ThemisDB v1.5.0 — Release Aggregation

Release Date: 2026-02-03 (stable) · 2026-03-03 (aggregation updated) Type: Feature Release Previous Version: v1.4.1-dev-alpha Milestone: v1.5.0

🎯 Overview

ThemisDB v1.5.0 is a major feature release that delivers production-ready GPU kernel sandboxing, comprehensive query-result masking, geo CPU/GPU benchmarks, full GeoJSON RFC 7946 support, advanced search components, shard repair/anti-entropy, write-amplification optimisations, and FAISS quantizer improvements. This document serves as the official release aggregation for v1.5.0 and as a template for future release cycles.

📦 Included Pull Requests

The table below lists all Pull Requests assigned to the v1.5.0 milestone, their module ownership, and a short purpose note. PRs are ordered by module.

PR #	Module	Purpose
#3049	geo	CPU vs GPU throughput benchmarks (`bench_geo_cpu_gpu.cpp`): measures `batchIntersects`, `exactIntersects`, `geodesicDistance` across CPU-exact, CPU-approximate, and GPU (CUDA/HIP with CPU fallback) backends
#3050	security	`QueryMaskingPolicy`: dynamic field-level PII masking of query results before they reach the client; role-based bypass for privileged roles
#3051	gpu	`WASMKernelSandbox`: isolated execution environment for untrusted third-party GPU kernel blobs; whitelist + FNV-1a checksum gate, memory ceiling, wall-clock timeout
#1383	acceleration	CUDA ANN + geospatial kernels production-ready — batch `intersects` and distance dispatch via `GpuBatchBackend`
#1384	acceleration	Vulkan compute shader pipeline — initial Vulkan backend wired into `ISpatialComputeBackend`
#1390	acceleration	Cross-backend L2 distance consistency validation — regression test suite confirming CPU/GPU numeric parity
#3420	updates	Update history log: persistent audit trail of who applied which update, from/to version, and timestamp
#3421	updates	Blue/green deployment support (Phase 4): zero-downtime major upgrades via traffic routing and automatic rollback
#3430	updates	fix: repair `hot_reload_engine.h` doc-comment syntax; mark blue/green deployment complete in module ROADMAP
#3422	replication / updates	`CoordinatedUpdateManager`: multi-node replication-safe update sequencing ensuring all replicas apply updates in consensus order
#3424	chimera	CI benchmark baseline: `run_ci_benchmarks` integration tests seed repeatable benchmark baselines for the Chimera adapter layer
#3425	gpu	Multi-node GPU cluster coordination marked production-ready; device management and topology discovery hardened
#3426	performance	Memory pressure monitor (Phase 3): wired into build system with dedicated test suite; triggers GC / eviction under configurable pressure thresholds
#3427	query	Per-query resource limits (`max_rows`, `max_memory_bytes`, `timeout_ms`) exposed via HTTP API and enforced at execution time
#3428	replication	CRDT library expansion: `FLAG_EW` (Enable-Wins Flag) and `FLAG_DW` (Disable-Wins Flag) types added alongside existing set types
#3434	voice	Real-time meeting transcription with action-item extraction: integrates STT pipeline with meeting-session management and NLP action extraction
#3435	performance	PMU cache-miss analysis: CPU performance-counter instrumentation wired into build system and test suite
#3437	performance / ci	Cross-module performance regression detection in CI: automated comparison of benchmark results against stored baselines
#3438	security / updates	HSM-backed `SigningService` for update bundle signing: replaces soft-key signing path with PKCS#11 hardware-backed operations
#3439	voice	Voice module unit-test coverage raised above 80% target
#3442	voice	Dedicated STT latency and TTS generation-speed benchmarks (`bench_voice_stt_tts.cpp`)
#3444	voice	Language detection and auto-locale switching: automatic BCP-47 locale resolution from audio stream characteristics
#3445	rpc	Replace all stub RPC service methods with real RocksDB-backed implementations (GET)
#3446	rpc	Unit tests for RPC GET operation with real RocksDB backend; CMake parse-error fixes
#3447	rpc	RPC PUT/INSERT operations with full transaction support
#3448	rpc	RPC DELETE with cascade logic and referential integrity enforcement
#3449	rpc	RPC Batch Read/Write operations; all remaining stubs removed
#3450	rpc	Integration tests covering all RPC paths (GET, PUT, DELETE, Batch)
#3453	security	Production safety guard for `TimestampAuthority` stub: hard-fails at startup if RFC 3161 real implementation is absent in production builds
#3454	security	Wire PKCS#11 HSM production path end-to-end: fix duplicate source file, harden stub gate, add HSM CI coverage
#3456	security	RFC 3161 TSA compliant API interface/wrapper design
#3457	security	RFC 3161 TSA request/response handling with full `TSAConfig` support (FreeTSA / DigiCert / Sectigo backends)
#3458	security	PKCS#11 C++ wrapper interface (`Pkcs11Provider`) with slot/session/key-object management
#3461	security	PKCS#11 token initialization and configuration (`Pkcs11Initializer`)
#3462	security	HSM production failsafe: hard rejection of stub HSM at runtime in `ENTERPRISE` and `HYPERSCALER` builds
#3463	security / observability	Audit log data-loss protection: `fsync` on write, log rotation with configurable size/age limits, secondary mirror path
#3464	sharding	Hardware migration support: stable `NodeIdentity` persistence across reboots/hardware replacements; safe endpoint replacement without data loss

