fix: flush all native providers on shutdown to prevent data loss

Shutdown/close/flush now properly flushes all 4 components in parallel:
metadataIndex, graphIndex, HNSW dirty nodes, and storage counts. Previously
only counts were flushed, causing native provider data loss on restart.

Also:
- Wire roaring, msgpack, entityIdMapper provider consumption from plugins
- Fix allOf filter O(n²) intersection → O(n) Set-based
- Fix ne/exists negation filter to use Set-based exclusion
- Add setMsgpackImplementation() swap in SSTable for native msgpack
- Add setRoaringImplementation() swap for native CRoaring bitmaps
- Add getAllIntIds() to EntityIdMapper for bitmap operations
- Remove TypeAwareHNSWIndex from default index creation path
- Export memory detection utilities from internals
- Clean up 26 permanently-skipped dead tests
This commit is contained in:
David Snelling 2026-02-01 16:23:49 -08:00
parent b87426409d
commit 773c5171c3
24 changed files with 297 additions and 1268 deletions

View file

@ -7,7 +7,8 @@
import { v4 as uuidv4 } from './universal/uuid.js'
import { HNSWIndex } from './hnsw/hnswIndex.js'
import { TypeAwareHNSWIndex } from './hnsw/typeAwareHNSWIndex.js'
// TypeAwareHNSWIndex removed from default path — single unified HNSWIndex is faster
// for the 99% of queries that don't filter by type (avoids searching 42 separate graphs)
import { createStorage } from './storage/storageFactory.js'
import { BaseStorage } from './storage/baseStorage.js'
import { StorageAdapter, Vector, DistanceFunction, EmbeddingFunction, GraphVerb } from './coreTypes.js'
@ -48,14 +49,12 @@ import {
import {
SaveNounMetadataOperation,
SaveNounOperation,
AddToTypeAwareHNSWOperation,
AddToHNSWOperation,
AddToMetadataIndexOperation,
SaveVerbMetadataOperation,
SaveVerbOperation,
AddToGraphIndexOperation,
RemoveFromHNSWOperation,
RemoveFromTypeAwareHNSWOperation,
RemoveFromMetadataIndexOperation,
RemoveFromGraphIndexOperation,
UpdateNounMetadataOperation,
@ -135,8 +134,7 @@ export class Brainy<T = any> implements BrainyInterface<T> {
private static instances: Brainy[] = []
// Core components
private index!: HNSWIndex | TypeAwareHNSWIndex
private indexIsTypeAware = false // true when index supports addItem(item, type) / search(v, k, type)
private index!: HNSWIndex
private storage!: BaseStorage
private metadataIndex!: MetadataIndexManager
private graphIndex!: GraphAdjacencyIndex
@ -275,7 +273,7 @@ export class Brainy<T = any> implements BrainyInterface<T> {
try {
// Auto-detect and activate plugins BEFORE storage setup
// so plugin-provided storage factories (e.g., mmap-filesystem) are available
// so plugin-provided storage factories (e.g., filesystem override from cortex) are available
await this.loadPlugins()
// Setup and initialize storage (checks plugin storage factories first)
@ -301,6 +299,20 @@ export class Brainy<T = any> implements BrainyInterface<T> {
setGlobalCache(cacheProvider)
}
// Provider: roaring bitmaps (native CRoaring replacement for WASM)
const roaringProvider = this.pluginRegistry.getProvider<any>('roaring')
if (roaringProvider) {
const { setRoaringImplementation } = await import('./utils/roaring/index.js')
setRoaringImplementation(roaringProvider)
}
// Provider: msgpack (native replacement for JS @msgpack/msgpack)
const msgpackProvider = this.pluginRegistry.getProvider<any>('msgpack')
if (msgpackProvider) {
const { setMsgpackImplementation } = await import('./graph/lsm/SSTable.js')
setMsgpackImplementation(msgpackProvider)
}
// Provider: distance function (resolve BEFORE setupIndex — index uses this.distance)
const nativeDistance = this.pluginRegistry.getProvider<DistanceFunction>('distance')
if (nativeDistance) {
@ -309,14 +321,18 @@ export class Brainy<T = any> implements BrainyInterface<T> {
// Provider: HNSW index factory (plugin or JS fallback)
this.index = this.createIndex()
this.indexIsTypeAware = this.index instanceof TypeAwareHNSWIndex
|| typeof (this.index as any).getIndexForType === 'function'
// Provider: metadata index factory
const metadataFactory = this.pluginRegistry.getProvider<(storage: StorageAdapter) => any>('metadataIndex')
this.metadataIndex = metadataFactory
? metadataFactory(this.storage)
: new MetadataIndexManager(this.storage)
if (metadataFactory) {
this.metadataIndex = metadataFactory(this.storage)
} else {
// JS fallback — inject native EntityIdMapper if cortex provides one
const entityIdMapperFactory = this.pluginRegistry.getProvider<(storage: StorageAdapter) => any>('entityIdMapper')
this.metadataIndex = new MetadataIndexManager(this.storage, {}, {
entityIdMapper: entityIdMapperFactory ? entityIdMapperFactory(this.storage) : undefined,
})
}
// Provider: graph index factory
const graphFactory = this.pluginRegistry.getProvider<(storage: StorageAdapter) => any>('graphIndex')
@ -440,21 +456,42 @@ export class Brainy<T = any> implements BrainyInterface<T> {
*/
private registerShutdownHooks(): void {
const flushOnShutdown = async () => {
console.log('⚠️ Shutdown signal received - flushing pending counts...')
console.log('Shutdown signal received - flushing pending data...')
try {
// Flush counts for all Brainy instances
let flushedCount = 0
for (const instance of Brainy.instances) {
if (instance.storage && typeof (instance.storage as any).flushCounts === 'function') {
await (instance.storage as any).flushCounts()
if (instance.initialized) {
// Full flush: counts + metadata index + graph index + HNSW dirty nodes
await Promise.all([
(async () => {
if (instance.storage && typeof (instance.storage as any).flushCounts === 'function') {
