Storage Backends

Implemented Backend: fs_split_v1

Layout:

• large chunk -> directory
• regular chunk -> .chk file
• per-regular-chunk WAL -> .wal file

Strengths

• simple implementation
• natural per-chunk isolation
• straightforward per-chunk locking
• corruption blast radius is localized to one regular chunk
• incremental recovery via per-chunk WAL replay

Weaknesses

• many files/inodes for very large worlds
• metadata overhead from filesystem operations
• directory scaling can become a factor for extremely large datasets

Experimental Layout Prototype (Benchmark-Only)

• fs_region_v1 exists as an experimental benchmark prototype.
• It is not the server default.
• It is used only for A/B layout measurements and decision support.
• It is not part of the stable 1.x support surface.
• It currently has no migration path and no compatibility guarantee.
• Latest measured A/B snapshot and gate decision are in PERFORMANCE_LAYOUT_AB.md.

Stable Backend Scope

For the stable 1.x line, the supported backend scope is fs_split_v1. This keeps stabilization focused on correctness, durability behavior, runtime scalability, and benchmark transparency.

No additional production-ready backend is part of the current stable support surface.