Architecture

Nimbus.kt / NimbusKt.kt   → main() + bootstrap wiring
NimbusVersion.kt          → Version read from JAR manifest / gradle.properties
LogRotation.kt            → Rotate latest.log to dated archives
api/                      → Ktor REST + WebSocket server (NimbusApi, auth, routes/)
cluster/                  → NodeManager, ClusterServer (Netty TLS), ClusterWebSocketHandler,
                            RemoteServiceHandle, RemoteFileProxy, TlsHelper, placement/
config/                   → NimbusConfig, GroupConfig, DedicatedConfig, loaders
console/                  → JLine3 REPL, CommandDispatcher, commands/
database/                 → DatabaseManager, MigrationManager, Tables, MetricsCollector, AuditCollector
doctor/                   → Core doctor checks (cluster, config, ports, etc.)
event/                    → EventBus (SharedFlow) + sealed NimbusEvent
group/                    → ServerGroup runtime state, GroupManager
loadbalancer/             → TcpLoadBalancer, BackendHealthManager, strategies
module/                   → ModuleManager (URLClassLoader + ServiceLoader), ModuleContextImpl
proxy/                    → ProxySyncManager (tab list, MOTD, chat, maintenance)
scaling/                  → Core ScalingEngine + reactive scaling rules
service/                  → ServiceManager, ServiceFactory, ProcessHandle, PortAllocator,
                            StateSyncManager, WarmPoolManager, DedicatedServiceManager
setup/                    → First-run SetupWizard
stress/                   → StressTestManager (simulated load)
system/                   → OS-level helpers (memory reader, process inspector)
template/                 → TemplateManager, ConfigPatcher, SoftwareResolver, ServiceDeployer,
                            ModScanner
update/                   → UpdateChecker (GitHub Releases auto-updater)
velocity/                 → VelocityConfigGen (auto-manage proxy server list)

1. Log rotation           → Rotate latest.log to dated archives
2. Setup wizard           → First-run interactive setup (if needed)
3. Config loading         → Parse config/nimbus.toml + config/groups/*.toml
4. API token generation   → Auto-generate if missing
5. Directory creation     → Ensure templates/, services/, logs/, etc. exist
6. Plugin deployment      → Extract nimbus-bridge.jar, nimbus-sdk.jar
7. Component init         → EventBus, ServiceRegistry, PortAllocator,
                            TemplateManager, GroupManager, ProxySyncManager
8. Group loading          → Parse group configs into GroupManager
9. Module loading         → ModuleManager scans modules/*.jar, creates one
                            URLClassLoader per JAR, discovers NimbusModule via
                            module.properties (fallback: ServiceLoader), calls
                            init(). Modules register commands, routes, plugin
                            deployments, event formatters, doctor checks, and
                            migrations during init.
10. Migrations run        → MigrationManager applies every registered migration
                            in ascending version order (core 1–999, modules
                            1000+). Baseline migrations are marked applied
                            without executing.
11. ServiceManager        → Wire up all dependencies, register in ModuleContext
                            via registerService() so late-binding loops in modules
                            (e.g. backup, scaling) can resolve it.
12. Modules enable()      → Called on every loaded module after ServiceManager
                            is registered. State-sync manager for [group.sync]
                            groups is initialized here.
11. ScalingEngine         → Create (but DON'T start yet — deferred until after boot)
12. NimbusApi             → Create (but don't start) Ktor server
13. NodeManager           → If cluster.enabled: init cluster coordination
14. ClusterServer         → If cluster.enabled: separate Ktor server on agent_port for agent WebSocket connections
15. TcpLoadBalancer       → If loadbalancer.enabled: init Layer-4 TCP proxy
                            with BackendHealthManager (health checks, circuit breaker,
                            connection draining, idle timeout, connection limit)
16. Shutdown hook         → Register SIGTERM/SIGINT handler
17. NimbusConsole.init()  → Banner, event listener
18. Api.start()           → Start Ktor HTTP server (+ /cluster WS if cluster enabled)
19. LoadBalancer.start()  → If enabled: start TCP listener on LB port + health check loop
20. VelocityUpdater       → Start periodic update check (first check after 60s)
21. startMinimumInstances → Phased startup:
                            Phase 1: Start proxy groups, wait for READY (120s timeout)
                            Phase 2: Start backend groups
22. ScalingEngine.start() → Start periodic scaling loop (AFTER boot completes)
23. Console.start()       → JLine3 REPL (blocks until shutdown)

// Root scope with SupervisorJob (child failures don't cancel siblings)
val scope = CoroutineScope(SupervisorJob() + Dispatchers.Default)

// EventBus uses the shared scope for event dispatch
val eventBus = EventBus(scope)

// Scaling engine launches a long-running coroutine
val scalingJob = scalingEngine.start()  // Returns a Job

// Emitting an event (suspend function)
eventBus.emit(NimbusEvent.ServiceReady(serviceName = "BedWars-1", groupName = "BedWars"))

// Subscribing to a specific event type
eventBus.on<NimbusEvent.ServiceReady> { event ->
    println("${event.serviceName} is ready!")
}

// Getting the raw SharedFlow
val flow: SharedFlow<NimbusEvent> = eventBus.subscribe()

PREPARING → PREPARED → STARTING → READY → DRAINING → STOPPING → STOPPED
                                    ↓                                 ↑
                                  CRASHED ── (restart?) ─────────────┘

1. Cancel scaling engine job
2. Send graceful shutdown to all agents (if cluster enabled, wait up to 30s)
3. Stop TCP load balancer (if enabled)
4. Stop REST API server
5. Disconnect cluster nodes (if enabled)
6. Stop all local services (via ServiceManager.stopAll()):
   a. Dynamic (game) services first
   b. Static backend (lobby) services
   c. Proxy services last
7. Stop cluster WebSocket server (if enabled)
8. Cancel coroutine scope