await (instance.storage as any).flushCounts()
}
})(),
(async () => {
if (instance.metadataIndex && typeof instance.metadataIndex.flush === 'function') {
await instance.metadataIndex.flush()
}
})(),
(async () => {
if (instance.graphIndex && typeof instance.graphIndex.flush === 'function') {
await instance.graphIndex.flush()
}
})(),
(async () => {
if (instance.index && typeof (instance.index as any).flush === 'function') {
await (instance.index as any).flush()
}
})()
])
flushedCount++
}
}
if (flushedCount > 0) {
console.log(`✅ Counts flushed successfully (${flushedCount} instance${flushedCount > 1 ? 's' : ''})`)
console.log(`Flushed successfully (${flushedCount} instance${flushedCount > 1 ? 's' : ''})`)
}
} catch (error) {
console.error('❌ Failed to flush counts on shutdown:', error)
console.error('Failed to flush on shutdown:', error)
}
}
@ -755,20 +792,9 @@ export class Brainy<T = any> implements BrainyInterface<T> {
)
// Operation 3: Add to HNSW index (after entity saved)
if (this.indexIsTypeAware) {
tx.addOperation(
new AddToTypeAwareHNSWOperation(
this.index as any,
id,
vector,
params.type as any
)
)
} else {
tx.addOperation(
new AddToHNSWOperation(this.index as any, id, vector)
)
}
tx.addOperation(
new AddToHNSWOperation(this.index as any, id, vector)
)
// Operation 4: Add to metadata index
tx.addOperation(
@ -1182,31 +1208,12 @@ export class Brainy<T = any> implements BrainyInterface<T> {
// Operation 3-4: Update HNSW index (remove and re-add if reindexing needed)
if (needsReindexing) {
if (this.indexIsTypeAware) {
tx.addOperation(
new RemoveFromTypeAwareHNSWOperation(
this.index as any,
params.id,
existing.vector,
existing.type as any
)
)
tx.addOperation(
new AddToTypeAwareHNSWOperation(
this.index as any,
params.id,
vector,
newType as any
)
)
} else {
tx.addOperation(
new RemoveFromHNSWOperation(this.index as any, params.id, existing.vector)
)
tx.addOperation(
new AddToHNSWOperation(this.index as any, params.id, vector)
)
}
tx.addOperation(
new RemoveFromHNSWOperation(this.index as any, params.id, existing.vector)
)
tx.addOperation(
new AddToHNSWOperation(this.index as any, params.id, vector)
)
}
// Operation 5-6: Update metadata index (remove old, add new)
@ -1264,21 +1271,10 @@ export class Brainy<T = any> implements BrainyInterface<T> {
// Execute atomically with transaction system
await this.transactionManager.executeTransaction(async (tx) => {
// Operation 1: Remove from vector index
if (noun && metadata) {
if (this.indexIsTypeAware && metadata.noun) {
tx.addOperation(
new RemoveFromTypeAwareHNSWOperation(
this.index as any,
id,
noun.vector,
metadata.noun as any
)
)
} else {
tx.addOperation(
new RemoveFromHNSWOperation(this.index as any, id, noun.vector)
)
}
if (noun) {
tx.addOperation(
new RemoveFromHNSWOperation(this.index as any, id, noun.vector)
)
}
// Operation 2: Remove from metadata index
@ -1975,6 +1971,42 @@ export class Brainy<T = any> implements BrainyInterface<T> {
return results
}
// Metadata-first optimization: pre-resolve filter IDs before vector search.
// This enables HNSW to search only within matching candidates instead of
// doing expensive post-filtering. Native HNSW uses searchWithCandidates (Rust
// bitmap), JS HNSW converts to a Set-based filter function.
let preResolvedMetadataIds: string[] | null = null
let preResolvedFilter: any = null
if (params.where || params.type || params.service || params.excludeVFS) {
preResolvedFilter = {}
if (params.where) Object.assign(preResolvedFilter, params.where)
if (params.service) preResolvedFilter.service = params.service
if (params.excludeVFS === true) {
preResolvedFilter.vfsType = { exists: false }
preResolvedFilter.isVFSEntity = { ne: true }
}
if (params.type) {
const types = Array.isArray(params.type) ? params.type : [params.type]
if (types.length === 1) {
preResolvedFilter.noun = types[0]
} else {
preResolvedFilter = {
anyOf: types.map(type => ({
noun: type,
...preResolvedFilter
}))
}
}
}
preResolvedMetadataIds = await this.metadataIndex.getIdsForFilter(preResolvedFilter)
// Short-circuit: if metadata filter matches nothing, skip expensive vector search
if (preResolvedMetadataIds.length === 0) {
return []
}
}
// Zero-Config Hybrid Search
// Determine search mode: auto (default) combines text + semantic for query searches
const searchMode = params.searchMode || 'auto'
@ -1986,14 +2018,14 @@ export class Brainy<T = any> implements BrainyInterface<T> {
}
// Handle semantic-only query (user explicitly wants vector search)
else if ((searchMode === 'semantic' || searchMode === 'vector') && (params.query || params.vector)) {
results = await this.executeVectorSearch(params)
results = await this.executeVectorSearch(params, preResolvedMetadataIds ?? undefined)
}
// Handle explicit hybrid or auto mode with query
else if ((searchMode === 'auto' || searchMode === 'hybrid') && params.query && params.query.trim() !== '' && !params.vector) {
// Zero-config hybrid: combine text + semantic search with RRF fusion
const [textResults, semanticResults] = await Promise.all([
this.executeTextSearch(params.query, limit * 2),
this.executeVectorSearch(params)
this.executeVectorSearch(params, preResolvedMetadataIds ?? undefined)
])
// Use user-specified alpha or auto-detect based on query length
@ -2007,7 +2039,7 @@ export class Brainy<T = any> implements BrainyInterface<T> {
}
// Handle direct vector search (no query text) - no hybrid needed
else if (params.vector && !params.query) {
results = await this.executeVectorSearch(params)
results = await this.executeVectorSearch(params, preResolvedMetadataIds ?? undefined)
