collection
- Delta-CRDTs (we do state-based for simplicity first)
Verification:
- CRDT property tests:
merge(a, a) === a (idempotent), merge(a, b) === merge(b, a) (commutative), merge(merge(a, b), c) === merge(a, merge(b, c)) (associative). Property-based via fast-check or written by hand.
- Multi-node tests: 3-node cluster, parallel updates from each, convergence check after gossip quiesces.
See the roadmap plan for full context (item 5 of 5).
Depends on #34 (Multi-Node-Spec) for convergence testing. This is the largest single roadmap item — recommended last.
Two related concepts in two phases:
Phase 5a: CRDT library (standalone, no persistence integration)
Conflict-free replicated data types — distributed structures that converge under replication without coordination. Akka has
ddata(GCounter, PNCounter, GSet, ORSet, LWWMap, ORMap, …); we ship a starter set:GCounter(grow-only counter)PNCounter(positive/negative counter)GSet(grow-only set)ORSet(observed-remove set)LWWRegister(last-writer-wins register)Each implements
merge(a, b) → c(idempotent, commutative, associative) andvalue(): T. ADistributedDataextension on the cluster wraps them in a gossip-replicated key-value store —update(key, fn),subscribe(key),delete(key).Phase 5b: Replicated Event Sourcing
On the persistence layer. Multiple nodes can write events for the same persistenceId in parallel; a conflict-resolution mechanism (vector clocks + pluggable resolver: LWW or custom merge) merges divergent histories deterministically.
Components:
Estimate: 8-12 days. This estimate is optimistic if we want provably-correct garbage collection of old vector-clock entries — which we explicitly defer to v2.
Out of scope (for first iteration):
Verification:
merge(a, a) === a(idempotent),merge(a, b) === merge(b, a)(commutative),merge(merge(a, b), c) === merge(a, merge(b, c))(associative). Property-based via fast-check or written by hand.See the roadmap plan for full context (item 5 of 5).