✨ New Features

GPU Module — WASM Kernel Sandbox (PR #3051)

Files: include/themis/gpu/wasm_kernel_sandbox.h, src/gpu/wasm_kernel_sandbox.cpp, tests/test_gpu_wasm_kernel_sandbox.cpp

WASMKernelSandbox provides an isolated execution environment for GPU kernel blobs submitted by third-party plugins.
Full validation pipeline: whitelist/checksum gate → GPUKernelValidator → sandbox resource enforcement → backend dispatch.
SandboxConfig: configurable memory ceiling (max_memory_bytes), wall-clock timeout (timeout_ms), and host-call toggle.
ExecutionResult: structured result with status, elapsed_ms, and optional error message.
Feature-gated via THEMIS_ENABLE_WASM_SANDBOX; falls back to GPUKernelValidator-only mode when WebAssembly runtime is absent.
Test coverage: tests/test_gpu_wasm_kernel_sandbox.cpp.

Security Module — QueryMaskingPolicy (PR #3050)

Files: include/security/query_masking_policy.h, src/security/query_masking_policy.cpp, tests/test_query_masking_policy.cpp

QueryMaskingPolicy applies field-level masking to JSON query-result objects before they are returned to clients.
Three complementary strategies: (1) field-name hint masking, (2) auto-detect PII via PIIDetector, (3) explicitly declared field paths with "strict" / "partial" / "none" masking modes.
Role-based bypass: roles listed in Config::privileged_roles receive the original unmasked result (default: "admin").
Thread-safe: all public methods are safe to call concurrently.
Integrated into QueryApiHandler and http_server.cpp.
Test coverage: tests/test_query_masking_policy.cpp.

Geo Module — CPU/GPU Benchmarks (PR #3049)

File: benchmarks/bench_geo_cpu_gpu.cpp

Google Benchmark suite measuring batchIntersects, exactIntersects, and geodesicDistance across:
- cpu_exact backend (Boost.Geometry or pure-C++ exact arithmetic)
- cpu_approx backend (S2-geometry approximate)
- gpu_spatial backend (CUDA/HIP with automatic CPU fallback)
Batch sizes: 256, 1024, 4096 point-in-polygon pairs.
Produces baseline CSV for regression tracking under benchmarks/baselines/.
Satisfies Production Readiness Checklist item [x] Performance benchmarks (CPU vs GPU throughput) in src/geo/ROADMAP.md.

Geo Module — Full GeoJSON RFC 7946 Support

Files: include/utils/geo/ewkb.h, src/utils/geo/ewkb.cpp, include/geo/spatial_backend.h

EWKBParser::parseGeoJSON() and toGeoJSON() now handle all 7 RFC 7946 geometry types: Point, MultiPoint, LineString, MultiLineString, Polygon, MultiPolygon, GeometryCollection (including 3D/Z variants).
GeometryCollection parsed recursively up to depth 8 to prevent stack overflow on adversarial input.
computeMBR() and computeCentroid() recurse into nested sub-geometries.
WGS84 coordinate range validation: longitude [-180, 180], latitude [-90, 90]; out-of-range values throw std::runtime_error. Compile with -DTHEMIS_GEO_COMPAT_LAX=1 for one-release migration window.

