fix: flush graph LSM-trees on close to prevent data loss across restarts

GraphAdjacencyIndex.flush() was a no-op — LSM MemTables were never
written to SSTables for datasets under the 100K auto-flush threshold.
This caused readdir, getRelations, and getDescendants to return empty
results after close + reopen.

Three fixes:
- LSMTree.get(): merge MemTable + SSTable results (data spans both
  after flush, old early-return missed SSTable data)
- GraphAdjacencyIndex.flush(): actually flush all 4 LSM-trees
- GraphAdjacencyIndex.close(): close all 4 trees (was only closing 2)

Also: brain.close() and shutdown hooks now call close() on graphIndex,
HNSW index, and metadataIndex to release timers and file handles.
This commit is contained in:
David Snelling 2026-02-01 17:55:40 -08:00
parent 0add0af45a
commit ab2493af02
4 changed files with 213 additions and 22 deletions

View file

@ -461,7 +461,7 @@ export class Brainy<T = any> implements BrainyInterface<T> {
let flushedCount = 0
for (const instance of Brainy.instances) {
if (instance.initialized) {
// Full flush: counts + metadata index + graph index + HNSW dirty nodes
// Flush all buffered data, then close to release resources (timers, handles)
await Promise.all([
(async () => {
if (instance.storage && typeof (instance.storage as any).flushCounts === 'function') {
@ -484,6 +484,24 @@ export class Brainy<T = any> implements BrainyInterface<T> {
}
})()
])
// Close components to stop timers that would prevent clean process exit
await Promise.all([
(async () => {
if (instance.graphIndex && typeof instance.graphIndex.close === 'function') {
await instance.graphIndex.close()
}
})(),
(async () => {
if (instance.index && typeof (instance.index as any).close === 'function') {
await (instance.index as any).close()
}
})(),
(async () => {
if (instance.metadataIndex && typeof (instance.metadataIndex as any).close === 'function') {
await (instance.metadataIndex as any).close()
}
})(),
])
flushedCount++
}
}
@ -6702,7 +6720,7 @@ export class Brainy<T = any> implements BrainyInterface<T> {
* This ensures deferred persistence mode data is saved
*/
async close(): Promise<void> {
// Flush ALL components before closing to prevent data loss
// Phase 1: Flush ALL components in parallel to persist buffered data
// This is critical when cortex native providers buffer data in Rust memory
await Promise.all([
// Flush HNSW dirty nodes (deferred persistence mode)
@ -6731,7 +6749,27 @@ export class Brainy<T = any> implements BrainyInterface<T> {
})()
])
// Deactivate plugins (safe — all data flushed above)
// Phase 2: Close components to release resources (timers, file handles)
// Data is already safe on disk from Phase 1
await Promise.all([
(async () => {
if (this.graphIndex && typeof this.graphIndex.close === 'function') {
await this.graphIndex.close()
}
})(),
(async () => {
if (this.index && typeof (this.index as any).close === 'function') {
await (this.index as any).close()
}
})(),
(async () => {
if (this.metadataIndex && typeof (this.metadataIndex as any).close === 'function') {
await (this.metadataIndex as any).close()
}
})(),
])
// Deactivate plugins (safe — all data flushed and resources released above)
await this.pluginRegistry.deactivateAll()
// Restore console methods if silent mode was enabled
@ -6751,8 +6789,6 @@ export class Brainy<T = any> implements BrainyInterface<T> {
}
}
// Storage doesn't have close in current interface
// We'll just mark as not initialized
this.initialized = false
}

View file

@ -800,10 +800,21 @@ export class GraphAdjacencyIndex {
const startTime = Date.now()
// Flush both LSM-trees
// Note: LSMTree.close() will handle flushing MemTable
// For now, we don't have an explicit flush method in LSMTree
// The MemTable will be flushed automatically when threshold is reached
// Flush all 4 LSM-trees in parallel (MemTables → SSTables on disk)
await Promise.all([
this.lsmTreeSource.flush().then(() => {
prodLog.debug(`GraphAdjacencyIndex: Flushed source tree`)
}),
this.lsmTreeTarget.flush().then(() => {
prodLog.debug(`GraphAdjacencyIndex: Flushed target tree`)
}),
this.lsmTreeVerbsBySource.flush().then(() => {
prodLog.debug(`GraphAdjacencyIndex: Flushed verbs-by-source tree`)
}),
this.lsmTreeVerbsByTarget.flush().then(() => {
prodLog.debug(`GraphAdjacencyIndex: Flushed verbs-by-target tree`)
}),
])
const elapsed = Date.now() - startTime
@ -819,10 +830,14 @@ export class GraphAdjacencyIndex {
this.flushTimer = undefined
}
// Close LSM-trees (will flush MemTables)
// Close all 4 LSM-trees (will flush MemTables to SSTables)
if (this.initialized) {
await this.lsmTreeSource.close()
await this.lsmTreeTarget.close()
await Promise.all([
this.lsmTreeSource.close(),
this.lsmTreeTarget.close(),
this.lsmTreeVerbsBySource.close(),
this.lsmTreeVerbsByTarget.close(),
])
}
prodLog.info('GraphAdjacencyIndex: Shutdown complete')