}
// Handle proximity search
else if (params.near) {
@ -2032,42 +2064,16 @@ export class Brainy<T = any> implements BrainyInterface<T> {
results = Array.from(uniqueResults.values())
}
// Apply O(log n) metadata filtering using core MetadataIndexManager
if (params.where || params.type || params.service || params.excludeVFS) {
// Build filter object for metadata index
let filter: any = {}
// Apply metadata filtering using pre-resolved IDs (metadata-first optimization).
// When vector search was performed, HNSW already filtered by candidateIds —
// this block handles pagination, entity loading, and metadata-only queries.
if (preResolvedMetadataIds && preResolvedFilter) {
const filteredIds = preResolvedMetadataIds
// Base filter from where and service
if (params.where) Object.assign(filter, params.where)
if (params.service) filter.service = params.service
// ExcludeVFS helper - exclude VFS infrastructure entities
// VFS files/folders have vfsType set, extracted entities do NOT
if (params.excludeVFS === true) {
filter.vfsType = { exists: false }
filter.isVFSEntity = { ne: true }
}
if (params.type) {
const types = Array.isArray(params.type) ? params.type : [params.type]
if (types.length === 1) {
filter.noun = types[0]
} else {
// For multiple types, create separate filter for each type with all conditions
filter = {
anyOf: types.map(type => ({
noun: type,
...filter
}))
}
}
}
const filteredIds = await this.metadataIndex.getIdsForFilter(filter)
// CRITICAL FIX: Handle both cases properly
if (results.length > 0) {
// OPTIMIZED: Filter existing results (from vector search) efficiently
// Filter results by pre-resolved metadata IDs.
// With metadata-first HNSW, most results already match — this is a safety net
// for text search results and proximity results that weren't pre-filtered.
const filteredIdSet = new Set(filteredIds)
results = results.filter((r) => filteredIdSet.has(r.id))
@ -2120,7 +2126,7 @@ export class Brainy<T = any> implements BrainyInterface<T> {
// Apply sorting if requested for metadata-only queries
if (params.orderBy) {
const sortedIds = await this.metadataIndex.getSortedIdsForFilter(
filter,
preResolvedFilter,
params.orderBy,
params.order || 'asc'
)
@ -2529,21 +2535,10 @@ export class Brainy<T = any> implements BrainyInterface<T> {
const allVerbs = [...verbs, ...targetVerbs]
// Add delete operations to transaction
if (noun && metadata) {
if (this.indexIsTypeAware && metadata.noun) {
tx.addOperation(
new RemoveFromTypeAwareHNSWOperation(
this.index as any,
id,
noun.vector,
metadata.noun as any
)
)
} else {
tx.addOperation(
new RemoveFromHNSWOperation(this.index as any, id, noun.vector)
)
}
if (noun) {
tx.addOperation(
new RemoveFromHNSWOperation(this.index as any, id, noun.vector)
)
}
if (metadata) {
@ -2948,9 +2943,8 @@ export class Brainy<T = any> implements BrainyInterface<T> {
// Create HNSW index (uses plugin factory when available)
clone.index = this.createIndex()
clone.indexIsTypeAware = this.indexIsTypeAware
// Enable COW (handle both HNSWIndex and TypeAwareHNSWIndex)
// Enable COW
if ('enableCOW' in clone.index && typeof clone.index.enableCOW === 'function') {
(clone.index as any).enableCOW(this.index)
}
@ -4188,7 +4182,7 @@ export class Brainy<T = any> implements BrainyInterface<T> {
* 1. Storage counts (entity/verb counts by type)
* 2. Metadata index (field indexes + EntityIdMapper)
* 3. Graph adjacency index (relationship cache)
* 4. HNSW vector index (no flush needed - saves directly)
* 4. HNSW vector index (deferred dirty nodes)
*
* @example
* // Flush after bulk operations
@ -4204,7 +4198,7 @@ export class Brainy<T = any> implements BrainyInterface<T> {
async flush(): Promise<void> {
await this.ensureInitialized()
console.log('🔄 Flushing Brainy indexes and caches to disk...')
console.log('Flushing Brainy indexes and caches to disk...')
const startTime = Date.now()
@ -4220,24 +4214,20 @@ export class Brainy<T = any> implements BrainyInterface<T> {
// 2. Flush metadata index (field indexes + EntityIdMapper)
this.metadataIndex.flush(),
// 3. Flush graph adjacency index (relationship cache)
// Note: Graph structure is already persisted via storage.saveVerb() calls
// This just flushes the in-memory cache for performance
this.graphIndex.flush()
// 3. Flush graph adjacency index (relationship cache + LSM trees)
this.graphIndex.flush(),
// Note: Write-through cache (storage.writeCache) is NOT cleared here
// Cache persists indefinitely for read-after-write consistency
// Provides safety net for:
// - Cloud storage eventual consistency (S3, GCS, Azure, R2)
// - Filesystem buffer cache timing
// - Type cache warming period (nounTypeCache population)
// Cache entries are removed only when explicitly deleted (deleteObjectFromBranch)
// Memory footprint is negligible for typical workloads (<10MB for 100k entities)
// 4. Flush HNSW dirty nodes (deferred persistence mode)
(async () => {
if (this.index && typeof (this.index as any).flush === 'function') {
await (this.index as any).flush()
}
})()
])
const elapsed = Date.now() - startTime
console.log(`All indexes flushed to disk in ${elapsed}ms`)
console.log(`All indexes flushed to disk in ${elapsed}ms`)
}
/**
@ -5619,15 +5609,22 @@ export class Brainy<T = any> implements BrainyInterface<T> {
/**
* Execute vector search component
*
* @param params Find parameters
* @param candidateIds Optional pre-resolved metadata filter IDs for metadata-first search.
* When provided, HNSW search is restricted to these candidates:
* - NativeHNSWWrapper: uses native searchWithCandidates (Rust bitmap filtering)
