open-brainy/tests/unit/brainy/warm.test.ts

399 lines
15 KiB
TypeScript
Raw Normal View History

feat: warm contract (warm/warmOnOpen/provider warm hook), configurable transact budget floor, backend-neutral vector index op names Three pieces addressing the cold-restart-write incident where a production deployment's first writes after every restart (33-35s each on a cold page cache) blew the op-count-scaled transact budget mid-batch: every write is itself a multi-op transaction, so one cold operation consumed the whole budget, the gate before the next operation tripped, and the write rolled back atomically - refused, retried, and refused again until the page cache warmed passively. - The budget's start-gating contract is now explicit and pinned: it gates STARTING the next operation, never rolling back completed work for elapsed time (the shipped schedule since 8.7.0, now stated in contract JSDoc, guarded by code for operation 0, and enforced by regression tests). The 30s floor is configurable via transactionBudgetFloorMs for stores whose cold operations legitimately run long. - New brain.warm() eagerly loads the vector index, metadata index, and graph adjacency so first operations after a cold restart run at steady-state cost. Returns a WarmReport with an honest per-surface outcome (warmed / probed / unavailable) - never reports a probe as a warm. warmOnOpen: true runs it during init(). New optional provider hook warm() on the vector and graph plugin contracts. - Vector-index transaction op classes renamed from the backend-specific AddToHNSWOperation / RemoveFromHNSWOperation to backend-neutral AddToVectorIndexOperation / RemoveFromVectorIndexOperation, stamping the active backend into the emitted op-name string (AddToVectorIndex(js-hnsw) vs a native provider's own identity) so journals never misdirect an operator toward an index that isn't running.
2026-07-22 16:42:26 -07:00
/**
* @module tests/unit/brainy/warm
* @description Coverage for `brain.warm()` / `warmOnOpen` / the provider
* `warm?()` contract (the cold-restart readiness fix): first operations after
* a cold restart should run at steady-state cost instead of paying
* demand-load latency on the critical path.
*
* Uses a real, in-process fake plugin provider implementing the actual
* `VectorIndexProvider` contract from `src/plugin.ts` the real seam a
* native provider (e.g. a disk-native accelerator) plugs into. The real
* built-in JS metadata index and graph adjacency index run against real
* (filesystem or in-memory) storage with pre-existing data, so their
* hydration paths are exercised for real, not mocked.
*/
import { describe, it, expect, afterEach } from 'vitest'
import * as fs from 'node:fs'
import * as os from 'node:os'
import * as path from 'node:path'
import { Brainy } from '../../../src/brainy.js'
import { NounType, VerbType } from '../../../src/types/graphTypes.js'
import type { VectorIndexProvider } from '../../../src/plugin.js'
import type { VectorDocument, Vector } from '../../../src/coreTypes.js'
import { MetadataIndexManager } from '../../../src/utils/metadataIndex.js'
import { GraphAdjacencyIndex } from '../../../src/graph/graphAdjacencyIndex.js'
const tmpDirs: string[] = []
function mkTmp(): string {
const d = fs.mkdtempSync(path.join(os.tmpdir(), 'brainy-warm-'))
tmpDirs.push(d)
return d
}
afterEach(() => {
for (const d of tmpDirs.splice(0)) fs.rmSync(d, { recursive: true, force: true })
})
// Brainy's ValidationConfig fixes vectors at exactly 384 dimensions
// (src/utils/paramValidation.ts) — match it so `add()` doesn't reject test data.
const DIM = 384
const V = (seed = 1): number[] => Array.from({ length: DIM }, (_, i) => Math.sin(seed + i))
/**
* A real (not mocked) VectorIndexProvider implementation, backed by a plain
* Map, that optionally implements `warm()` the exact seam
* `AddToVectorIndexOperation` / `brain.warm()` call through.