View file

@ -296,18 +296,21 @@ export class LSMTree {
async get(sourceId: string): Promise<string[] | null> {
const startTime = performance.now()
// Check MemTable first (hot data)
// Merge results from MemTable AND SSTables
// Data can span both after a flush (old data in SSTables, new in MemTable)
const allTargets = new Set<string>()
// Check MemTable (hot data)
const memResult = this.memTable.get(sourceId)
if (memResult !== null) {
return memResult
for (const target of memResult) {
allTargets.add(target)
}
}
// Check SSTables from newest to oldest
// Newer levels (L0, L1, L2) checked first for better cache locality
const maxLevel = Math.max(...Array.from(this.sstablesByLevel.keys()), 0)
const allTargets = new Set<string>()
for (let level = 0; level <= maxLevel; level++) {
const sstables = this.sstablesByLevel.get(level) || []
@ -608,16 +611,20 @@ export class LSMTree {
}
/**
* Flush MemTable and stop compaction
* Called during shutdown
* Flush MemTable to SSTables without closing
* Called by GraphAdjacencyIndex.flush() and brain.close()
*/
async flush(): Promise<void> {
if (!this.memTable.isEmpty()) {
await this.flushMemTable()
}
}
async close(): Promise<void> {
this.stopCompactionTimer()
// Final MemTable flush
if (!this.memTable.isEmpty()) {
await this.flushMemTable()
}
await this.flush()
prodLog.info('LSMTree: Closed successfully')
}

View file

@ -0,0 +1,133 @@
/**
* Regression test: VFS data persists across restart (close + reopen)
*
* Verifies the fix for GraphAdjacencyIndex.flush() and LSMTree.get()
* which previously caused graph relationships to be lost on restart
* because LSM MemTables were never flushed to SSTables on close.
*/
import { describe, it, expect } from 'vitest'
import { Brainy } from '../../src/brainy.js'
import { VerbType } from '../../src/types/graphTypes.js'
import { mkdirSync, rmSync } from 'fs'
describe('VFS restart persistence', () => {
it('entities and relationships survive close + reopen', async () => {
const dir = '/tmp/brainy-restart-fix-' + Date.now()
mkdirSync(dir, { recursive: true })
let entityId: string | undefined
const rootId = '00000000-0000-0000-0000-000000000000'
try {
// === SESSION 1: Write data ===
let brain = new Brainy({
storage: { type: 'filesystem', rootDirectory: dir },
disableAutoRebuild: true,
plugins: [],
silent: true,
})
await brain.init()
await brain.vfs.writeFile('/chapter-1.txt', 'Once upon a time...')
entityId = await brain.vfs.resolvePathToId('/chapter-1.txt')
expect(entityId).toBeTruthy()
// Verify within same session
const entity1 = await brain.get(entityId!)
expect(entity1).not.toBeNull()
const entries1 = await brain.vfs.readdir('/')
expect(entries1.length).toBeGreaterThan(0)
await brain.close()
// === SESSION 2: Read data after restart ===
brain = new Brainy({
storage: { type: 'filesystem', rootDirectory: dir },
disableAutoRebuild: true,
plugins: [],
silent: true,
})
await brain.init()
// Entity should be retrievable
const entity2 = await brain.get(entityId!)
expect(entity2).not.toBeNull()
// Root entity should exist
const rootEntity = await brain.get(rootId)
expect(rootEntity).not.toBeNull()
// readdir should return the file
const entries2 = await brain.vfs.readdir('/')
expect(entries2.length).toBeGreaterThan(0)
expect(entries2).toContain('chapter-1.txt')
// getRelations should find the Contains relationship
const relations2 = await brain.getRelations({ from: rootId, type: VerbType.Contains })
expect(relations2.length).toBeGreaterThan(0)
await brain.close()
} finally {
try { rmSync(dir, { recursive: true }) } catch {}
}
})
it('multiple files and subdirectories survive restart', async () => {
const dir = '/tmp/brainy-restart-multi-' + Date.now()
mkdirSync(dir, { recursive: true })
const rootId = '00000000-0000-0000-0000-000000000000'
try {
// === SESSION 1: Write multiple files ===
let brain = new Brainy({
storage: { type: 'filesystem', rootDirectory: dir },
disableAutoRebuild: true,
plugins: [],
silent: true,
})
await brain.init()
await brain.vfs.writeFile('/readme.txt', 'Project readme')
await brain.vfs.writeFile('/docs/guide.txt', 'User guide')
await brain.vfs.writeFile('/docs/api.txt', 'API reference')
const entries1 = await brain.vfs.readdir('/')
expect(entries1.length).toBe(2) // readme.txt + docs/
const docEntries1 = await brain.vfs.readdir('/docs')
expect(docEntries1.length).toBe(2) // guide.txt + api.txt
await brain.close()
// === SESSION 2: Verify all data persisted ===
brain = new Brainy({
storage: { type: 'filesystem', rootDirectory: dir },
disableAutoRebuild: true,
plugins: [],
silent: true,
})
await brain.init()
// Root children
const entries2 = await brain.vfs.readdir('/')
expect(entries2.length).toBe(2)
expect(entries2.sort()).toEqual(['docs', 'readme.txt'])
// Subdirectory children
const docEntries2 = await brain.vfs.readdir('/docs')
expect(docEntries2.length).toBe(2)
expect(docEntries2.sort()).toEqual(['api.txt', 'guide.txt'])
// Root relations
const rootRelations = await brain.getRelations({ from: rootId, type: VerbType.Contains })
expect(rootRelations.length).toBeGreaterThan(0)
await brain.close()
} finally {
try { rmSync(dir, { recursive: true }) } catch {}
}
})
})