feat(nbstore): add indexer storage (#10953)

This commit is contained in:
EYHN
2025-03-31 12:59:51 +00:00
parent c9e14ac0db
commit 8957d0645f
82 changed files with 3393 additions and 4753 deletions
@@ -3,15 +3,26 @@ import { IndexedDBBlobStorage } from './blob';
import { IndexedDBBlobSyncStorage } from './blob-sync';
import { IndexedDBDocStorage } from './doc';
import { IndexedDBDocSyncStorage } from './doc-sync';
import { IndexedDBIndexerStorage } from './indexer';
import { IndexedDBIndexerSyncStorage } from './indexer-sync';
export * from './blob';
export * from './blob-sync';
export * from './doc';
export * from './doc-sync';
export * from './indexer';
export * from './indexer-sync';
export const idbStorages = [
IndexedDBDocStorage,
IndexedDBBlobStorage,
IndexedDBDocSyncStorage,
IndexedDBBlobSyncStorage,
IndexedDBIndexerStorage,
IndexedDBIndexerSyncStorage,
] satisfies StorageConstructor[];
export const idbStoragesIndexerOnly = [
IndexedDBIndexerStorage,
IndexedDBIndexerSyncStorage,
] satisfies StorageConstructor[];
@@ -0,0 +1,38 @@
import { share } from '../../connection';
import type { DocClock } from '../../storage/doc';
import { IndexerSyncStorageBase } from '../../storage/indexer-sync';
import { IDBConnection, type IDBConnectionOptions } from './db';
export class IndexedDBIndexerSyncStorage extends IndexerSyncStorageBase {
static readonly identifier = 'IndexedDBIndexerSyncStorage';
readonly connection = share(new IDBConnection(this.options));
constructor(private readonly options: IDBConnectionOptions) {
super();
}
async getDocIndexedClock(docId: string): Promise<DocClock | null> {
const tx = this.connection.inner.db.transaction('indexerSync', 'readonly');
const store = tx.store;
const result = await store.get(docId);
return result
? { docId: result.docId, timestamp: result.indexedClock }
: null;
}
async setDocIndexedClock(docClock: DocClock): Promise<void> {
const tx = this.connection.inner.db.transaction('indexerSync', 'readwrite');
const store = tx.store;
await store.put({
docId: docClock.docId,
indexedClock: docClock.timestamp,
});
}
async clearDocIndexedClock(docId: string): Promise<void> {
const tx = this.connection.inner.db.transaction('indexerSync', 'readwrite');
const store = tx.store;
await store.delete(docId);
}
}
@@ -0,0 +1,10 @@
import { expect, test } from 'vitest';
import { bm25 } from '../bm25';
test('bm25', () => {
expect(bm25(1, 1, 10, 10, 15)).toEqual(3.2792079793859643);
expect(bm25(2, 1, 10, 10, 15) > bm25(1, 1, 10, 10, 15)).toBeTruthy();
expect(bm25(1, 1, 10, 10, 15) > bm25(2, 1, 10, 100, 15)).toBeTruthy();
expect(bm25(1, 1, 10, 10, 15) > bm25(1, 1, 10, 100, 15)).toBeTruthy();
});
@@ -0,0 +1,32 @@
import { expect, test } from 'vitest';
import { highlighter } from '../highlighter';
test('highlighter', () => {
expect(highlighter('0123456789', '<b>', '</b>', [[3, 5]])).toEqual(
'012<b>34</b>56789'
);
expect(
highlighter(
'012345678901234567890123456789012345678901234567890123456789',
'<b>',
'</b>',
[[59, 60]]
)
).toEqual('...0123456789012345678901234567890123456789012345678<b>9</b>');
expect(
highlighter(
'012345678901234567890123456789012345678901234567890123456789',
'<b>',
'</b>',
[
[10, 11],
[49, 51],
]
)
).toEqual(
'0123456789<b>0</b>12345678901234567890123456789012345678<b>9</b>...'