Geo Module — In-Memory R-Tree Spatial Index

Files: include/geo/geo_rtree.h, src/geo/geo_rtree.cpp

GeoRTree: in-memory R-tree index for GeometryInfo objects enabling sub-linear intersects / contains queries.
With THEMIS_GEO_BOOST_BACKEND: uses boost::geometry::index::rtree with rstar<16> splitting strategy.
Without Boost: falls back to O(n) linear MBR scan — semantically identical, no extra dependency.
bulkLoad(): STR (Sort-Tile-Recursive) packing for 3–5× faster cold-start load versus incremental insert().
20 unit tests covering insert, bulk-load, remove, clear, intersects/contains, memory reporting, move semantics.

Geo Module — ST_UNION and ST_DIFFERENCE

Files: include/geo/spatial_backend.h, include/query/functions/geo_functions.h, tests/geo/test_geo_st_union_difference.cpp, tests/geo/test_aql_st_functions.cpp

ISpatialComputeBackend::stUnion(geom1, geom2) and stDifference(geom1, geom2) added to the spatial-backend interface.
CpuExactBackend: Greiner-Hormann polygon clipping (ACM TOG 1998) with fast-paths for containment, disjoint, and B-inside-A cases.
BoostCpuExactBackend: delegates to boost::geometry::union_ / boost::geometry::difference.
GpuBatchBackend: CPU fallback with audit log and metrics — same pattern as stBuffer.
AQL functions ST_UNION and ST_DIFFERENCE registered and return GeoJSON geometry.
15 unit tests + 7 AQL-level tests.

Search Module — 7 New Search Components

Files: include/search/

Component	Purpose
`QueryExpander`	Synonym expansion, Levenshtein spelling correction, zero-result relaxation
`FuzzyMatcher`	Levenshtein, Soundex, Metaphone, N-gram (Dice) similarity
`FacetedSearch`	Per-field value-count facets, numeric range buckets, drill-down filtering
`SearchAnalytics`	Thread-safe query log; avg/p95/p99 latency, zero-result rate, top-20 queries
`AutocompleteEngine`	Prefix-index + popular-query suggestions, deduplicated and score-ranked
`LearningToRank`	Dot-product linear re-ranker; online pairwise gradient-descent training from click events
`MultiModalSearch`	TEXT/IMAGE/AUDIO/CUSTOM modalities; weighted RRF fusion

Combined test coverage: 162+ tests across 7 new test files.

Sharding Module — Shard Repair / Anti-Entropy Engine

File: include/sharding/shard_repair_engine.h

Background anti-entropy scan: periodic checkDocumentHealth() across all shards; degraded documents automatically queued for recovery.
Repair worker thread drains job queue via RedundancyStrategy::recoverDocument() (RAID-5/6 + Mirror modes).
On-demand triggers with trackable job IDs: triggerRepair(), triggerFullScan(), triggerDocumentRepair().
Per-shard ShardHealthReport: HEALTHY / DEGRADED / FAILED / REBUILDING + scan/repair counters.
Prometheus metrics forwarded via ShardingMetricsHandler.
Admin API endpoints: POST /admin/repair, POST /admin/repair/scan, GET /admin/repair/{job_id}.

Sharding Module — Improved Reed-Solomon Erasure Decoder

Replaced XOR-only parity with Vandermonde matrix systematic codec over GF(2⁸).
Recovers up to parity_shards simultaneously lost chunks — enables true RAID-6 dual-parity recovery.
Both ReedSolomonCoder and CauchyReedSolomonCoder validate missing_indices.size() <= parity_shards.

Security Module — HSM Security Warning System (FIND-002)

Startup warning banner when stub HSM provider is active.
Periodic ERROR-level logging every 5 minutes in insecure configuration.
Prometheus metrics: themis_hsm_insecure_config, themis_hsm_provider_type, hsm_compliance_status.
CLI flag --allow-stub-hsm for development environments.
Addresses FIND-002 from v1.4.1 audit (NIST SP 800-53, ISO 27001, PCI DSS, GDPR).

Index Module — FAISS Quantizer Integration (v1.5.0)

ProductQuantizer: FAISS K-means clustering for 20–30% faster training with SIMD optimisations.
BinaryQuantizer: SIMD-optimised popcount (__builtin_popcount / __popcnt) for 10–15% faster Hamming distance.
ResidualQuantizer: inherits FAISS acceleration from ProductQuantizer stages (30% faster training).
Graceful fallback to custom implementation when FAISS unavailable.
getBackend() method reports actual backend in use at runtime.

