fix(aggregation): the lifecycle cluster — flush stamps, behind-stamp catches up incrementally, the native rebuild finally gets invoked, deletes are never silently skipped
Some checks failed
CI / Node 22 (push) Has been cancelled
CI / Node 24 (push) Has been cancelled
CI / Bun (latest) (push) Has been cancelled

SELF-ENGINE-LIFECYCLE-SPRINT + BRAINY-PROD-LATENCY-TRIAD, the four asks:

(a) brain.flush() persists aggregation state stamped at the committed
    generation. The stamp used to advance only at close(), so a long-lived
    writer that flushes but never closes — the primary production shape —
    left every write window behind the stamp, and ANY unclean exit forced a
    whole-store backfill walk (per-entity work, measured >60s and
    door-starving on a 9k-row production brain) on the first stats call.

(b) BEHIND-stamp adoption becomes adopt + INCREMENTAL CATCH-UP: the exact
    missing window (stamp, committed] resolves its affected-id set from the
    fact log and reconciles each entity with time-travel before/after reads
    (asOf at both window bounds) through the same delta algebra the live
    hooks use — cost bounded by writes since the last flush, never store
    size, and exact under interleaving because reconciliation targets the
    FIXED window end while later writes chain through hooks. Oversized
    windows (>5000 affected) and unreadable windows demote to the announced
    rescan — never a silent partial serve.

(c) The native provider's parallel rebuildAggregate — on the contract since
    8.x but never invoked anywhere — is now the backfill walk's preferred
    door: one call per aggregate with source-matched entities, replacing
    the per-entity FFI stream.

(d) A delete whose before-image is unavailable can no longer SKIP the
    aggregation hook silently (counts drifted upward forever): both delete
    paths (remove() and transact) flag an exact rescan, loudly.

Pins: integration (flush stamp; unclean-exit reopen → exact counts through
an add + group-move + delete window with the walk spy proving ZERO
whole-store walks) + unit (provider rebuild invoked once with filtered
entities; flagAllForRescan; reconcile delta algebra). Gates: unit 1913/1913
· integration 760 · conformance 27/27.
This commit is contained in:
David Snelling 2026-08-05 15:49:12 -07:00
parent 607b6b56f2
commit 1dc861d299
5 changed files with 796 additions and 38 deletions