);
});
@@ -0,0 +1,128 @@
import { expect, test } from 'vitest';
import { GeneralTokenizer } from '../tokenizer';
test('tokenizer', () => {
{
const tokens = new GeneralTokenizer().tokenize('hello world,\n AFFiNE');
expect(tokens).toEqual([
{ term: 'hello', start: 0, end: 5 },
{ term: 'world', start: 7, end: 12 },
{ term: 'affine', start: 15, end: 21 },
]);
}
{
const tokens = new GeneralTokenizer().tokenize('你好世界,阿芬');
expect(tokens).toEqual([
{
end: 2,
start: 0,
term: '你好',
},
{
end: 3,
start: 1,
term: '好世',
},
{
end: 4,
start: 2,
term: '世界',
},
{
end: 7,
start: 5,
term: '阿芬',
},
]);
}
{
const tokens = new GeneralTokenizer().tokenize('1阿2芬');
expect(tokens).toEqual([
{ term: '1', start: 0, end: 1 },
{ term: '阿', start: 1, end: 2 },
{ term: '2', start: 2, end: 3 },
{ term: '芬', start: 3, end: 4 },
]);
}
{
const tokens = new GeneralTokenizer().tokenize('안녕하세요 세계');
expect(tokens).toEqual([
{
end: 2,
start: 0,
term: '안녕',
},
{
end: 3,
start: 1,
term: '녕하',
},
{
end: 4,
start: 2,
term: '하세',
},
{
end: 5,
start: 3,
term: '세요',
},
{
end: 8,
start: 6,
term: '세계',
},
]);
}
{
const tokens = new GeneralTokenizer().tokenize('ハローワールド');
expect(tokens).toEqual([
{ term: 'ハロ', start: 0, end: 2 },
{ term: 'ロー', start: 1, end: 3 },
{ term: 'ーワ', start: 2, end: 4 },
{ term: 'ワー', start: 3, end: 5 },
{ term: 'ール', start: 4, end: 6 },
{ term: 'ルド', start: 5, end: 7 },
]);
}
{
const tokens = new GeneralTokenizer().tokenize('はろーわーるど');
expect(tokens).toEqual([
{ term: 'はろ', start: 0, end: 2 },
{ term: 'ろー', start: 1, end: 3 },
{ term: 'ーわ', start: 2, end: 4 },
{ term: 'わー', start: 3, end: 5 },
{ term: 'ーる', start: 4, end: 6 },
{ term: 'るど', start: 5, end: 7 },
]);
}
{
const tokens = new GeneralTokenizer().tokenize('👋1️⃣🚪👋🏿');
expect(tokens).toEqual([
{ term: '👋', start: 0, end: 2 },
{ term: '1️⃣', start: 2, end: 5 },
{ term: '🚪', start: 5, end: 7 },
{ term: '👋🏿', start: 7, end: 11 },
]);
}
{
const tokens = new GeneralTokenizer().tokenize('1');
expect(tokens).toEqual([{ term: '1', start: 0, end: 2 }]);
}
});
@@ -0,0 +1,62 @@
/**
* Parameters of the BM25+ scoring algorithm. Customizing these is almost never
* necessary, and finetuning them requires an understanding of the BM25 scoring
* model.
*
* Some information about BM25 (and BM25+) can be found at these links:
*
* - https://en.wikipedia.org/wiki/Okapi_BM25
* - https://opensourceconnections.com/blog/2015/10/16/bm25-the-next-generation-of-lucene-relevation/
*/
export type BM25Params = {
/** Term frequency saturation point.
*
* Recommended values are between `1.2` and `2`. Higher values increase the
* difference in score between documents with higher and lower term
* frequencies. Setting this to `0` or a negative value is invalid. Defaults
* to `1.2`
*/
k: number;
/**
* Length normalization impact.