Security Module — RFC 3161 Timestamp Authority (v1.5.0)

Full RFC 3161 TSA client with OpenSSL cryptographic operations.
Integration with FreeTSA, DigiCert, Sectigo TSA providers.
eIDAS compliance for qualified electronic timestamps; LTV support for 30-year retention.
CMake option THEMIS_USE_OPENSSL_TSA (default: ON); 10+ compliance tests.

Storage Module — Write-Amplification Optimisation

Default memtable_size_mb 256 MB → 512 MB: ~50% fewer L0 flushes, 30–40% lower write amplification.
Default max_write_buffer_number 3 → 6: fewer write stalls under sustained load.
db_write_buffer_size_mb cap added at 2 GB (was unbounded).
enable_async_io now defaults to true; async_io_readahead_size_mb 64 MB → 128 MB: 2–5× faster sequential scans.

Query Module — v1.5.x Distributed Query Optimizer

DistributedQueryCostModel::getShardRowCount(): dynamic row-count estimates replacing hardcoded 10K constant.
calculatePredicateSelectivity(): histogram-based selectivity estimation with column-specific heuristics.
measureShardLatency(): latency integration hook for future network-aware query planning.
Integration tests: tests/test_optimizer_v1_5_x_integration.cpp.

Index Module — Multi-GPU Vector Indexing API (v2.4 scaffolding)

MultiGPUVectorIndex: logical multi-device API with round-robin/hash/range/balanced partition strategies.
Query fan-out and centralized top-k merge logic.
Current execution uses CPU-based GPUVectorIndex; actual GPU offload planned for v2.5+.
Full API guide: docs/MULTI_GPU_VECTOR_INDEXING.md.

Core — Git-Like Features (PITR, Merge, Branch)

SnapshotManager re-enabled: named snapshots via 5 REST endpoints.
PITR API handler: restore by sequence, tag, or timestamp; dry-run preview.
MergeEngine integrated: 3-way merge with OURS / THEIRS / MANUAL / FAST_FORWARD resolution.
BranchManager: non-fast-forward merges via MergeEngine; conflict detection.

✨ Additional Features (Post-Initial Aggregation)

The following features were merged after the initial v1.5.0 feature set was defined. They are included in v1.5.0.

Updates Module — History Log, Blue/Green Deployment, CoordinatedUpdateManager (PRs #3420, #3421, #3422, #3430, #3433)

Files: include/updates/update_history_log.h, include/updates/blue_green_manager.h, include/updates/coordinated_update_manager.h

UpdateHistoryLog (#3420): Persistent audit trail of all update operations — records operator identity, source/target version, timestamp, and outcome. Retained across restarts.
Blue/Green Deployment (#3421, #3430): Zero-downtime major upgrades. Traffic is shifted to a parallel "green" environment while "blue" serves live traffic; automatic rollback on failure. Roadmap item [x] blue-green deployment support marked complete.
CoordinatedUpdateManager (#3422): Multi-node replication-safe update sequencing using Raft-compatible consensus ordering. Ensures all replicas apply updates in the same order and prevents split-brain during rolling upgrades.
Schema migration regression tests (#3433): validateMigration test suite added; Stubs:1 path in schema migration testing framework resolved.

Voice Module — Language Detection, Meeting Transcription, Benchmarks (PRs #3434, #3439, #3442, #3444)

Files: include/voice/language_detector.h, include/voice/meeting_transcriber.h, benchmarks/bench_voice_stt_tts.cpp

Language Detection (#3444): Automatic BCP-47 locale resolution from audio stream characteristics; switches transcription model and tokenizer locale on-the-fly. Supports 40+ languages.
Real-Time Meeting Transcription (#3434): Speaker-diarized transcription with action-item extraction using NLP rule engine. Emits structured MeetingSession records with participant turn boundaries and extracted tasks.
Voice Unit-Test Coverage (#3439): Test coverage for voice module raised above 80% production threshold.
STT/TTS Benchmarks (#3442): Dedicated Google Benchmark suite measuring STT latency (p50/p95/p99) and TTS generation speed across engine configurations.

GPU Module — Multi-Node Cluster Coordination (PR #3425)

Multi-node GPU cluster coordination declared production-ready: device management, topology discovery, and health-check probes hardened; memory-safety issues from pre-production phase resolved.