* - JS HNSWIndex: converts to filter function with O(1) Set lookups
*/
private async executeVectorSearch(params: FindParams<T>): Promise<Result<T>[]> {
private async executeVectorSearch(params: FindParams<T>, candidateIds?: string[]): Promise<Result<T>[]> {
const vector = params.vector || (await this.embed(params.query!))
const limit = params.limit || 10
// Pass type for type-aware indexes (10x faster for type-specific queries)
const searchResults: [string, number][] = this.indexIsTypeAware && params.type
? await (this.index as any).search(vector, limit * 2, params.type as any)
: await this.index.search(vector, limit * 2)
// Build search options for metadata-first candidate filtering
const searchOptions = candidateIds ? { candidateIds } : undefined
// HNSW search with optional metadata-first candidate filtering
const searchResults: [string, number][] = await this.index.search(vector, limit * 2, undefined, searchOptions)
// Batch-load entities for 10-50x faster cloud storage performance
// GCS: 10 results = 1×50ms vs 10×50ms = 500ms (10x faster)
@ -5655,10 +5652,7 @@ export class Brainy<T = any> implements BrainyInterface<T> {
const nearEntity = await this.get(params.near.id)
if (!nearEntity) return []
// Pass type for type-aware indexes
const nearResults: [string, number][] = this.indexIsTypeAware && params.type
? await (this.index as any).search(nearEntity.vector, params.limit || 10, params.type as any)
: await this.index.search(nearEntity.vector, params.limit || 10)
const nearResults: [string, number][] = await this.index.search(nearEntity.vector, params.limit || 10)
// Filter by threshold first to minimize batch fetch
const threshold = params.near.threshold || 0.7
@ -6252,7 +6246,7 @@ export class Brainy<T = any> implements BrainyInterface<T> {
const storageConfig = (this.config.storage || {}) as Record<string, unknown>
const storageType = (storageConfig.type as string) || 'auto'
// Check plugin-provided storage factories (e.g., 'mmap-filesystem' from cortex)
// Check plugin-provided storage factories (e.g., 'filesystem' override from cortex)
const pluginFactory = this.pluginRegistry.getStorageFactory(storageType)
if (pluginFactory) {
const adapter = await pluginFactory.create(storageConfig)
@ -6290,18 +6284,13 @@ export class Brainy<T = any> implements BrainyInterface<T> {
}
/**
* Setup index
* Setup index single unified HNSW graph.
*
* Phase 2: Uses TypeAwareHNSWIndex for billion-scale optimization
* - 87% memory reduction through separate graphs per entity type
* - 10x faster type-specific queries
* - Automatic type routing
*
* Smart defaults for HNSW persistence mode
* - Cloud storage (GCS/S3/R2/Azure): 'deferred' for 30-50× faster adds
* Smart defaults for HNSW persistence mode:
* - Cloud storage (GCS/S3/R2/Azure): 'deferred' for 30-50x faster adds
* - Local storage (FileSystem/Memory/OPFS): 'immediate' (already fast)
*/
private setupIndex(): HNSWIndex | TypeAwareHNSWIndex {
private setupIndex(): HNSWIndex {
const indexConfig = {
...this.config.index,
distanceFunction: this.distance,
@ -6316,24 +6305,18 @@ export class Brainy<T = any> implements BrainyInterface<T> {
const persistMode = this.resolveHNSWPersistMode()
// Use TypeAwareHNSWIndex for non-memory storage (billion-scale optimization)
const detectedStorageType = this.config.storage?.type || this.getStorageType()
if (detectedStorageType !== 'memory') {
return new TypeAwareHNSWIndex(indexConfig, this.distance, {
storage: this.storage,
useParallelization: true,
persistMode
})
}
return new HNSWIndex(indexConfig as any, this.distance, { persistMode })
return new HNSWIndex(indexConfig as any, this.distance, {
storage: this.storage,
useParallelization: true,
persistMode
})
}
/**
* Create an HNSW index, using plugin factory when available.
* Shared by init(), fork(), checkout(), and clear() to avoid duplication.
*/
private createIndex(): HNSWIndex | TypeAwareHNSWIndex {
private createIndex(): HNSWIndex {
const hnswFactory = this.pluginRegistry.getProvider<(config: any, distance: DistanceFunction, options: any) => any>('hnsw')
if (hnswFactory) {
const persistMode = this.resolveHNSWPersistMode()
@ -6719,13 +6702,36 @@ export class Brainy<T = any> implements BrainyInterface<T> {
* This ensures deferred persistence mode data is saved
*/
async close(): Promise<void> {
// Flush HNSW dirty nodes before closing
// In deferred persistence mode, this persists all pending HNSW graph data
if (this.index && typeof this.index.flush === 'function') {
await this.index.flush()
}
// Flush ALL components before closing to prevent data loss
// This is critical when cortex native providers buffer data in Rust memory
await Promise.all([
// Flush HNSW dirty nodes (deferred persistence mode)
(async () => {
if (this.index && typeof (this.index as any).flush === 'function') {
await (this.index as any).flush()
}
})(),
// Flush metadata index (field indexes + EntityIdMapper)
(async () => {
if (this.metadataIndex && typeof this.metadataIndex.flush === 'function') {
await this.metadataIndex.flush()
}
})(),
// Flush graph adjacency index (LSM trees)
(async () => {
if (this.graphIndex && typeof this.graphIndex.flush === 'function') {
await this.graphIndex.flush()
}
})(),
// Flush storage adapter counts
(async () => {
if (this.storage && typeof (this.storage as any).flushCounts === 'function') {
await (this.storage as any).flushCounts()
}
})()
])
// Deactivate plugins
// Deactivate plugins (safe — all data flushed above)
await this.pluginRegistry.deactivateAll()
// Restore console methods if silent mode was enabled