*
* Recommended values are around `0.75`. Higher values increase the weight
* that field length has on scoring. Setting this to `0` (not recommended)
* means that the field length has no effect on scoring. Negative values are
* invalid. Defaults to `0.7`.
*/
b: number;
/**
* BM25+ frequency normalization lower bound (usually called δ).
*
* Recommended values are between `0.5` and `1`. Increasing this parameter
* increases the minimum relevance of one occurrence of a search term
* regardless of its (possibly very long) field length. Negative values are
* invalid. Defaults to `0.5`.
*/
d: number;
};
const defaultBM25params: BM25Params = { k: 1.2, b: 0.7, d: 0.5 };
export const bm25 = (
termFreq: number,
matchingCount: number,
totalCount: number,
fieldLength: number,
avgFieldLength: number,
bm25params: BM25Params = defaultBM25params
): number => {
const { k, b, d } = bm25params;
const invDocFreq = Math.log(
1 + (totalCount - matchingCount + 0.5) / (matchingCount + 0.5)
);
return (
invDocFreq *
(d +
(termFreq * (k + 1)) /
(termFreq + k * (1 - b + b * (fieldLength / avgFieldLength))))
);
};
@@ -0,0 +1,493 @@
import { type IDBPDatabase, type IDBPTransaction, type StoreNames } from 'idb';
import {
type AggregateOptions,
type AggregateResult,
type IndexerDocument,
type IndexerFieldSchema,
IndexerSchema,
type Query,
type SearchOptions,
type SearchResult,
} from '../../../storage';
import type { DocStorageSchema } from '../schema';
import { highlighter } from './highlighter';
import {
BooleanInvertedIndex,
FullTextInvertedIndex,
IntegerInvertedIndex,
type InvertedIndex,
StringInvertedIndex,
} from './inverted-index';
import { Match } from './match';
export type DataStructRWTransaction = IDBPTransaction<
DocStorageSchema,
ArrayLike<StoreNames<DocStorageSchema>>,
'readwrite'
>;
export type DataStructROTransaction = IDBPTransaction<
DocStorageSchema,
ArrayLike<StoreNames<DocStorageSchema>>,
'readonly' | 'readwrite'
>;
export class DataStruct {
database: IDBPDatabase<DocStorageSchema> = null as any;
invertedIndex = new Map<string, Map<string, InvertedIndex>>();
constructor() {
for (const [tableName, table] of Object.entries(IndexerSchema)) {
const tableInvertedIndex = new Map<string, InvertedIndex>();
for (const [fieldName, type] of Object.entries(table)) {
const typeInfo: IndexerFieldSchema =
typeof type === 'string' ? { type } : type;
if ('index' in typeInfo && typeInfo.index === false) {
// If index is false, we don't need to create an inverted index for this field.
continue;
}
if (typeInfo.type === 'String') {
tableInvertedIndex.set(
fieldName,
new StringInvertedIndex(tableName, fieldName)
);
} else if (typeInfo.type === 'Integer') {
tableInvertedIndex.set(
fieldName,
new IntegerInvertedIndex(tableName, fieldName)
);
} else if (typeInfo.type === 'FullText') {
tableInvertedIndex.set(
fieldName,
new FullTextInvertedIndex(tableName, fieldName)
);
} else if (typeInfo.type === 'Boolean') {
tableInvertedIndex.set(
fieldName,
new BooleanInvertedIndex(tableName, fieldName)
);
} else {
throw new Error(`Field type '${typeInfo.type}' not supported`);
}
}
this.invertedIndex.set(tableName, tableInvertedIndex);
}
}
private async update(
trx: DataStructRWTransaction,
table: keyof IndexerSchema,
document: IndexerDocument
) {
const existsNid = await trx
.objectStore('indexerRecords')
.index('id')
.getKey([table, document.id]);
const exists = existsNid
? await trx.objectStore('indexerRecords').get(existsNid)
: null;
if (!existsNid || !exists) {
// if not exists, return
return;
}
