coldkeep v0.6.0 — Storage Model Evolution

[franchoy]

coldkeep v0.6 introduces a major evolution of the storage engine, focusing on
a cleaner, safer, and more extensible container model.

This release removes legacy assumptions (whole-container compression/encryption)
and establishes a deterministic append-based storage API that prepares the system
for future block-level features.

Highlights

• New container API with explicit Append / ReadAt / Sync / Close operations
• Removal of whole-container compression and encryption flags
• Deterministic append-only write model (offset-driven, no implicit file pointer)
• Safer container lifecycle: create → append → sync → close → seal
• Container sealing now includes full-file integrity hashing
• Improved concurrency handling with row-level locking and retry strategy
• Container rotation under concurrent workloads (best-effort size enforcement)
• Refactored storage flow to use container abstraction consistently
• Simplified container header format, ready for future block layout

Verification & Testing

• New multi-layer verification system:

• Standard: logical structure and ordering
• Full: container boundaries and metadata consistency
• Deep: chunk-level hash validation

• Extensive stress tests covering:

• Concurrent storage of identical files
• Folder-wide concurrent ingestion
• Retry after aborted file and chunk operations
• Container rotation under high concurrency

• End-to-end restore validation across all scenarios

Notes

• coldkeep remains an experimental research project and is not production ready
• On-disk format may still change before v1.0
• Container max size is enforced on a best-effort basis under concurrency

This version establishes the foundation for upcoming features such as
block-based layouts, per-block compression/encryption, and simulation tools.

---

This discussion was created from the release coldkeep v0.6.0 — Storage Model Evolution.