feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
/**
|
|
|
|
|
* @module column-store.test
|
|
|
|
|
* @description Tests for the ColumnStore coordinator: full lifecycle including
|
|
|
|
|
* add, flush, reload, filter, sort, range query, and removal.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
import { describe, it, expect, beforeEach, afterEach } from 'vitest'
|
|
|
|
|
import { ColumnStore } from '../../../../src/indexes/columnStore/ColumnStore.js'
|
|
|
|
|
import { MemoryStorage } from '../../../../src/storage/adapters/memoryStorage.js'
|
|
|
|
|
import { EntityIdMapper } from '../../../../src/utils/entityIdMapper.js'
|
|
|
|
|
import { RoaringBitmap32 } from '../../../../src/utils/roaring/index.js'
|
|
|
|
|
|
|
|
|
|
describe('ColumnStore', () => {
|
|
|
|
|
let storage: MemoryStorage
|
|
|
|
|
let idMapper: EntityIdMapper
|
|
|
|
|
let store: ColumnStore
|
|
|
|
|
|
|
|
|
|
beforeEach(async () => {
|
|
|
|
|
storage = new MemoryStorage()
|
|
|
|
|
await storage.init()
|
|
|
|
|
idMapper = new EntityIdMapper({ storage, storageKey: 'test:idMapper' })
|
|
|
|
|
await idMapper.init()
|
|
|
|
|
|
|
|
|
|
store = new ColumnStore({ flushThreshold: 10 }) // low threshold for tests
|
|
|
|
|
await store.init(storage, idMapper)
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
afterEach(async () => {
|
|
|
|
|
await store.close()
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
// Basic add + sort
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
describe('sortTopK', () => {
|
|
|
|
|
it('returns entities sorted by numeric field ascending', async () => {
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('a')), { createdAt: 300 })
|
|
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('b')), { createdAt: 100 })
|
|
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('c')), { createdAt: 200 })
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
await store.flush()
|
|
|
|
|
|
|
|
|
|
const result = await store.sortTopK('createdAt', 'asc', 10)
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
const uuids = result.map(id => idMapper.getUuid(Number(id)))
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
expect(uuids).toEqual(['b', 'c', 'a'])
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
it('returns entities sorted by numeric field descending', async () => {
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('a')), { createdAt: 300 })
|
|
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('b')), { createdAt: 100 })
|
|
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('c')), { createdAt: 200 })
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
await store.flush()
|
|
|
|
|
|
|
|
|
|
const result = await store.sortTopK('createdAt', 'desc', 10)
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
const uuids = result.map(id => idMapper.getUuid(Number(id)))
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
expect(uuids).toEqual(['a', 'c', 'b'])
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
it('respects limit K', async () => {
|
|
|
|
|
for (let i = 0; i < 50; i++) {
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign(`e${i}`)), { createdAt: i * 1000 })
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
}
|
|
|
|
|
await store.flush()
|
|
|
|
|
|
|
|
|
|
const result = await store.sortTopK('createdAt', 'desc', 5)
|
|
|
|
|
expect(result).toHaveLength(5)
|
|
|
|
|
// Should be the 5 highest timestamps
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
const uuids = result.map(id => idMapper.getUuid(Number(id)))
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
expect(uuids).toEqual(['e49', 'e48', 'e47', 'e46', 'e45'])
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
it('sorts string fields lexicographically', async () => {
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('a')), { status: 'pending' })
|
|
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('b')), { status: 'active' })
|
|
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('c')), { status: 'inactive' })
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
await store.flush()
|
|
|
|
|
|
|
|
|
|
const result = await store.sortTopK('status', 'asc', 10)
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
const uuids = result.map(id => idMapper.getUuid(Number(id)))
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
expect(uuids).toEqual(['b', 'c', 'a']) // active, inactive, pending
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
it('returns empty for unknown field', async () => {
|
|
|
|
|
const result = await store.sortTopK('nonexistent', 'asc', 10)
|
|
|
|
|
expect(result).toEqual([])
|
|
|
|
|
})
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
// Filtered sort
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
describe('filteredSortTopK', () => {
|
|
|
|
|