Performance Module — Memory Pressure Monitor & PMU Cache-Miss Analysis (PRs #3426, #3435, #3437)

Files: include/performance/memory_pressure_monitor.h, include/performance/pmu_cache_miss_analyzer.h

Memory Pressure Monitor (#3426): Monitors process RSS and system-level free-memory headroom; triggers configurable callbacks (cache eviction, GC, reject-new-writes) at configurable pressure thresholds. Wired into build system with test suite.
PMU Cache-Miss Analysis (#3435): CPU performance-counter instrumentation using perf_event_open (Linux) / VTune-compatible counters; exposes L1/L2/LLC miss rates per query execution. Wired into build and test pipeline.
Cross-Module Performance Regression Detection in CI (#3437): Automated benchmark comparison stage; builds store result JSON under benchmarks/baselines/; CI job fails if any metric regresses beyond configurable tolerance.

Query Module — Per-Query Resource Limits (PR #3427)

Files: include/query/query_resource_limiter.h, src/server/http_server.cpp

QueryResourceLimiter enforces per-request ceilings: max_rows (result set size), max_memory_bytes (working-set allocation), timeout_ms (wall-clock deadline).
Limits configurable per-query via HTTP request header X-Themis-Resource-Limits (JSON) or globally via config/core/config.yaml.
Returns HTTP 429 (resource limit exceeded) or 408 (timeout) with structured error body.

Replication Module — CRDT FLAG Types (PR #3428)

Files: include/replication/crdt_flag.h

FLAG_EW (Enable-Wins Flag): concurrent enable and disable → enable wins; semantically correct for permission grants.
FLAG_DW (Disable-Wins Flag): concurrent enable and disable → disable wins; semantically correct for feature kill-switches and circuit breakers.
Both types integrate with existing CRDTManager merge/serialisation pipeline.

Security Module — Full PKCS#11 HSM, RFC 3161 TSA, Audit Log Hardening (PRs #3453–#3463)

Files: include/security/pkcs11_provider.h, include/security/pkcs11_initializer.h, include/security/timestamp_authority.h, include/security/audit_log.h

PKCS#11 C++ Wrapper (#3458): Pkcs11Provider encapsulates slot enumeration, session lifecycle, key-object management, and CKM mechanism dispatch. Tested against SoftHSMv2 and Thales Luna.
PKCS#11 Token Initialization (#3461): Pkcs11Initializer handles token provisioning, PIN management, and key-object lifecycle for first-time HSM setup.
HSM Production Failsafe (#3462): ENTERPRISE and HYPERSCALER builds hard-fail at startup when stub HSM provider is detected; prevents accidental production deployments without hardware key storage.
PKCS#11 Production Path (#3454): Duplicate source file removed; stub guard hardened; HSM CI coverage extended to include PKCS#11 signing/verification round-trips against SoftHSMv2.
RFC 3161 TSA Interface (#3456): Compliant API interface for TimestampAuthority; abstracts FreeTSA, DigiCert, and Sectigo backends behind a uniform requestTimestamp() / verifyTimestamp() contract.
RFC 3161 TSA Implementation (#3457): Full request/response serialisation and verification using OpenSSL ASN.1 primitives. TSAConfig selects backend, retry policy, and certificate pinning.
TSA Production Guard (#3453): TimestampAuthority stub hard-fails at startup in production builds; development builds log a suppressible warning.
Audit Log Data-Loss Protection (#3463): Writes are fsync-ed immediately; log rotation triggered at configurable size/age limits; secondary mirror path for dual-write redundancy. Addresses potential data loss under crash scenarios.

Sharding Module — Hardware Migration Support (PR #3464)

Files: include/sharding/node_identity.h, src/sharding/node_identity.cpp

NodeIdentity persisted to stable storage (LMDB sidecar); survives reboots and hardware component replacements without triggering unnecessary shard rebalancing.
Safe endpoint replacement API: NodeIdentity::updateEndpoint() atomically updates the network address and notifies the cluster without data movement.
Cluster controller reconciliation: nodes with matching NodeIdentity but changed endpoint are recognised as the same logical node; no false NODE_LOST events.

Chimera Module — CI Benchmark Baseline (PR #3424)

Files: tests/chimera/test_chimera_ci_benchmarks.cpp, benchmarks/baselines/chimera_baseline.json

run_ci_benchmarks integration-test target seeds a repeatable baseline under benchmarks/baselines/chimera_baseline.json.
Benchmark results are committed and compared in CI to detect adapter-layer performance regressions.