View file

@ -0,0 +1,134 @@
/**
* @module tests/unit/aggregation/aggregation-provider-rebuild
* @description Pins for SELF-ENGINE-LIFECYCLE-SPRINT asks (c) + (d):
* (c) the native provider's parallel `rebuildAggregate` on the provider
* contract since 8.x but NEVER invoked (the JS walk streamed per-entity
* FFI calls instead) is now the backfill walk's preferred door;
* (d) a write-path hook that cannot see its entity (before-image-less
* delete) flags an exact rescan LOUDLY instead of silently skipping the
* decrement (the skip let counts drift upward forever).
*/
import { describe, it, expect, vi } from 'vitest'
import { AggregationIndex } from '../../../src/aggregation/AggregationIndex.js'
import { NounType } from '../../../src/types/graphTypes.js'
import type { AggregationProvider, AggregateGroupState } from '../../../src/types/brainy.types.js'
const DEF = {
name: 'by_subtype',
source: { type: NounType.Document },
groupBy: ['system.subtype'] as string[],
metrics: { count: { op: 'count' as const } }
}
/** Minimal in-memory storage double for the index's persistence surface. */
function memStorage() {
const store = new Map<string, unknown>()
return {
saveMetadata: async (k: string, v: unknown) => void store.set(k, v),
getMetadata: async (k: string) => store.get(k) ?? null
} as never
}
function providerDouble(): AggregationProvider & { rebuildAggregate: ReturnType<typeof vi.fn> } {
return {
defineAggregate: vi.fn(),
removeAggregate: vi.fn(),
incrementalUpdate: vi.fn(() => []),
computeGroupKey: vi.fn(() => ({})),
rebuildAggregate: vi.fn((): Map<string, AggregateGroupState> => {
return new Map([
[
'system.subtype=invoice',
{
groupKey: { 'system.subtype': 'invoice' },
metrics: { count: { sum: 0, count: 2, min: Infinity, max: -Infinity, m2: 0 } }
} as AggregateGroupState
]
])
}),
queryAggregate: vi.fn(() => [])
} as never
}
describe('ask (c) — the native parallel rebuild is invoked, never dead code', () => {
it('rebuildWithProvider hands SOURCE-MATCHED entities to the provider once and swaps state in', () => {
const provider = providerDouble()
const index = new AggregationIndex(memStorage(), provider)
index.defineAggregate(DEF)
expect(index.hasProviderRebuild()).toBe(true)
const entities = [
{ type: NounType.Document, subtype: 'invoice', metadata: {} },
{ type: NounType.Document, subtype: 'invoice', metadata: {} },
// Source-filter mismatch: a different noun type must be filtered OUT
// before the provider sees the batch.
{ type: NounType.Person, subtype: 'invoice', metadata: {} }
]
const handled = index.rebuildWithProvider(DEF.name, entities)
expect(handled).toBe(true)
expect(provider.rebuildAggregate).toHaveBeenCalledTimes(1)
const [defArg, entArg] = provider.rebuildAggregate.mock.calls[0]
expect(defArg.name).toBe(DEF.name)
expect(entArg).toHaveLength(2)
// The rebuilt state serves — and the aggregate is no longer pending.
expect(index.getPendingBackfills()).not.toContain(DEF.name)
})
it('returns false without a provider rebuild — the caller streams the JS walk', () => {
const index = new AggregationIndex(memStorage())
index.defineAggregate(DEF)
expect(index.hasProviderRebuild()).toBe(false)
expect(index.rebuildWithProvider(DEF.name, [])).toBe(false)
})
})
describe('ask (d) — the before-image-less delete is LOUD, never a silent skip', () => {
it('flagAllForRescan puts every defined aggregate back on the backfill list', () => {
const index = new AggregationIndex(memStorage())
index.defineAggregate(DEF)
index.defineAggregate({ ...DEF, name: 'second' })
// Simulate settled state: nothing pending.
for (const n of index.getPendingBackfills()) {
index.beginBackfill(n)
index.finishBackfill(n)
}
expect(index.getPendingBackfills()).toEqual([])
index.flagAllForRescan('delete of X carried no before-image metadata')
expect(index.getPendingBackfills().sort()).toEqual(['by_subtype', 'second'])
})
})
describe('reconcileEntity — the exact delta algebra at the catch-up boundary', () => {
it('before-only removes, after-only adds, both reconciles a group move', () => {
const index = new AggregationIndex(memStorage())
index.defineAggregate(DEF)
for (const n of index.getPendingBackfills()) {
index.beginBackfill(n)
index.finishBackfill(n)
}
const doc = (subtype: string) => ({ type: NounType.Document, subtype, metadata: {} })
// Pre-window state, applied through the LIVE hooks (as adoption would
// have counted it): c and seed exist as drafts, x1 as an invoice.
index.onEntityAdded('c', doc('draft'))
index.onEntityAdded('seed', doc('draft'))
index.onEntityAdded('x1', doc('invoice'))
// The window's reconciliation: two adds, one group move, one delete.
index.reconcileEntity(DEF.name, 'a', null, doc('invoice'))
index.reconcileEntity(DEF.name, 'b', null, doc('invoice'))
index.reconcileEntity(DEF.name, 'c', doc('draft'), doc('invoice'))
index.reconcileEntity(DEF.name, 'seed', doc('draft'), null)
const rows = index.queryAggregate({ name: DEF.name })
const count = (st: string) =>
Number(rows.find(r => r.groupKey['system.subtype'] === st)?.metrics.count ?? 0)
expect(count('invoice')).toBe(4) // x1 + a + b + moved c
expect(count('draft')).toBe(0) // c moved out, seed deleted
})
})

