perf(store): evaluate compact, concurrency-safe observation timestamp cache #490

Assessment baseline: fork master b6852ba0876b02426b6200c1c79c7e0ebe3b9389, checked on 16 October 2026. Recheck master and competing work before implementation.

Problem and priority

Medium priority: investigate replacing each observation's sync.Once, cached time.Time and success flag with a compact, concurrency-safe parsed-time cache.

Upstream uses an atomic 64-bit value with explicit sentinel states. Its reported structure-size and heap savings concern its own layout; our StoreObs also contains fork-specific per-observation routing evidence, so the benefit must be measured here.

Fork adaptation

• Preserve the complete ParsedTime contract, including parse order, returned location, precision, zero value and success/failure distinction.
• Define unambiguous encodings for uninitialized, invalid and fallback states. Valid epoch, negative timestamps or sentinel-colliding values must never become invalid or incorrect cached results.
• Preserve offset locations and dates outside the exact Unix-nanosecond range, using a correct fallback when necessary.
• Preserve wireRoute and all fork-specific fields. Update structure-size guards and memory accounting from the actual resulting layout, not upstream's stated 160-byte target.
• Coordinate with observer string sharing and other store-layout/accounting work before implementation. No schema, config, workflow or deployment changes.

Acceptance criteria

• Differential tests compare first and cached calls with the existing parser for RFC3339/RFC3339Nano, SQL timestamps, offsets, empty/invalid input, epoch, negative/zero times, range boundaries and sentinel collisions.
• Concurrent first reads and repeated reads pass race tests. Existing retransmission, observer analytics and route-evidence tests remain green.
• Measure unsafe.Sizeof, effective heap size class, per-observation/accounted bytes, first-parse and cache-hit costs on our supported architecture(s).
• Demonstrate real retained-heap savings on representative fork data and report trade-offs for fallback timestamps. Mutation tests must catch incorrect sentinel handling and lost offset/route semantics.

Source provenance

Repository: Kpa-clawbot/CoreScope
PR: 2196
Title: perf(store): cache an observation's parsed time in 8 bytes
Head: 682c40a1ed84d600e94fa433d97ffb22be7261de
Merge: 8d2ae405e8f2ba4b9bdb09866f563053375a908b
Changed files:
cmd/server/obs_parsed_time_test.go
cmd/server/retransmission_pressure_test.go
cmd/server/store.go