*/
class FakeVectorProvider implements VectorIndexProvider {
readonly name = 'fake-vector-provider'
feat: warm contract (warm/warmOnOpen/provider warm hook), configurable transact budget floor, backend-neutral vector index op names Three pieces addressing the cold-restart-write incident where a production deployment's first writes after every restart (33-35s each on a cold page cache) blew the op-count-scaled transact budget mid-batch: every write is itself a multi-op transaction, so one cold operation consumed the whole budget, the gate before the next operation tripped, and the write rolled back atomically - refused, retried, and refused again until the page cache warmed passively. - The budget's start-gating contract is now explicit and pinned: it gates STARTING the next operation, never rolling back completed work for elapsed time (the shipped schedule since 8.7.0, now stated in contract JSDoc, guarded by code for operation 0, and enforced by regression tests). The 30s floor is configurable via transactionBudgetFloorMs for stores whose cold operations legitimately run long. - New brain.warm() eagerly loads the vector index, metadata index, and graph adjacency so first operations after a cold restart run at steady-state cost. Returns a WarmReport with an honest per-surface outcome (warmed / probed / unavailable) - never reports a probe as a warm. warmOnOpen: true runs it during init(). New optional provider hook warm() on the vector and graph plugin contracts. - Vector-index transaction op classes renamed from the backend-specific AddToHNSWOperation / RemoveFromHNSWOperation to backend-neutral AddToVectorIndexOperation / RemoveFromVectorIndexOperation, stamping the active backend into the emitted op-name string (AddToVectorIndex(js-hnsw) vs a native provider's own identity) so journals never misdirect an operator toward an index that isn't running.
2026-07-22 16:42:26 -07:00
readonly items = new Map<string, Vector>()
warmCalls = 0
searchCalls: Array<{ k?: number }> = []
// Only present on the instance when the constructor is told to — mirrors a
// real provider that may or may not implement the optional hook. Assigned
// in the constructor BODY (not as a field initializer): under native ES
// class fields, field initializers run before constructor-body statements
// — including the parameter-property assignment — so referencing a
// parameter property from a field initializer would see it as still
// `undefined`.
warm?: () => Promise<void>
constructor(hasWarm: boolean) {
if (hasWarm) {
this.warm = async (): Promise<void> => {
this.warmCalls++
}
}
}
async addItem(item: VectorDocument): Promise<string> {
this.items.set(item.id, item.vector)
return item.id
}
async removeItem(id: string): Promise<boolean> {
return this.items.delete(id)
}
async search(_queryVector: Vector, k?: number): Promise<Array<[string, number]>> {
this.searchCalls.push({ k })
return [...this.items.keys()].slice(0, k ?? 10).map((id) => [id, 0])
}
size(): number {
return this.items.size
}
clear(): void {
this.items.clear()
}
async rebuild(): Promise<void> {}
async flush(): Promise<number> {
return 0
}
getPersistMode(): 'immediate' | 'deferred' {
return 'deferred'
}
}
/** Registers `provider` under the `'vector'` plugin key, before `init()`. */
function useFakeVectorProvider(brain: Brainy<any>, provider: FakeVectorProvider): void {
brain.use({
name: 'fake-vector-provider',
activate: async (ctx: any) => {
ctx.registerProvider('vector', () => provider)
return true
}
})
}
describe('brain.warm()', () => {
it('(a) calls the vector provider\'s warm() when present and reports "warmed"', async () => {
const brain = new Brainy({
requireSubtype: false,
storage: { type: 'memory' },
silent: true
})
const provider = new FakeVectorProvider(true)
useFakeVectorProvider(brain, provider)
await brain.init()
await brain.add({ data: 'a', type: NounType.Thing, vector: V(1) })
const report = await brain.warm()
expect(provider.warmCalls).toBe(1)
expect(provider.searchCalls.length).toBe(0) // warm() ran — no probe fallback
expect(report.vector.outcome).toBe('warmed')
await brain.close()
})
it('(b) falls back to a probe search() when warm() is absent and reports "probed", never "warmed"', async () => {
const brain = new Brainy({
requireSubtype: false,
storage: { type: 'memory' },
silent: true
})
const provider = new FakeVectorProvider(false)
useFakeVectorProvider(brain, provider)
await brain.init()
await brain.add({ data: 'a', type: NounType.Thing, vector: V(1) })
await brain.add({ data: 'b', type: NounType.Thing, vector: V(2) })
const report = await brain.warm()
expect(provider.warmCalls).toBe(0) // no warm() on this provider
expect(provider.searchCalls.length).toBe(1) // the probe ran
expect(provider.searchCalls[0].k).toBe(Math.min(100, provider.size())) // k = min(100, size)
expect(report.vector.outcome).toBe('probed')
expect(report.vector.outcome).not.toBe('warmed') // never conflated
await brain.close()
})
it('vector: reports "unavailable" when there is nothing to probe (empty index, unknown dimension)', async () => {
const brain = new Brainy({
requireSubtype: false,
storage: { type: 'memory' },
silent: true
})
// A disk-native provider MAY honestly report 0 resident entries while
// durable data exists on disk (the same posture documented on
// `VectorIndexProvider.isReady` — "an mmap/disk-native index may
// legitimately report 0 resident entries"). warm()'s probe fallback
// reads `size()`, so this is the real trigger for "nothing to probe":
// never a k=0 search, an honest skip.