it('returns only entities in the filter bitmap, sorted', async () => {
|
|
|
|
|
const idA = idMapper.getOrAssign('a')
|
|
|
|
|
const idB = idMapper.getOrAssign('b')
|
|
|
|
|
const idC = idMapper.getOrAssign('c')
|
|
|
|
|
const idD = idMapper.getOrAssign('d')
|
|
|
|
|
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
store.addEntity(BigInt(idA), { createdAt: 400 })
|
|
|
|
|
store.addEntity(BigInt(idB), { createdAt: 100 })
|
|
|
|
|
store.addEntity(BigInt(idC), { createdAt: 300 })
|
|
|
|
|
store.addEntity(BigInt(idD), { createdAt: 200 })
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
await store.flush()
|
|
|
|
|
|
|
|
|
|
// Filter: only B and C
|
|
|
|
|
const bitmap = new RoaringBitmap32([idB, idC])
|
|
|
|
|
const result = await store.filteredSortTopK(bitmap, 'createdAt', 'desc', 10)
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
const uuids = result.map(id => idMapper.getUuid(Number(id)))
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
expect(uuids).toEqual(['c', 'b']) // 300, 100
|
|
|
|
|
})
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
// Point filter
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
describe('filter', () => {
|
|
|
|
|
it('returns matching entities as roaring bitmap', async () => {
|
|
|
|
|
const idA = idMapper.getOrAssign('a')
|
|
|
|
|
const idB = idMapper.getOrAssign('b')
|
|
|
|
|
const idC = idMapper.getOrAssign('c')
|
|
|
|
|
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
store.addEntity(BigInt(idA), { status: 'active' })
|
|
|
|
|
store.addEntity(BigInt(idB), { status: 'inactive' })
|
|
|
|
|
store.addEntity(BigInt(idC), { status: 'active' })
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
await store.flush()
|
|
|
|
|
|
|
|
|
|
const bitmap = await store.filter('status', 'active')
|
|
|
|
|
expect(bitmap.has(idA)).toBe(true)
|
|
|
|
|
expect(bitmap.has(idC)).toBe(true)
|
|
|
|
|
expect(bitmap.has(idB)).toBe(false)
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
it('returns empty bitmap for no match', async () => {
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('a')), { status: 'active' })
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
await store.flush()
|
|
|
|
|
|
|
|
|
|
const bitmap = await store.filter('status', 'deleted')
|
|
|
|
|
expect(bitmap.size).toBe(0)
|
|
|
|
|
})
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
// Range query
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
describe('rangeQuery', () => {
|
|
|
|
|
it('returns entities within numeric range', async () => {
|
|
|
|
|
const ids: number[] = []
|
|
|
|
|
for (let i = 0; i < 10; i++) {
|
|
|
|
|
ids.push(idMapper.getOrAssign(`e${i}`))
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
store.addEntity(BigInt(ids[i]), { price: i * 10 })
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
}
|
|
|
|
|
await store.flush()
|
|
|
|
|
|
|
|
|
|
const bitmap = await store.rangeQuery('price', 30, 70)
|
|
|
|
|
// price 30, 40, 50, 60, 70 → entities e3, e4, e5, e6, e7
|
|
|
|
|
expect(bitmap.size).toBe(5)
|
|
|
|
|
expect(bitmap.has(ids[3])).toBe(true)
|
|
|
|
|
expect(bitmap.has(ids[7])).toBe(true)
|
|
|
|
|
expect(bitmap.has(ids[2])).toBe(false)
|
|
|
|
|
expect(bitmap.has(ids[8])).toBe(false)
|
|
|
|
|
})
|
fix(8.0): column-store range queries honor exclusive bounds (lessThan/greaterThan)
getIdsForRange computed includeMin/includeMax correctly for every operator
(gt→false/true, lt→true/false, between→true/true) but DROPPED them when the
column store served the query — it called columnStore.rangeQuery(field,min,max)
with no flags, and the column-store path was inclusive-only. So whenever a field
lived in the column store, strict lessThan/greaterThan silently behaved as
lte/gte. The sparse-fallback path already forwarded the flags, so this was
column-store-only.
Thread includeMin/includeMax end to end:
- ColumnSegmentCursor: add textbook lowerBound/upperBound helpers and express
binarySearchRange in terms of them. Inclusive/inclusive is byte-identical to
the old impl (lowerBound(lo) == old binarySearchValue(lo).index; upperBound(hi)
== old "first position after hi"). Exclusive lower advances past ALL duplicates
of the boundary value; exclusive upper stops before them.
- ColumnStore.rangeQuery: accept the flags, apply them in the segment cursor AND
the tail-buffer linear scan (strict > / < when exclusive). A bound taken from a
segment's own min/max stays inclusive — it is a real stored value.
- VectorIndex/types interface + getIdsForRange call site updated.
Surfaced by brainy-complete dual-bound test (year/popularity exclusive both ends
→ ['Express','Vue.js']; React@95 and Angular@75 correctly excluded). Added 5
column-store unit tests: exclusive lower, exclusive upper, both, duplicate
boundary values, and the unflushed tail-buffer path. 1469 unit green.
