- hardening and refinement of storage runtime behavior
- deterministic reliability improvements for durability/recovery paths
- stronger benchmark-backed guardrails for storage hot paths
- storage contracts remain backward compatible within major release line.
- persistence and recovery outcomes remain explicit and deterministic.
- degraded backend/tiering paths remain observable and non-silent.
- maintenance and replay behavior remains bounded and diagnosable.
| Interface | Requirement |
|---|---|
| persistence interfaces | deterministic lifecycle and durable write behavior |
| versioning interfaces | stable MVCC and replay semantics |
| recovery interfaces | explicit backup/PITR result contracts |
| maintenance interfaces | bounded compaction/pruning/index-maintenance behavior |
- tighten parity between replay, compaction, and recovery diagnostics.
- standardize incident taxonomy for durability and maintenance classes.
- expand resilience tests for prolonged mixed write and recovery traffic.
- broaden benchmark depth for allocator and mount-latency scenarios.
- unit and integration suites for persistence, recovery, and maintenance paths.
- regressions for WAL replay, PITR faults, and tiered/blob edge failures.
- deterministic stress runs for sustained mixed read/write + maintenance load.
- release-profile benchmark runs for mapped storage targets.
- storage hot paths remain inside regression budgets.
- write/replay/recovery-sensitive operations remain stable at p95/p99 envelopes.
- mapped benchmark manifests reach no-missing-case status for release gating.
- maintain strict bounded behavior for durability and recovery transitions.
- preserve explicit failure signaling for storage and maintenance faults.
- enforce predictable degradation under storage pressure and backend faults.
- keep diagnostics actionable for production storage incidents.