const provider = new FakeVectorProvider(false)
provider.size = () => 0
useFakeVectorProvider(brain, provider)
await brain.init() // the VFS root bootstrap write sets `dimensions`, but size() still reports 0
const report = await brain.warm()
expect(provider.searchCalls.length).toBe(0) // nothing probed
expect(report.vector.outcome).toBe('unavailable')
await brain.close()
})
it('(c) metadata + graph hydration paths actually execute against filesystem storage with pre-existing data', async () => {
const dir = mkTmp()
// Build a brain with real data (including a field NOT in the metadata
// index's own common-fields warm subset — 'wave' — so hydrateAll()'s
// FULL hydration is distinguishable from init()'s partial warmCache()),
// and real graph edges, then close it (persisting everything).
const seed = new Brainy({
requireSubtype: false,
storage: { type: 'filesystem', path: dir },
silent: true
})
await seed.init()
const ids: string[] = []
for (let i = 0; i < 6; i++) {
ids.push(
await seed.add({
data: `entity ${i}`,
type: NounType.Thing,
metadata: { wave: i % 3 },
vector: V(i + 1)
})
)
}
for (let i = 0; i + 1 < ids.length; i++) {
await seed.relate({ from: ids[i], to: ids[i + 1], type: VerbType.RelatedTo })
}
await seed.close()
// Cold-reopen a FRESH instance and spy on the real hydration seams before
// calling warm(), so we assert they actually ran (not just that the
// report claims they did).
const metaHydrateCalls: number[] = []
const loadedFields: string[] = []
const graphInitCalls: number[] = []
const origHydrateAll = MetadataIndexManager.prototype.hydrateAll
const origLoadSparseIndex = (MetadataIndexManager.prototype as any).loadSparseIndex
const origGraphInit = GraphAdjacencyIndex.prototype.init
MetadataIndexManager.prototype.hydrateAll = async function (...args: any[]) {
metaHydrateCalls.push(1)
return origHydrateAll.apply(this, args as any)
}
;(MetadataIndexManager.prototype as any).loadSparseIndex = async function (
field: string,
...args: any[]
) {
loadedFields.push(field)
return origLoadSparseIndex.apply(this, [field, ...args] as any)
}
GraphAdjacencyIndex.prototype.init = async function (...args: any[]) {
graphInitCalls.push(1)
return origGraphInit.apply(this, args as any)
}
try {
const brain = new Brainy({
requireSubtype: false,
storage: { type: 'filesystem', path: dir },
silent: true
})
await brain.init()
const report = await brain.warm()
expect(metaHydrateCalls.length).toBe(1) // hydrateAll() actually ran
expect(loadedFields).toContain('wave') // a NON-common field was loaded — full hydration, not the heuristic subset
expect(report.metadata.outcome).toBe('warmed')
expect(graphInitCalls.length).toBeGreaterThanOrEqual(1) // graph's full-load seam ran (once at brain init, once via warm())
expect(report.graph.outcome).toBe('warmed')
// Correctness survives: the hydrated data still answers queries.
const byWhere = await brain.find({ type: NounType.Thing, where: { wave: 1 } })
expect(byWhere.length).toBe(2) // waves 1,4 of 0..5
await brain.close()
} finally {
MetadataIndexManager.prototype.hydrateAll = origHydrateAll
;(MetadataIndexManager.prototype as any).loadSparseIndex = origLoadSparseIndex
GraphAdjacencyIndex.prototype.init = origGraphInit
}
})
it('(d) warmOnOpen: true runs warm() during init() — observable via the fake provider', async () => {
const brain = new Brainy({
requireSubtype: false,
storage: { type: 'memory' },
warmOnOpen: true,
silent: true
})
const provider = new FakeVectorProvider(true)
useFakeVectorProvider(brain, provider)
// No explicit brain.warm() call — warmOnOpen must have run it as part of init().