Security / Updates — HSM-Backed SigningService (PR #3438)

Files: include/security/hsm_signing_service.h, src/security/hsm_signing_service.cpp

HsmSigningService replaces the soft-key signing path in update bundle signing with PKCS#11 hardware-backed operations.
Integrates with Pkcs11Provider; supports RSA-PSS and ECDSA signing mechanisms.
Update bundles now carry a hardware-attested signature; rejected by the update verifier if signed with a soft key in production builds.

RPC Layer — Full Production Implementation (PRs #3445–#3450)

Files: src/rpc/rpc_service.cpp, tests/rpc/

All stub RPC service methods replaced with real RocksDB-backed implementations:
- GET (#3445, #3446): key lookup with optional consistency-level parameter.
- PUT / INSERT (#3447): transactional write with conflict detection.
- DELETE (#3448): cascade delete with referential integrity checks across column families.
- Batch Read/Write (#3449): atomic multi-key operations with all-or-nothing semantics.
Full integration-test coverage across all RPC paths (#3450).

⚠️ Breaking Changes

Area	Change	Migration
GeoJSON parsing	Coordinates outside WGS84 range now throw `std::runtime_error`	Compile with `-DTHEMIS_GEO_COMPAT_LAX=1` for one-release window
GeoJSON unknown types	Unknown geometry types throw instead of returning empty geometry	Update clients to send valid RFC 7946 geometry types
GPU Vector Index stubs	`gpu_vector_index_cuda.cpp`, `_vulkan.cpp`, `_hip.cpp` removed from `src/index/`	See `docs/FUTURE_GPU_SUPPORT.md`; use `GPUVectorIndex` (CPU-SIMD)

🔒 Security

RFC 3161 TSA enables legally binding digital signatures (eIDAS).
QueryMaskingPolicy prevents PII exposure in query results.
WASMKernelSandbox isolates untrusted third-party GPU kernel blobs.
HSM security warning system prevents accidental production deployment with stub provider.

✅ QA Criteria for Inclusion in v1.5.0

The following criteria must be satisfied for a PR to be included in this release:

Criterion	Requirement
Feature tests completed	All new code paths covered by unit and/or integration tests; no failing tests in CI
Compatibility with v1.4.x	No breaking changes beyond those listed in the Breaking Changes table; all existing API contracts maintained
Documentation present	Header-level Doxygen/comment blocks; release notes entry in CHANGELOG.md; module ROADMAP.md updated
Migration guide available	For every breaking change a migration note exists in CHANGELOG.md and this document
Security review	Security-sensitive features (crypto, masking, sandboxing) reviewed and signed off
Performance baseline	Performance-critical features benchmarked and baseline stored under `benchmarks/baselines/`
Roadmap item linkage	Every PR is traced to at least one roadmap item (`ROADMAP.md` or module `ROADMAP.md`)

QA Status per PR Group

PR(s)	Module	Tests	Docs	Security Review	Performance Baseline	QA Flag
#3049	geo	✅	✅	✅	✅ baselines/	✅
#3050	security	✅	✅	✅ signed off	—	✅
#3051	gpu	✅	✅	✅ signed off	—	✅
#1383	acceleration	⚠️ PR open	⚠️ partial	⚠️ pending	—	⚠️ Needs QA sign-off
#1384	acceleration	⚠️ PR open	⚠️ partial	⚠️ pending	—	⚠️ Needs QA sign-off
#1390	acceleration	⚠️ PR open	✅	⚠️ pending	—	⚠️ Needs QA sign-off
#3420	updates	✅	✅	—	—	✅
#3421, #3430	updates	✅	✅	—	—	✅
#3422	replication/updates	✅	✅	—	—	✅
#3424	chimera	✅	✅	—	✅ baseline seeded	✅
#3425	gpu	✅	✅	✅	—	✅
#3426	performance	✅	✅	—	—	✅
#3427	query	✅	✅	—	—	✅
#3428	replication	✅	✅	—	—	✅
#3434, #3439	voice	✅ >80%	✅	⚠️ audio security audit WIP (#3443)	—	⚠️ Audio security audit pending
#3435, #3437	performance	✅	✅	—	✅	✅
#3438	security/updates	✅	✅	✅	—	✅
#3442	voice	✅	✅	—	✅ STT/TTS baselines	✅
#3444	voice	✅	✅	—	—	✅
#3445–#3450	rpc	✅ integration	✅	—	—	✅
#3453–#3462	security (HSM/TSA)	✅	✅	✅	—	✅
#3463	security/observability	✅	✅	✅	—	✅
#3464	sharding	✅	✅	—	—	✅