View file

@ -14,9 +14,23 @@
* - Footer: checksum, stats
*/
import { encode, decode } from '@msgpack/msgpack'
import { encode as defaultEncode, decode as defaultDecode } from '@msgpack/msgpack'
import { BloomFilter, SerializedBloomFilter } from './BloomFilter.js'
// Swappable msgpack implementation — defaults to @msgpack/msgpack JS,
// can be replaced with native msgpack (e.g., cortex's Rust-backed encoder)
let _encode: (data: unknown) => Uint8Array = defaultEncode as (data: unknown) => Uint8Array
let _decode: (data: Uint8Array) => unknown = defaultDecode as (data: Uint8Array) => unknown
/**
* Replace the msgpack encode/decode implementation at runtime.
* Called by brainy.ts when a cortex 'msgpack' provider is registered.
*/
export function setMsgpackImplementation(impl: { encode: (data: unknown) => Uint8Array, decode: (data: Uint8Array) => unknown }) {
_encode = impl.encode
_decode = impl.decode
}
/**
* Entry in the SSTable
* Maps a source node to its target nodes
@ -299,7 +313,7 @@ export class SSTable {
checksum: this.calculateChecksum(this.entries)
}
const serialized = encode(data)
const serialized = _encode(data)
// Update size in metadata
this.metadata.sizeBytes = serialized.length
@ -331,7 +345,7 @@ export class SSTable {
* @returns SSTable instance
*/
static deserialize(data: Uint8Array): SSTable {
const decoded = decode(data) as SerializedSSTable
const decoded = _decode(data) as SerializedSSTable
// Verify checksum
const sstable = new SSTable(

View file

@ -681,12 +681,18 @@ export class HNSWIndex {
queryVector: Vector,
k: number = 10,
filter?: (id: string) => Promise<boolean>,
options?: { rerank?: { multiplier: number } }
options?: { rerank?: { multiplier: number }; candidateIds?: string[] }
): Promise<Array<[string, number]>> {
if (this.nouns.size === 0) {
return []
}
// Metadata-first: convert candidateIds to filter function if no explicit filter
if (!filter && options?.candidateIds && options.candidateIds.length > 0) {
const candidateSet = new Set(options.candidateIds)
filter = async (id: string) => candidateSet.has(id)
}
// Check if query vector is defined
if (!queryVector) {
throw new Error('Query vector is undefined or null')