2026-06-19 11:01:41 -07:00
|
|
|
|
|
|
|
|
// Exclusive-bound coverage — regression guard for the column-store bug where
|
|
|
|
|
// includeMin/includeMax were dropped, so strict greaterThan/lessThan silently
|
|
|
|
|
// behaved as gte/lte whenever the column store served the query.
|
|
|
|
|
const seedPrices = async (): Promise<number[]> => {
|
|
|
|
|
const ids: number[] = []
|
|
|
|
|
for (let i = 0; i < 10; i++) {
|
|
|
|
|
ids.push(idMapper.getOrAssign(`x${i}`))
|
|
|
|
|
store.addEntity(BigInt(ids[i]), { price: i * 10 }) // 0,10,...,90
|
|
|
|
|
}
|
|
|
|
|
await store.flush()
|
|
|
|
|
return ids
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
it('honors exclusive lower bound — greaterThan (30, 70]', async () => {
|
|
|
|
|
const ids = await seedPrices()
|
|
|
|
|
const bm = await store.rangeQuery('price', 30, 70, false, true)
|
|
|
|
|
// 40,50,60,70 → e4..e7 (30 excluded)
|
|
|
|
|
expect(bm.size).toBe(4)
|
|
|
|
|
expect(bm.has(ids[3])).toBe(false)
|
|
|
|
|
expect(bm.has(ids[4])).toBe(true)
|
|
|
|
|
expect(bm.has(ids[7])).toBe(true)
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
it('honors exclusive upper bound — lessThan [30, 70)', async () => {
|
|
|
|
|
const ids = await seedPrices()
|
|
|
|
|
const bm = await store.rangeQuery('price', 30, 70, true, false)
|
|
|
|
|
// 30,40,50,60 → e3..e6 (70 excluded)
|
|
|
|
|
expect(bm.size).toBe(4)
|
|
|
|
|
expect(bm.has(ids[7])).toBe(false)
|
|
|
|
|
expect(bm.has(ids[3])).toBe(true)
|
|
|
|
|
expect(bm.has(ids[6])).toBe(true)
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
it('honors both exclusive bounds — (30, 70)', async () => {
|
|
|
|
|
const ids = await seedPrices()
|
|
|
|
|
const bm = await store.rangeQuery('price', 30, 70, false, false)
|
|
|
|
|
// 40,50,60 → e4..e6
|
|
|
|
|
expect(bm.size).toBe(3)
|
|
|
|
|
expect(bm.has(ids[3])).toBe(false)
|
|
|
|
|
expect(bm.has(ids[7])).toBe(false)
|
|
|
|
|
expect(bm.has(ids[4])).toBe(true)
|
|
|
|
|
expect(bm.has(ids[6])).toBe(true)
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
it('exclusive lower excludes ALL duplicates of the boundary value', async () => {
|
|
|
|
|
const a = idMapper.getOrAssign('dup_a'); store.addEntity(BigInt(a), { price: 30 })
|
|
|
|
|
const b = idMapper.getOrAssign('dup_b'); store.addEntity(BigInt(b), { price: 30 })
|
|
|
|
|
const c = idMapper.getOrAssign('dup_c'); store.addEntity(BigInt(c), { price: 30 })
|
|
|
|
|
const d = idMapper.getOrAssign('dup_d'); store.addEntity(BigInt(d), { price: 40 })
|
|
|
|
|
await store.flush()
|
|
|
|
|
const bm = await store.rangeQuery('price', 30, 100, false, true) // (30, 100]
|
|
|
|
|
expect(bm.has(a)).toBe(false)
|
|
|
|
|
expect(bm.has(b)).toBe(false)
|
|
|
|
|
expect(bm.has(c)).toBe(false)
|
|
|
|
|
expect(bm.has(d)).toBe(true)
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
it('honors exclusive bounds in the unflushed tail buffer', async () => {
|
|
|
|
|
// No flush → values live only in the tail buffer (separate code path from
|
|
|
|
|
// the persisted segment cursor).
