Apollo Architecture: Current Code and Target
Keep one target, with responsibilities and actual gaps defined by the BigWorld–Apollo mapping.
Conclusions
Apollo prioritizes tower defense and temporary group activities while reserving seamless maps, without claiming complete BigWorld capabilities. Retain six agreed role names and add the discussed cell executor; Cell means a partition.
worldServer overlaps CellAppMgr registration, Space/Cell placement, topology, and migration management. cell overlaps CellApp simulation. Base persistent objects and Cell entities are distinct; database owns storage. loginServer hosts authentication, service directories, and location projections. gateway retains sockets and forwards directly to Base/cell; machined supervises locally.
Space matches current Scene semantics and Instance is temporary gameplay, without a synonymous Scene→Space layer. Initially use one executor and one Cell per Space; reserve multiple Cells, Real/Ghost, continuous views, and complete authority handoff.
Unfinished boundaries
WorldApp, World, and Scene currently combine management/execution; split profile/storage duties in PlayerStore. Base/database prototypes provide remote load/save and a file journal. cell, registration/routing, complete business commits, cross-server associations, dual-version decoders, and three state handoffs remain unfinished. SceneTransfer recreates Avatar and supports explicit scene changes rather than same-Space seamless handoff.
Cross-server play retains original Base/storage affiliation with idempotent durable settlement. Base, cell, and World management transfer persistent objects, simulation, and management records respectively. Gateway survival/replacement have separate TCP outcomes; activity continuity needs complete necessary Instance state.
Dedicated Profiler and OpenTelemetry SDK/OTLP exporter are not integrated. Custom logging/MetricRegistry/span primitives are not integration evidence.
Implementation
Architecture, applications, objects, routing, cross-server play, and updates follow these conclusions. Root todo.md tracks implementation/acceptance; planning is not completion evidence.
