e411fde1a9
Sprint: SPRINT_20251229_049_BE_csproj_audit_maint_tests Tasks: AUDIT-0001 through AUDIT-0147 APPLY tasks (approved decisions 1-9) Changes: - Set TreatWarningsAsErrors=true for all production .NET projects - Fixed nullable warnings in Scanner.EntryTrace, Scanner.Evidence, Scheduler.Worker, Concelier connectors, and other modules - Injected TimeProvider/IGuidProvider for deterministic time/ID generation - Added path traversal validation in AirGap.Bundle - Fixed NULL handling in various cursor classes - Third-party GostCryptography retains TreatWarningsAsErrors=false (preserves original) - Test projects excluded per user decision (rejected decision 10) Note: All 17 ACSC connector tests pass after snapshot fixture sync
Link-Not-Merge Bench
Synthetic workload that measures advisory observation ingestion and linkset correlation throughput for the Link-Not-Merge program.
Scenarios
config.json defines three scenarios that vary observation volume, alias density, and correlation fan-out. Each scenario captures:
- Total latency (ingest + correlation) and p95/max percentiles
- Insert latency against an ephemeral MongoDB instance
- Correlator-only latency, tracking fan-out costs
- Observation and Mongo insert throughput (ops/sec)
- Peak managed heap allocations
Running locally
dotnet run \
--project src/Bench/StellaOps.Bench/LinkNotMerge/StellaOps.Bench.LinkNotMerge/StellaOps.Bench.LinkNotMerge.csproj \
-- \
--csv out/linknotmerge-bench.csv \
--json out/linknotmerge-bench.json \
--prometheus out/linknotmerge-bench.prom
The benchmark exits non-zero if latency exceeds configured thresholds, throughput falls below the floor, Mongo insert throughput regresses, allocations exceed the ceiling, or regression ratios breach the baseline.