|
|
|
|
|
const ids: number[] = []
|
|
|
|
|
for (let i = 0; i < 5; i++) {
|
|
|
|
|
ids.push(idMapper.getOrAssign(`tail${i}`))
|
|
|
|
|
store.addEntity(BigInt(ids[i]), { score: i * 10 }) // 0,10,20,30,40
|
|
|
|
|
}
|
|
|
|
|
const bm = await store.rangeQuery('score', 0, 40, false, false) // (0, 40)
|
|
|
|
|
expect(bm.size).toBe(3) // 10,20,30
|
|
|
|
|
expect(bm.has(ids[0])).toBe(false)
|
|
|
|
|
expect(bm.has(ids[4])).toBe(false)
|
|
|
|
|
expect(bm.has(ids[1])).toBe(true)
|
|
|
|
|
expect(bm.has(ids[3])).toBe(true)
|
|
|
|
|
})
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
})
|
|
|
|
|
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
// Removal
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
describe('removeEntity', () => {
|
|
|
|
|
it('excluded from sort after removal', async () => {
|
|
|
|
|
const idA = idMapper.getOrAssign('a')
|
|
|
|
|
const idB = idMapper.getOrAssign('b')
|
|
|
|
|
const idC = idMapper.getOrAssign('c')
|
|
|
|
|
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
store.addEntity(BigInt(idA), { createdAt: 100 })
|
|
|
|
|
store.addEntity(BigInt(idB), { createdAt: 200 })
|
|
|
|
|
store.addEntity(BigInt(idC), { createdAt: 300 })
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
await store.flush()
|
|
|
|
|
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
store.removeEntity(BigInt(idB))
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
await store.flush()
|
|
|
|
|
|
|
|
|
|
const result = await store.sortTopK('createdAt', 'asc', 10)
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
const uuids = result.map(id => idMapper.getUuid(Number(id)))
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
expect(uuids).toEqual(['a', 'c'])
|
|
|
|
|
})
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
// Multi-segment merge sort
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
describe('multi-segment', () => {
|
|
|
|
|
it('sorts correctly across multiple L0 segments', async () => {
|
|
|
|
|
// Flush threshold is 10, so 25 entities creates 3 segments
|
|
|
|
|
for (let i = 0; i < 25; i++) {
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign(`e${i}`)), { ts: i })
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
}
|
|
|
|
|
await store.flush()
|
|
|
|
|
|
|
|
|
|
const result = await store.sortTopK('ts', 'desc', 5)
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
const uuids = result.map(id => idMapper.getUuid(Number(id)))
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
expect(uuids).toEqual(['e24', 'e23', 'e22', 'e21', 'e20'])
|
|
|
|
|
})
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
// Persistence (flush + reload)
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
describe('persistence', () => {
|
|
|
|
|
it('survives close + reload', async () => {
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('a')), { createdAt: 300 })
|
|
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('b')), { createdAt: 100 })
|
|
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('c')), { createdAt: 200 })
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
await store.flush()
|
|
|
|
|
await store.close()
|
|
|
|
|
|
|
|
|
|
// Create a new store pointing at the same storage
|
|
|
|
|
const store2 = new ColumnStore({ flushThreshold: 10 })
|
|
|
|
|
await store2.init(storage, idMapper)
|
|
|
|
|
|
|
|
|
|
const result = await store2.sortTopK('createdAt', 'desc', 10)
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
const uuids = result.map(id => idMapper.getUuid(Number(id)))
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
expect(uuids).toEqual(['a', 'c', 'b'])
|
|
|
|
|
|
|
|
|
|
await store2.close()
|
|
|
|
|
})
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
// Multi-value fields (words)
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
describe('multi-value fields', () => {
|
|
|
|
|
it('indexes multiple values per entity', async () => {
|
|
|
|
|
const idA = idMapper.getOrAssign('docA')
|
|
|
|
|
const idB = idMapper.getOrAssign('docB')
|
|
|
|
|
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
store.addEntity(BigInt(idA), { __words__: ['machine', 'learning', 'algorithm'] })
|
|
|
|
|
store.addEntity(BigInt(idB), { __words__: ['neural', 'network', 'machine'] })
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
await store.flush()
|
|
|
|
|
|
|
|
|
|
// Point filter: "machine" should match both
|
|
|
|
|
const machineBitmap = await store.filter('__words__', 'machine')
|
|
|
|
|
expect(machineBitmap.has(idA)).toBe(true)
|
|
|
|
|
expect(machineBitmap.has(idB)).toBe(true)
|
|
|
|
|
|
|
|
|
|
// "algorithm" should match only docA
|
|
|
|
|
const algoBitmap = await store.filter('__words__', 'algorithm')
|
|
|
|
|
expect(algoBitmap.has(idA)).toBe(true)
|
|
|
|
|
expect(algoBitmap.has(idB)).toBe(false)
|
|
|
|
|
|
|
|
|
|
// AND intersection: "machine" AND "neural" → only docB
|
|
|
|
|
const neuralBitmap = await store.filter('__words__', 'neural')
|
|
|
|
|