// delete exists one
await this.deleteByNid(trx, existsNid);
const dataMap = new Map([...exists.data, ...document.fields]); // merge exists data with new data
const nid = await trx
.objectStore('indexerRecords')
.put({ table, id: document.id, data: dataMap });
for (const [key, values] of dataMap) {
const type = IndexerSchema[table][
key as keyof IndexerSchema[typeof table]
] as IndexerFieldSchema;
if (!type) {
continue;
}
const typeInfo = typeof type === 'string' ? { type } : type;
if (typeInfo.index !== false) {
// If index is false, the field will not be indexed
const iidx = this.invertedIndex.get(table)?.get(key);
if (!iidx) {
continue;
}
await iidx.insert(trx, nid, values);
}
}
}
private async insert(
trx: DataStructRWTransaction,
table: keyof IndexerSchema,
document: IndexerDocument
) {
const existsNid = await trx
.objectStore('indexerRecords')
.index('id')
.getKey([table, document.id]);
if (existsNid) {
// delete exists one
await this.deleteByNid(trx, existsNid);
}
const dataMap = document.fields;
const nid = await trx
.objectStore('indexerRecords')
.put({ table, id: document.id, data: dataMap });
for (const [key, values] of dataMap) {
const type = IndexerSchema[table][
key as keyof IndexerSchema[typeof table]
] as IndexerFieldSchema;
if (!type) {
continue;
}
const typeInfo = typeof type === 'string' ? { type } : type;
if (typeInfo.index !== false) {
// If index is false, the field will not be indexed
const iidx = this.invertedIndex.get(table)?.get(key);
if (!iidx) {
continue;
}
await iidx.insert(trx, nid, values);
}
}
}
private async deleteByNid(trx: DataStructRWTransaction, nid: number) {
await trx.objectStore('indexerRecords').delete(nid);
const indexIds = await trx
.objectStore('invertedIndex')
.index('nid')
.getAllKeys(nid);
for (const indexId of indexIds) {
await trx.objectStore('invertedIndex').delete(indexId);
}
}
private async delete(
trx: DataStructRWTransaction,
table: keyof IndexerSchema,
id: string
) {
const nid = await trx
.objectStore('indexerRecords')
.index('id')
.getKey([table, id]);
if (nid) {
await this.deleteByNid(trx, nid);
} else {
return;
}
}
async deleteByQuery(
trx: DataStructRWTransaction,
table: keyof IndexerSchema,
query: Query<any>
) {
const match = await this.queryRaw(trx, table, query);
for (const nid of match.scores.keys()) {
await this.deleteByNid(trx, nid);
}
}
async batchWrite(
trx: DataStructRWTransaction,
table: keyof IndexerSchema,
deleteByQueries: Query<any>[],
deletes: string[],
inserts: IndexerDocument<any>[],
updates: IndexerDocument<any>[]
) {
for (const query of deleteByQueries) {
await this.deleteByQuery(trx, table, query);
}
for (const del of deletes) {
await this.delete(trx, table, del);
}
for (const inst of inserts) {
await this.insert(trx, table, inst);
}
for (const update of updates) {
await this.update(trx, table, update);
}
}
async matchAll(
trx: DataStructROTransaction,
table: keyof IndexerSchema
): Promise<Match> {
const allNids = await trx
.objectStore('indexerRecords')
.index('table')
.getAllKeys(table);
const match = new Match();
for (const nid of allNids) {
match.addScore(nid, 1);
}
return match;
}
async queryRaw(
trx: DataStructROTransaction,
table: keyof IndexerSchema,
query: Query<any>
): Promise<Match> {
if (query.type === 'match') {
const iidx = this.invertedIndex.get(table)?.get(query.field as string);
if (!iidx) {
return new Match();
}
return await iidx.match(trx, query.match);
} else if (query.type === 'boolean') {
const weights = [];
for (const q of query.queries) {
weights.push(await this.queryRaw(trx, table, q));