await brain.init()
expect(provider.warmCalls).toBe(1)
await brain.close()
})
it('warmOnOpen defaults to false — init() does NOT run warm() unless opted in', async () => {
const brain = new Brainy({
requireSubtype: false,
storage: { type: 'memory' },
silent: true
})
const provider = new FakeVectorProvider(true)
useFakeVectorProvider(brain, provider)
await brain.init()
expect(provider.warmCalls).toBe(0)
await brain.close()
})
it('(e) WarmReport shape: outcome literal + durationMs per surface, plus totalDurationMs', async () => {
const brain = new Brainy({
requireSubtype: false,
storage: { type: 'memory' },
silent: true
})
const provider = new FakeVectorProvider(true)
useFakeVectorProvider(brain, provider)
await brain.init()
await brain.add({ data: 'a', type: NounType.Thing, vector: V(1) })
const report = await brain.warm()
for (const surface of ['vector', 'metadata', 'graph'] as const) {
expect(['warmed', 'probed', 'unavailable']).toContain(report[surface].outcome)
expect(typeof report[surface].durationMs).toBe('number')
expect(report[surface].durationMs).toBeGreaterThanOrEqual(0)
}
expect(typeof report.totalDurationMs).toBe('number')
expect(report.totalDurationMs).toBeGreaterThanOrEqual(0)
await brain.close()
})
fix: warm() metadata surface routes through the active provider (warm hook added to the metadata contract); add maintenanceDebt() observability surface A production deployment's warm report showed metadata: 'unavailable' under a native metadata provider. brain.warm()'s metadata leg only duck-typed the built-in JS manager's hydrateAll() method, which a native provider has no reason to implement. - MetadataIndexProvider (src/plugin.ts) gains an optional warm?(): Promise<void> hook, mirroring the existing vector and graph provider hooks. brain.warm() now checks the active provider's own warm() FIRST, falls back to the JS manager's hydrateAll() when absent, and reports 'unavailable' only when neither exists -- never init() as a stand-in, since a native provider's init() may be a cheap verify rather than a real warm. - Tests (tests/unit/brainy/warm.test.ts): a live provider instance shaped to have warm() reports 'warmed' and the hook called with no hydrateAll fallback; shaped to have neither hook reports 'unavailable' (pins the honest branch); the unmodified built-in JS manager still reports 'warmed' via hydrateAll(), unchanged. Additive scope agreed mid-flight with the native-provider team: a maintenance-debt observability seam so an operator sees a grind coming instead of discovering it as a CPU storm. - New optional maintenanceDebt?(): Promise<ProviderMaintenanceDebt> hook on all three provider contracts (vector, metadata, graph -- the same three warm?() lives on). ProviderMaintenanceDebt is fields-all-optional: a provider reports only what it truly measures (pendingBytes, pendingItems, lastPassCompletedAt, lastPassOutcome, converging), never an estimate dressed as fact. - New public brain.maintenanceDebt(): a pure passthrough -- for each surface it calls only the active provider's own hook and reports the payload verbatim, or 'unavailable' when absent. No thresholds, no polling, no JS-side estimation; the provider owns the numbers, the operator owns the policy. - ProviderMaintenanceDebt, MaintenanceDebtReport, and MaintenanceDebtOutcome are exported from the package root. - Tests (tests/unit/brainy/maintenance-debt.test.ts): hook present reports 'reported' with the exact payload passed through; hook absent reports 'unavailable' on every surface; mixed surfaces resolve independently of each other. RELEASES.md gains the 8.10.1 entry covering both fixes above and this feature, including the no-hot-retry contract from the prior commit.
2026-07-24 16:02:01 -07:00
// --- Metadata leg routes through the ACTIVE provider (8.10.1) -----------
//
// `MetadataIndexProvider` is a ~50-method interface (src/plugin.ts) — far
// too large to hand-write a compliant fake class the way `FakeVectorProvider`
// fakes the ~8-method `VectorIndexProvider` above. Test (c) already
// establishes this file's pattern for the metadata leg: exercise the REAL
// `MetadataIndexManager` instance and shape just the probe points brain.ts
// reads (`typeof provider.warm === 'function'` /
// `typeof provider.hydrateAll === 'function'`) directly on that instance.
