- Add ErrVersionConflict error type and VersionConflictError for detailed conflict information - Implement version validation in InMemoryEventStore.SaveEvent that rejects events with version <= current latest version - Implement version validation in JetStreamEventStore.SaveEvent with version caching for performance - Add comprehensive tests for version conflict detection including concurrent writes to same actor - Document versioning semantics in EventStore interface and CLAUDE.md This ensures events have monotonically increasing versions per actor and provides clear error messages for version conflicts, enabling optimistic concurrency control patterns. Closes #6 Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
4.4 KiB
4.4 KiB
Aether
Distributed actor system with event sourcing for Go, powered by NATS.
Organization Context
This repo is part of Flowmade. See:
- Organization manifesto - who we are, what we believe
- Repository map - how this fits in the bigger picture
- Vision - what this specific product does
Setup
git clone git@git.flowmade.one:flowmade-one/aether.git
cd aether
go mod download
Requires NATS server for integration tests:
# Install NATS
brew install nats-server
# Run with JetStream enabled
nats-server -js
Project Structure
aether/
├── event.go # Event, ActorSnapshot, EventStore interface
├── eventbus.go # EventBus, EventBroadcaster interface
├── nats_eventbus.go # NATSEventBus - cross-node event broadcasting
├── store/
│ ├── memory.go # InMemoryEventStore (testing)
│ └── jetstream.go # JetStreamEventStore (production)
├── cluster/
│ ├── manager.go # ClusterManager
│ ├── discovery.go # NodeDiscovery
│ ├── hashring.go # ConsistentHashRing
│ ├── shard.go # ShardManager
│ ├── leader.go # LeaderElection
│ └── types.go # Cluster types
└── model/
└── model.go # EventStorming model types
Development
make build # Build the library
make test # Run tests
make lint # Run linters
Architecture
Event Sourcing
Events are the source of truth. State is derived by replaying events.
// Create an event
event := &aether.Event{
ID: uuid.New().String(),
EventType: "OrderPlaced",
ActorID: "order-123",
Version: 1,
Data: map[string]interface{}{"total": 100.00},
Timestamp: time.Now(),
}
// Persist to event store
store.SaveEvent(event)
// Replay events to rebuild state
events, _ := store.GetEvents("order-123", 0)
Event Versioning
Events for each actor must have monotonically increasing versions. This ensures event stream integrity and enables optimistic concurrency control.
Version Semantics
- Each actor has an independent version sequence
- Version must be strictly greater than the current latest version
- For new actors (no events), the first event must have version > 0
- Non-consecutive versions are allowed (gaps are permitted)
Optimistic Concurrency Pattern
// 1. Get current version
currentVersion, _ := store.GetLatestVersion("order-123")
// 2. Create event with next version
event := &aether.Event{
ID: uuid.New().String(),
EventType: "OrderUpdated",
ActorID: "order-123",
Version: currentVersion + 1,
Data: map[string]interface{}{"status": "shipped"},
Timestamp: time.Now(),
}
// 3. Attempt to save
err := store.SaveEvent(event)
if errors.Is(err, aether.ErrVersionConflict) {
// Another writer won - reload and retry if appropriate
var versionErr *aether.VersionConflictError
errors.As(err, &versionErr)
log.Printf("Conflict: actor %s has version %d, attempted %d",
versionErr.ActorID, versionErr.CurrentVersion, versionErr.AttemptedVersion)
}
Error Types
ErrVersionConflict- Sentinel error for version conflicts (use witherrors.Is)VersionConflictError- Detailed error with ActorID, CurrentVersion, and AttemptedVersion
Namespace Isolation
Namespaces provide logical boundaries for events and subscriptions:
// Subscribe to events in a namespace
ch := eventBus.Subscribe("tenant-abc")
// Events are isolated per namespace
eventBus.Publish("tenant-abc", event) // Only tenant-abc subscribers see this
Clustering
Aether handles node discovery, leader election, and shard distribution:
// Create cluster manager
manager := cluster.NewClusterManager(natsConn, nodeID)
// Join cluster
manager.Start()
// Leader election happens automatically
if manager.IsLeader() {
// Coordinate shard assignments
}
Key Patterns
- Events are immutable - Never modify, only append
- Versions are monotonic - Each event must have version > previous for same actor
- Snapshots for performance - Periodically snapshot state to avoid full replay
- Namespaces for isolation - Not multi-tenancy, just logical boundaries
- NATS for everything - Events, pub/sub, clustering all use NATS