}
if (query.occur === 'must') {
return weights.reduce((acc, w) => acc.and(w));
} else if (query.occur === 'must_not') {
const total = weights.reduce((acc, w) => acc.and(w));
return (await this.matchAll(trx, table)).exclude(total);
} else if (query.occur === 'should') {
return weights.reduce((acc, w) => acc.or(w));
}
} else if (query.type === 'all') {
return await this.matchAll(trx, table);
} else if (query.type === 'boost') {
return (await this.queryRaw(trx, table, query.query)).boost(query.boost);
} else if (query.type === 'exists') {
const iidx = this.invertedIndex.get(table)?.get(query.field as string);
if (!iidx) {
return new Match();
}
return await iidx.all(trx);
}
throw new Error(`Query type '${query.type}' not supported`);
}
async clear(trx: DataStructRWTransaction) {
await trx.objectStore('indexerRecords').clear();
await trx.objectStore('invertedIndex').clear();
await trx.objectStore('indexerMetadata').clear();
}
async search(
trx: DataStructROTransaction,
table: keyof IndexerSchema,
query: Query<any>,
options: SearchOptions<any> = {}
): Promise<SearchResult<any, any>> {
const pagination = {
skip: options.pagination?.skip ?? 0,
limit: options.pagination?.limit ?? 100,
};
const match = await this.queryRaw(trx, table, query);
const nids = match
.toArray()
.slice(pagination.skip, pagination.skip + pagination.limit);
const nodes = [];
for (const nid of nids) {
const record = await trx.objectStore('indexerRecords').get(nid);
if (!record) {
continue;
}
nodes.push(this.resultNode(record, options, match, nid));
}
return {
pagination: {
count: match.size(),
hasMore: match.size() > pagination.limit + pagination.skip,
limit: pagination.limit,
skip: pagination.skip,
},
nodes: nodes,
};
}
async aggregate(
trx: DataStructROTransaction,
table: keyof IndexerSchema,
query: Query<any>,
field: string,
options: AggregateOptions<any> = {}
): Promise<AggregateResult<any, any>> {
const pagination = {
skip: options.pagination?.skip ?? 0,
limit: options.pagination?.limit ?? 100,
};
const hitPagination = options.hits
? {
skip: options.hits.pagination?.skip ?? 0,
limit: options.hits.pagination?.limit ?? 3,
}
: { skip: 0, limit: 0 };
const match = await this.queryRaw(trx, table, query);
const nids = match.toArray();
const buckets: {
key: string;
nids: number[];
hits: SearchResult<any, any>['nodes'];
}[] = [];
for (const nid of nids) {
const record = await trx.objectStore('indexerRecords').get(nid);
if (!record) {
continue;
}
const values = record.data.get(field);
for (const value of values ?? []) {
let bucket;
let bucketIndex = buckets.findIndex(b => b.key === value);
if (bucketIndex === -1) {
bucket = { key: value, nids: [], hits: [] };
buckets.push(bucket);
bucketIndex = buckets.length - 1;
} else {
bucket = buckets[bucketIndex];
}
if (
bucketIndex >= pagination.skip &&
bucketIndex < pagination.skip + pagination.limit
) {
bucket.nids.push(nid);
if (
bucket.nids.length - 1 >= hitPagination.skip &&
bucket.nids.length - 1 < hitPagination.skip + hitPagination.limit
) {
bucket.hits.push(
this.resultNode(record, options.hits ?? {}, match, nid)
);
}
}
}
}
return {
buckets: buckets
.slice(pagination.skip, pagination.skip + pagination.limit)
.map(bucket => {
const result = {
key: bucket.key,
score: match.getScore(bucket.nids[0]),
count: bucket.nids.length,
} as AggregateResult<any, any>['buckets'][number];
if (options.hits) {
(result as any).hits = {
pagination: {
count: bucket.nids.length,
hasMore:
bucket.nids.length > hitPagination.limit + hitPagination.skip,