View file

@ -0,0 +1,142 @@
/**
* @module tests/unit/utils/metadataIndex-nested-orderby
* @description THE NESTED-FIELD ADDRESSING PIN for ordered reads (the
* field-addressing law, dotted-path clause). The defect this keeps dead:
* `orderBy` on a nested user metadata field (dotted path, e.g.
* `orderBy: 'profile.score'` over `metadata: { profile: { score: 7 } }`)
* silently returned insertion order a no-op sort because the sort
* path's value resolution read flat bag keys only. The law: a dotted user
* address is either SERVED CORRECTLY (the batched resolver walks inside
* the bag) or REFUSED with a typed UnresolvableFieldError never a silent
* pass-through. Both spellings (`profile.score` / `metadata.profile.score`)
* are the same address; the filter side (`where: { 'profile.score': … }`)
* obeys the same law.
*/
import { describe, it, expect, beforeAll, afterAll } from 'vitest'
import { Brainy, UnresolvableFieldError } from '../../../src/index.js'
import { NounType } from '../../../src/types/graphTypes.js'
const ROWS = 30
describe('nested (dotted-path) user field orderBy — the field-addressing law', () => {
let brain: Brainy
/** id → nested score, for the rows that carry profile.score */
const scoreById = new Map<string, number>()
/** ids of the two rows WITHOUT a profile bag */
let noProfileIds: string[] = []
beforeAll(async () => {
brain = new Brainy({ storage: { type: 'memory' }, requireSubtype: false })
await brain.init()
for (let i = 0; i < ROWS; i++) {
// (i * 11) % 30 is a permutation of 0..29 (gcd(11,30)=1): every score
// distinct, insertion order maximally different from value order — a
// silent insertion-order pass-through cannot accidentally look sorted.
const score = (i * 11) % ROWS
const id = await brain.add({
data: `row ${i}`,
type: NounType.Document,
metadata: { profile: { score }, plain: i }
})
scoreById.set(id, score)
}
const a = await brain.add({
data: 'no-profile a',
type: NounType.Document,
metadata: { plain: 1000 }
})
const b = await brain.add({
data: 'no-profile b',
type: NounType.Document,
metadata: { plain: 1001 }
})
noProfileIds = [a, b].sort()
}, 120000)
afterAll(async () => {
await brain.close().catch(() => {})
})
/** Assert one complete ordered read against the sealed ordering contract. */
function assertOrdered(
rows: Array<{ id: string }>,
order: 'asc' | 'desc',
label: string
): void {
// Rows are NEVER dropped: all 30 scored + 2 profile-less rows come back.
expect(rows.length, `${label}: complete result`).toBe(ROWS + 2)
// Missing-value rows sort LAST in BOTH directions, ties by id ascending.
const lastTwo = rows.slice(-2).map((r) => r.id)
expect(lastTwo, `${label}: missing-value rows LAST, id asc`).toEqual(noProfileIds)
// The scored 30 are ordered by the NESTED value — the exact permutation,
// not insertion order.
const observed = rows.slice(0, ROWS).map((r) => scoreById.get(r.id))
const wanted = [...scoreById.values()].sort((x, y) =>
order === 'asc' ? x - y : y - x
)
expect(observed, `${label}: nested values in ${order} order`).toEqual(wanted)
}
it('orderBy: "profile.score" desc — served correctly, missing rows LAST (never a silent insertion-order no-op)', async () => {
const rows = await brain.find({
type: NounType.Document,
orderBy: 'profile.score',
order: 'desc',
limit: 40
})
assertOrdered(rows, 'desc', 'bare dotted, desc')
})
it('orderBy: "profile.score" asc — same law in the other direction', async () => {
const rows = await brain.find({
type: NounType.Document,
orderBy: 'profile.score',
order: 'asc',
limit: 40
})
assertOrdered(rows, 'asc', 'bare dotted, asc')
})
it('explicit spelling "metadata.profile.score" is the SAME address — identical result', async () => {
const bare = await brain.find({
type: NounType.Document,
orderBy: 'profile.score',
order: 'desc',
limit: 40
})
const explicit = await brain.find({
type: NounType.Document,
orderBy: 'metadata.profile.score',
order: 'desc',
limit: 40
})
assertOrdered(explicit, 'desc', 'metadata.-prefixed, desc')
expect(
explicit.map((r) => r.id),
'both spellings resolve to the identical ordered id sequence'
).toEqual(bare.map((r) => r.id))
})
it('a dotted path carried by NO entity REFUSES with UnresolvableFieldError — never a silent insertion-order return', async () => {
await expect(
brain.find({
type: NounType.Document,
orderBy: 'no.such.path',
order: 'desc',
limit: 40
})
).rejects.toThrow(UnresolvableFieldError)
})
it('dotted where: { "profile.score": 7 } finds exactly the right row — the filter side of the same law', async () => {
const wantedId = [...scoreById.entries()].find(([, s]) => s === 7)![0]
const rows = await brain.find({
type: NounType.Document,
where: { 'profile.score': 7 },
limit: 40
})
expect(rows.map((r) => r.id)).toEqual([wantedId])
})
})