View file

@ -257,7 +257,7 @@ export class TypeAwareHNSWIndex {
k: number = 10,
type?: NounType | NounType[],
filter?: (id: string) => Promise<boolean>,
options?: { rerank?: { multiplier: number } }
options?: { rerank?: { multiplier: number }; candidateIds?: string[] }
): Promise<Array<[string, number]>> {
// Single-type search (fast path)
if (type && typeof type === 'string') {
@ -288,7 +288,7 @@ export class TypeAwareHNSWIndex {
k: number,
types: NounType[],
filter?: (id: string) => Promise<boolean>,
options?: { rerank?: { multiplier: number } }
options?: { rerank?: { multiplier: number }; candidateIds?: string[] }
): Promise<Array<[string, number]>> {
const allResults: Array<[string, number]> = []
@ -323,7 +323,7 @@ export class TypeAwareHNSWIndex {
queryVector: Vector,
k: number,
filter?: (id: string) => Promise<boolean>,
options?: { rerank?: { multiplier: number } }
options?: { rerank?: { multiplier: number }; candidateIds?: string[] }
): Promise<Array<[string, number]>> {
const allResults: Array<[string, number]> = []

View file

@ -9,3 +9,5 @@ export { FieldTypeInference, FieldType } from './utils/fieldTypeInference.js'
export type { FieldTypeInfo } from './utils/fieldTypeInference.js'
export { EntityIdMapper } from './utils/entityIdMapper.js'
export type { EntityIdMapperOptions, EntityIdMapperData } from './utils/entityIdMapper.js'
export { getRecommendedCacheConfig, formatBytes, checkMemoryPressure } from './utils/memoryDetection.js'
export type { MemoryInfo, CacheAllocationStrategy } from './utils/memoryDetection.js'

View file

@ -2,8 +2,7 @@
* Index Operations with Rollback Support
*
* Provides transactional operations for all indexes:
* - TypeAwareHNSWIndex (per-type vector indexes)
* - HNSWIndex (legacy/fallback vector index)
* - HNSWIndex (unified vector index)
* - MetadataIndexManager (roaring bitmap filtering)
* - GraphAdjacencyIndex (LSM-tree graph storage)
*
@ -11,10 +10,8 @@
*/
import type { HNSWIndex } from '../../hnsw/hnswIndex.js'
import type { TypeAwareHNSWIndex } from '../../hnsw/typeAwareHNSWIndex.js'
import type { MetadataIndexManager } from '../../utils/metadataIndex.js'
import type { GraphAdjacencyIndex } from '../../graph/graphAdjacencyIndex.js'
import type { NounType } from '../../types/graphTypes.js'
import type { GraphVerb } from '../../coreTypes.js'
import type { Operation, RollbackAction } from '../types.js'
@ -67,55 +64,6 @@ export class AddToHNSWOperation implements Operation {
}
}
/**
* Add to TypeAwareHNSW index with rollback support
*
* Rollback strategy:
* - Remove item from type-specific index
*/
export class AddToTypeAwareHNSWOperation implements Operation {
readonly name = 'AddToTypeAwareHNSW'
constructor(
private readonly index: TypeAwareHNSWIndex,
private readonly id: string,
private readonly vector: number[],
private readonly nounType: NounType
) {}
async execute(): Promise<RollbackAction> {
// Check if item already exists
const existed = await this.itemExists(this.id, this.nounType)
// Add to type-specific index
await this.index.addItem({ id: this.id, vector: this.vector }, this.nounType)
// Return rollback action
return async () => {
if (!existed) {
// Remove newly added item from type-specific index
await this.index.removeItem(this.id, this.nounType)
}
}
}
/**
* Check if item exists in type-specific index
*/
private async itemExists(id: string, type: NounType): Promise<boolean> {
try {
const index = (this.index as any).getIndexForType?.(type)
if (index) {
await index.getItem?.(id)
return true
}
return false
} catch {
return false
}
}
}
/**
* Remove from HNSW index with rollback support
*
@ -145,34 +93,6 @@ export class RemoveFromHNSWOperation implements Operation {
}
}
/**
* Remove from TypeAwareHNSW index with rollback support
*
* Rollback strategy:
* - Re-add item to type-specific index with original vector
*/
export class RemoveFromTypeAwareHNSWOperation implements Operation {
readonly name = 'RemoveFromTypeAwareHNSW'
constructor(
private readonly index: TypeAwareHNSWIndex,
private readonly id: string,
private readonly vector: number[], // Required for rollback
private readonly nounType: NounType
) {}
async execute(): Promise<RollbackAction> {
// Remove from type-specific index
await this.index.removeItem(this.id, this.nounType)
// Return rollback action
return async () => {
// Re-add item with original vector to type-specific index
await this.index.addItem({ id: this.id, vector: this.vector }, this.nounType)
}
}
}
/**
* Add to metadata index with rollback support
*