const intersection = RoaringBitmap32.and(machineBitmap, neuralBitmap)
|
|
|
|
|
expect(intersection.has(idB)).toBe(true)
|
|
|
|
|
expect(intersection.has(idA)).toBe(false)
|
|
|
|
|
})
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
// getFilterValues
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
describe('getFilterValues', () => {
|
|
|
|
|
it('returns distinct values sorted', async () => {
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('a')), { status: 'pending' })
|
|
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('b')), { status: 'active' })
|
|
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('c')), { status: 'active' })
|
|
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('d')), { status: 'inactive' })
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
await store.flush()
|
|
|
|
|
|
|
|
|
|
const values = await store.getFilterValues('status')
|
|
|
|
|
expect(values).toEqual(['active', 'inactive', 'pending'])
|
|
|
|
|
})
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
// hasField
|
|
|
|
|
// -----------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
describe('hasField', () => {
|
|
|
|
|
it('returns false for unknown field', () => {
|
|
|
|
|
expect(store.hasField('nonexistent')).toBe(false)
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
it('returns true after adding data', async () => {
|
feat(8.0): u64 BigInt graph provider contract — punch list a-d,g,h
GraphIndexProvider now speaks BigInt at the boundary (D.2 mirror):
- getNeighbors/getVerbIdsBySource/getVerbIdsByTarget take entity ints and
return entity/verb ints as bigint[]
- new REQUIRED verbIntsToIds(bigint[]) batch reverse resolver (L.7
identity-fingerprint design — verb ids are UUIDs by contract, so the
provider-side interning is losslessly reversible)
- addVerb(verb, sourceInt, targetInt) returns the interned verb int;
removeVerb(verbId) joins the contract
Coordinator (brainy.ts) owns ALL UUID <-> int conversion: getOrAssign on
writes, getInt on reads (unmapped UUID -> empty result without calling the
provider), getUuid / verbIntsToIds on returns, plus a bounded ~100k-entry
insertion-order warm cache for verb-int -> verb-id pairs fed by addVerb
returns and resolver results. GraphVerb gains derived sourceInt/targetInt
(populated at add time, never persisted). findConnectedSubtype gains a
native fast path that routes single-type single-subtype outgoing BFS
through the provider when available.
JS GraphAdjacencyIndex satisfies the contract while staying string/u32-keyed
internally: entity ints resolve through the shared entity-id mapper (threaded
in by the coordinator on init/fork/checkout), verb ints come from an
in-process append-only interning map re-derived from storage on
rebuild/cold-start. ColumnStoreProvider widens the same way: addEntity/
removeEntity take bigint, sortTopK/filteredSortTopK return bigint[].
relate() now rejects a caller-supplied id with a teaching error — verb ids
are brainy-generated UUIDs by contract in 8.0 (previously a passed id was
silently ignored). No Roaring64 provider-boundary decode site exists yet;
the JS-internal column store stays Roaring32 and the Treemap decoder lands
with the first consumer of provider-returned filter buffers.
Public brain API unchanged. 1413 tests green (+10 new BigInt contract tests).
2026-06-10 10:45:45 -07:00
|
|
|
store.addEntity(BigInt(idMapper.getOrAssign('a')), { createdAt: 100 })
|
feat: unified column store for filtering + sorting at billion scale
Adds a per-field sorted column store (Lucene doc values + roaring bitmap
architecture) that replaces the MetadataIndex sparse index internals for
both filtering and sorting. One system for all field types with exact
precision — no bucketing, no per-entity storage reads.
Column store: binary .cidx segment format, in-memory tail buffers,
LSM-style compaction, k-way merge sort, multi-value support (__words__).
All queries (filter, range, sort, filtered sort) route through the column
store when data is available, falling back to sparse index otherwise.
Key unlocks:
- find({ orderBy: 'createdAt' }) works WITHOUT a filter (previously threw)
- find({ orderBy: 'metadata.price' }) works for custom numeric fields
- Exact timestamp precision (no 1-minute bucketing)
- O(K log S) sort independent of total entity count
New files: src/indexes/columnStore/ (types, format, tail buffer, cursor,
manifest, coordinator — ~700 lines). 101 new unit tests covering binary
format round-trips, CRC validation, sort, filter, range, deletion,
multi-segment merge, persistence, and multi-value (words) fields.
Deleted: metadataIndex-automatic-bucketing.test.ts (bucketing behavior
eliminated by exact-precision column store). Sparse index write path
removed from addToIndex/removeFromIndex. Sparse index legacy code still
present as dead code pending cleanup in next commit.
2026-04-10 11:22:19 -07:00
|
|
|
expect(store.hasField('createdAt')).toBe(true)
|
|
|
|
|
})
|
|
|
|
|
})
|
|
|
|
|
})
|