📋 Roadmap Traceability

Roadmap Item	Status	PR / Issue
CUDA ANN + geospatial kernels production-ready	[P]	#1383
Vulkan compute shader pipeline	[P]	#1384
Cross-backend L2 distance consistency validation	[P]	#1390
Geo CPU/GPU throughput benchmarks	[x]	#3049
QueryMaskingPolicy for PII field masking	[x]	#3050
WASM GPU Kernel Sandbox for third-party kernels	[x]	#3051
Full GeoJSON RFC 7946 parsing	[x]	Unreleased → v1.5.0
In-memory R-tree spatial index	[x]	Unreleased → v1.5.0
ST_UNION / ST_DIFFERENCE geometry operations	[x]	Unreleased → v1.5.0
Shard repair / anti-entropy engine	[x]	Unreleased → v1.5.0
Write-amplification optimisation (RocksDB)	[x]	Unreleased → v1.5.0
RFC 3161 Timestamp Authority	[x]	#3453, #3456, #3457
FAISS Quantizer Integration	[x]	v1.5.0
Update history log (who/when/from/to version)	[x]	#3420
Blue/green deployment support for zero-downtime upgrades	[x]	#3421, #3430
Multi-node replication-safe update sequencing	[x]	#3422
Multi-node GPU cluster coordination production-ready	[x]	#3425
Memory pressure monitor (Phase 3)	[x]	#3426
Per-query resource limits (max_rows, max_memory_bytes, timeout_ms)	[x]	#3427
CRDT FLAG_EW and FLAG_DW types	[x]	#3428
Real-time meeting transcription with action-item extraction	[x]	#3434
PMU cache-miss analysis and cross-module regression CI	[x]	#3435, #3437
HSM-backed SigningService for update bundles	[x]	#3438
Voice language detection and auto-locale switching	[x]	#3444
RPC layer: all stubs replaced with real database-backed implementations	[x]	#3445–#3450
PKCS#11 HSM production path (wrapper, init, failsafe, CI)	[x]	#3454, #3458, #3461, #3462
Audit log data-loss protection (fsync, rotation, mirror)	[x]	#3463
Sharding hardware migration support / NodeIdentity persistence	[x]	#3464
Chimera CI benchmark baseline	[x]	#3424

🚩 PRs / Features Requiring Further Clarification or QA

The following items are included in the v1.5.0 PR aggregation but require additional action before the final release sign-off:

Item	Issue	Action Required
#1383 CUDA ANN + geospatial kernels	PR still open	Merge and run full GPU regression suite; confirm acceptance tests pass
#1384 Vulkan compute shader pipeline	PR still open	Merge; shader integrity verification (#1384 roadmap item) to be completed
#1390 Cross-backend L2 distance validation	PR still open	Merge; confirm numeric parity test baseline committed
Voice audio security audit	#3443 WIP	Complete audio pipeline security audit before final release
Voice SIP/WebRTC integration	#3431 WIP	Descoped from v1.5.0 unless completed; target v1.6.0
Build modularisation	#3429 WIP	Descoped from v1.5.0 unless completed; target v1.6.0
Chimera adapter capabilities matrix	#3436 WIP	Documentation-only; low risk; can follow in patch
Pre-flight health checks update	#3441 WIP	Confirm coverage; no blocking dependency on release

🔗 Related Documentation

RELEASE_TARGET_v1.5.0.md — Pre-release target definition: deliverables, scope, completion criteria, QA/performance requirements, and out-of-scope items
CHANGELOG.md — Full change log
ROADMAP.md — Top-level project roadmap
src/geo/ROADMAP.md — Geo module roadmap
src/gpu/ROADMAP.md — GPU module roadmap
src/security/ROADMAP.md — Security module roadmap
docs/FUTURE_GPU_SUPPORT.md — GPU support v2.x roadmap
docs/GPU_MASTER_TRACKING.md — GPU implementation tracking

This document was produced as part of the v1.5.0 Release Aggregation (Issue: #3070) and serves as a template for future release cycles. Last updated: 2026-03-03.

FilesExpand file tree

RELEASE_NOTES_v1.5.0.md

Latest commit

History