View file

@ -22,9 +22,7 @@ export {
// Index Operations
export {
AddToHNSWOperation,
AddToTypeAwareHNSWOperation,
RemoveFromHNSWOperation,
RemoveFromTypeAwareHNSWOperation,
AddToMetadataIndexOperation,
RemoveFromMetadataIndexOperation,
AddToGraphIndexOperation,

View file

@ -14,7 +14,6 @@
*/
import { HNSWIndex } from '../hnsw/hnswIndex.js'
import { TypeAwareHNSWIndex } from '../hnsw/typeAwareHNSWIndex.js'
import { MetadataIndexManager } from '../utils/metadataIndex.js'
import { Vector } from '../coreTypes.js'
import { NounType } from '../types/graphTypes.js'
@ -231,7 +230,7 @@ class QueryPlanner {
*/
export class TripleIntelligenceSystem {
private metadataIndex: MetadataIndexManager
private hnswIndex: HNSWIndex | TypeAwareHNSWIndex
private hnswIndex: HNSWIndex
private graphIndex: GraphAdjacencyIndex
private metrics: PerformanceMetrics
private planner: QueryPlanner
@ -240,7 +239,7 @@ export class TripleIntelligenceSystem {
constructor(
metadataIndex: MetadataIndexManager,
hnswIndex: HNSWIndex | TypeAwareHNSWIndex,
hnswIndex: HNSWIndex,
graphIndex: GraphAdjacencyIndex,
embedder: (text: string) => Promise<Vector>,
storage: any
@ -318,10 +317,8 @@ export class TripleIntelligenceSystem {
? await this.embedder(query)
: query
// Phase 3: Pass types to TypeAwareHNSWIndex for optimized search
const searchResults = this.hnswIndex instanceof TypeAwareHNSWIndex
? await this.hnswIndex.search(vector, limit, types)
: await this.hnswIndex.search(vector, limit)
// Single unified HNSW search (type filtering handled by metadata-first optimization)
const searchResults = await this.hnswIndex.search(vector, limit)
// Convert to result format
const results: TripleResult[] = []

View file

@ -150,6 +150,15 @@ export class EntityIdMapper {
return this.uuidToInt.size
}
/**
* Get all mapped integer IDs without storage reads.
* Used for bitmap negation operations (ne, exists:false, missing:true)
* to avoid full-table getAllIds() scans.
*/
getAllIntIds(): number[] {
return Array.from(this.intToUuid.keys())
}
/**
* Convert array of UUIDs to array of integers
*/