// Shadowing an own property on the live object stands in for "a different
// provider implementation" without needing a hand-rolled full fake — the
// rest of the real manager (used by add()/init() above) is untouched.
describe('metadata leg — warm() routes through the active provider', () => {
it('(f) calls the ACTIVE metadata provider\'s warm() when present and reports "warmed", never falling back to hydrateAll', async () => {
const brain = new Brainy({
requireSubtype: false,
storage: { type: 'memory' },
silent: true
})
await brain.init()
await brain.add({ data: 'a', type: NounType.Thing, vector: V(1) })
const metadataIndex = (brain as any).metadataIndex
let warmCalls = 0
let hydrateAllCalls = 0
const origHydrateAll = metadataIndex.hydrateAll.bind(metadataIndex)
metadataIndex.hydrateAll = async (...args: unknown[]) => {
hydrateAllCalls++
return origHydrateAll(...args)
}
// Simulates a native metadata provider declaring the optional `warm()`
// hook added to `MetadataIndexProvider` (src/plugin.ts) in 8.10.1.
metadataIndex.warm = async () => {
warmCalls++
}
const report = await brain.warm()
expect(warmCalls).toBe(1)
expect(hydrateAllCalls).toBe(0) // warm() ran — no hydrateAll fallback
expect(report.metadata.outcome).toBe('warmed')
await brain.close()
})
it('reports "unavailable" when the active metadata provider implements neither warm() nor hydrateAll() (the honest branch a native provider without either hook must hit)', async () => {
const brain = new Brainy({
requireSubtype: false,
storage: { type: 'memory' },
silent: true
})
await brain.init()
await brain.add({ data: 'a', type: NounType.Thing, vector: V(1) })
const metadataIndex = (brain as any).metadataIndex
// Shadow away BOTH optional hooks — models a genuinely native provider
// that (unlike the built-in JS manager) offers neither seam. This must
// never fall back to calling init() as a stand-in for warmth.
metadataIndex.warm = undefined
metadataIndex.hydrateAll = undefined
const report = await brain.warm()
expect(report.metadata.outcome).toBe('unavailable')
await brain.close()
})
it('the built-in JS manager (no warm()) still reports "warmed" via its existing hydrateAll() duck-type — unchanged by the new provider hook', async () => {
const brain = new Brainy({
requireSubtype: false,
storage: { type: 'memory' },
silent: true
})
await brain.init()
await brain.add({ data: 'a', type: NounType.Thing, vector: V(1) })
// No patching at all — the default built-in MetadataIndexManager has
// hydrateAll() but no warm(), exactly as it did before this change.
const metadataIndex = (brain as any).metadataIndex
expect(typeof metadataIndex.warm).not.toBe('function')
expect(typeof metadataIndex.hydrateAll).toBe('function')
const report = await brain.warm()
expect(report.metadata.outcome).toBe('warmed')
await brain.close()
})
})
feat: warm contract (warm/warmOnOpen/provider warm hook), configurable transact budget floor, backend-neutral vector index op names Three pieces addressing the cold-restart-write incident where a production deployment's first writes after every restart (33-35s each on a cold page cache) blew the op-count-scaled transact budget mid-batch: every write is itself a multi-op transaction, so one cold operation consumed the whole budget, the gate before the next operation tripped, and the write rolled back atomically - refused, retried, and refused again until the page cache warmed passively. - The budget's start-gating contract is now explicit and pinned: it gates STARTING the next operation, never rolling back completed work for elapsed time (the shipped schedule since 8.7.0, now stated in contract JSDoc, guarded by code for operation 0, and enforced by regression tests). The 30s floor is configurable via transactionBudgetFloorMs for stores whose cold operations legitimately run long. - New brain.warm() eagerly loads the vector index, metadata index, and graph adjacency so first operations after a cold restart run at steady-state cost. Returns a WarmReport with an honest per-surface outcome (warmed / probed / unavailable) - never reports a probe as a warm. warmOnOpen: true runs it during init(). New optional provider hook warm() on the vector and graph plugin contracts. - Vector-index transaction op classes renamed from the backend-specific AddToHNSWOperation / RemoveFromHNSWOperation to backend-neutral AddToVectorIndexOperation / RemoveFromVectorIndexOperation, stamping the active backend into the emitted op-name string (AddToVectorIndex(js-hnsw) vs a native provider's own identity) so journals never misdirect an operator toward an index that isn't running.
2026-07-22 16:42:26 -07:00
})