View file

@ -56,6 +56,10 @@ export interface MetadataIndexConfig {
excludeFields?: string[] // Never index these fields
}
export interface MetadataIndexOptions {
entityIdMapper?: EntityIdMapper // Optional pre-configured EntityIdMapper (e.g., native from cortex)
}
/**
* Manages metadata indexes for fast filtering
* Maintains inverted indexes: field+value -> list of IDs
@ -143,7 +147,7 @@ export class MetadataIndexManager {
// Replaces unreliable pattern matching with DuckDB-inspired value analysis
private fieldTypeInference: FieldTypeInference
constructor(storage: StorageAdapter, config: MetadataIndexConfig = {}) {
constructor(storage: StorageAdapter, config: MetadataIndexConfig = {}, options: MetadataIndexOptions = {}) {
this.storage = storage
this.config = {
maxIndexSize: config.maxIndexSize ?? 10000,
@ -185,8 +189,8 @@ export class MetadataIndexManager {
// Get global unified cache for coordinated memory management
this.unifiedCache = getGlobalCache()
// Initialize EntityIdMapper for roaring bitmap UUID ↔ integer mapping
this.idMapper = new EntityIdMapper({
// Use injected EntityIdMapper (e.g., native from cortex) or create JS fallback
this.idMapper = options.entityIdMapper ?? new EntityIdMapper({
storage,
storageKey: 'brainy:entityIdMapper'
})
@ -1868,11 +1872,15 @@ export class MetadataIndexManager {
if (allIds.length === 0) return []
if (allIds.length === 1) return allIds[0]
// Intersection of all sets
return allIds.reduce((intersection, currentSet) =>
intersection.filter(id => currentSet.includes(id))
)
// Set-based intersection O(n) — start with smallest set for optimal perf
const sorted = allIds.sort((a, b) => a.length - b.length)
let result = new Set(sorted[0])
for (let i = 1; i < sorted.length; i++) {
const current = new Set(sorted[i])
result = new Set([...result].filter(id => current.has(id)))
}
return Array.from(result)
}
if (filter.anyOf && Array.isArray(filter.anyOf)) {
@ -1930,21 +1938,22 @@ export class MetadataIndexManager {
// Canonical: 'ne' | Alias: 'notEquals' | Deprecated: 'isNot'
case 'isNot': // DEPRECATED: Use 'ne' instead
case 'notEquals': // Alias for 'ne'
case 'ne':
case 'ne': {
// For notEquals, we need all IDs EXCEPT those matching the value
// This is especially important for soft delete: deleted !== true
// should include items without a deleted field
// First, get all IDs in the database
const allItemIds = await this.getAllIds()
// Use EntityIdMapper universe (in-memory) instead of getAllIds() storage scan
const allKnownIds = this.idMapper.intsIterableToUuids(this.idMapper.getAllIntIds())
// Then get IDs that match the value we want to exclude
const excludeIds = await this.getIds(field, operand)
const excludeSet = new Set(excludeIds)
// Return all IDs except those to exclude
fieldResults = allItemIds.filter(id => !excludeSet.has(id))
fieldResults = allKnownIds.filter(id => !excludeSet.has(id))
break
}
// ===== MULTI-VALUE OPERATORS =====
// Canonical: 'in' | Alias: 'oneOf'
@ -2034,8 +2043,7 @@ export class MetadataIndexManager {
fieldResults = this.idMapper.intsIterableToUuids(allIntIds)
} else {
// exists: false - Get all IDs that DON'T have this field
// Fixed excludeVFS bug (was returning empty array)
const allItemIds = await this.getAllIds()
// Uses EntityIdMapper universe (in-memory) instead of getAllIds() storage scan
const existsIntIds = new Set<number>()
// Get IDs that HAVE this field
@ -2053,10 +2061,11 @@ export class MetadataIndexManager {
}
}
// Convert to UUIDs and subtract from all IDs
// Use EntityIdMapper universe and subtract IDs that have this field
const allKnownIds = this.idMapper.intsIterableToUuids(this.idMapper.getAllIntIds())
const existsUuids = this.idMapper.intsIterableToUuids(existsIntIds)
const existsSet = new Set(existsUuids)
fieldResults = allItemIds.filter(id => !existsSet.has(id))
fieldResults = allKnownIds.filter(id => !existsSet.has(id))
}
break
@ -2068,7 +2077,7 @@ export class MetadataIndexManager {
// Added for API consistency with in-memory metadataFilter
if (operand) {
// missing: true - field does NOT exist (same as exists: false)
const allItemIds = await this.getAllIds()
// Uses EntityIdMapper universe (in-memory) instead of getAllIds() storage scan
const existsIntIds = new Set<number>()
const sparseIndex = await this.loadSparseIndex(field)
@ -2085,9 +2094,11 @@ export class MetadataIndexManager {
}
}
// Use EntityIdMapper universe and subtract IDs that have this field
const allKnownIds = this.idMapper.intsIterableToUuids(this.idMapper.getAllIntIds())
const existsUuids = this.idMapper.intsIterableToUuids(existsIntIds)
const existsSet = new Set(existsUuids)
fieldResults = allItemIds.filter(id => !existsSet.has(id))
fieldResults = allKnownIds.filter(id => !existsSet.has(id))
} else {
// missing: false - field DOES exist (same as exists: true)
const allIntIds = new Set<number>()
@ -2126,11 +2137,15 @@ export class MetadataIndexManager {
if (idSets.length === 0) return []
if (idSets.length === 1) return idSets[0]
// Intersection of all field criteria (implicit AND)
return idSets.reduce((intersection, currentSet) =>
intersection.filter(id => currentSet.includes(id))
)
// Set-based intersection O(n) — start with smallest set for optimal perf
const sortedSets = idSets.sort((a, b) => a.length - b.length)
let resultSet = new Set(sortedSets[0])
for (let i = 1; i < sortedSets.length; i++) {
const current = new Set(sortedSets[i])
resultSet = new Set([...resultSet].filter(id => current.has(id)))
}
return Array.from(resultSet)
}
/**

View file

@ -12,7 +12,7 @@
*/
import {
RoaringBitmap32,
RoaringBitmap32 as WasmRoaringBitmap32,
RoaringBitmap32Iterator,
roaringLibraryInitialize,
roaringLibraryIsReady,
@ -24,9 +24,35 @@ import {
// This ensures RoaringBitmap32 is ready to use immediately after import
await roaringLibraryInitialize()
// Re-export all types and values
// Swappable implementation — defaults to WASM, can be replaced with native CRoaring
// Uses a wrapper class that delegates to the active implementation so that the exported
// RoaringBitmap32 remains a class (usable as both value and type).
let _impl: typeof WasmRoaringBitmap32 = WasmRoaringBitmap32
/**
* Replace the RoaringBitmap32 implementation at runtime.
* Called by brainy.ts when a cortex 'roaring' provider is registered.
*/
export function setRoaringImplementation(impl: typeof WasmRoaringBitmap32) {
_impl = impl
}
/**
* Get the current RoaringBitmap32 implementation (WASM or native).
* Use this instead of direct `new RoaringBitmap32()` when constructing bitmaps
* in hot paths that should benefit from native replacement.
*/
export function getRoaringBitmap32(): typeof WasmRoaringBitmap32 {
return _impl
}
// Re-export the original WASM class as RoaringBitmap32 for type compatibility
// Consumers use this as both a type (Map<string, RoaringBitmap32>) and value (new RoaringBitmap32())
// The WASM and native implementations are API-compatible
export { WasmRoaringBitmap32 as RoaringBitmap32 }
// Re-export remaining types and values
export {
RoaringBitmap32,
RoaringBitmap32Iterator,
roaringLibraryInitialize,
roaringLibraryIsReady,