limit: hitPagination.limit,
skip: hitPagination.skip,
},
nodes: bucket.hits,
} as SearchResult<any, any>;
}
return result;
}),
pagination: {
count: buckets.length,
hasMore: buckets.length > pagination.limit + pagination.skip,
limit: pagination.limit,
skip: pagination.skip,
},
};
}
async readonly(database: IDBPDatabase<DocStorageSchema>) {
return database.transaction(
['indexerRecords', 'invertedIndex', 'indexerMetadata'],
'readonly',
{ durability: 'relaxed' }
);
}
async readwrite(database: IDBPDatabase<DocStorageSchema>) {
return database.transaction(
['indexerRecords', 'invertedIndex', 'indexerMetadata'],
'readwrite',
{ durability: 'relaxed' }
);
}
private resultNode(
record: { id: string; data: Map<string, string[]> },
options: SearchOptions<any>,
match?: Match,
nid?: number
): SearchResult<any, any>['nodes'][number] {
const node = {
id: record.id,
score: match && nid ? match.getScore(nid) : 1,
} as any;
if (options.fields) {
const fields = {} as Record<string, string | string[]>;
for (const field of options.fields as string[]) {
fields[field] = record.data.get(field) ?? [''];
if (fields[field].length === 1) {
fields[field] = fields[field][0];
}
}
node.fields = fields;
}
if (match && nid && options.highlights) {
const highlights = {} as Record<string, string[]>;
for (const { field, before, end } of options.highlights) {
const highlightValues = match.getHighlighters(nid, field);
if (highlightValues) {
const rawValues = record.data.get(field) ?? [];
highlights[field] = Array.from(highlightValues)
.map(([index, ranges]) => {
const raw = rawValues[index];
if (raw) {
return (
highlighter(raw, before, end, ranges, {
maxPrefix: 20,
maxLength: 50,
}) ?? ''
);
}
return '';
})
.filter(Boolean);
}
}
node.highlights = highlights;
}
return node;
}
}
@@ -0,0 +1,77 @@
export function highlighter(
originText: string,
before: string,
after: string,
matches: [number, number][],
{
maxLength = 50,
maxPrefix = 20,
}: { maxLength?: number; maxPrefix?: number } = {}
) {
const merged = mergeRanges(matches);
if (merged.length === 0) {
return null;
}
const firstMatch = merged[0][0];
const start = Math.max(
0,
Math.min(firstMatch - maxPrefix, originText.length - maxLength)
);
const end = Math.min(start + maxLength, originText.length);
const text = originText.substring(start, end);
let result = '';
let pointer = 0;
for (const match of merged) {
const matchStart = match[0] - start;
const matchEnd = match[1] - start;
if (matchStart >= text.length) {
break;
}
result += text.substring(pointer, matchStart);
pointer = matchStart;
const highlighted = text.substring(matchStart, matchEnd);
if (highlighted.length === 0) {
continue;
}
result += `${before}${highlighted}${after}`;
pointer = matchEnd;
}
result += text.substring(pointer);
if (start > 0) {
result = `...${result}`;
}
if (end < originText.length) {
result = `${result}...`;
}
return result;
}
function mergeRanges(intervals: [number, number][]) {
if (intervals.length === 0) return [];
intervals.sort((a, b) => a[0] - b[0]);
const merged = [intervals[0]];
for (let i = 1; i < intervals.length; i++) {
const last = merged[merged.length - 1];
const current = intervals[i];
if (current[0] <= last[1]) {
last[1] = Math.max(last[1], current[1]);
} else {
merged.push(current);
}
}
return merged;
}
@@ -0,0 +1,203 @@
import { merge, Observable, of, Subject, throttleTime } from 'rxjs';
import type {
AggregateOptions,
AggregateResult,
IndexerDocument,
IndexerSchema,
Query,
SearchOptions,
SearchResult,
} from '../../../storage';
import { IndexerStorageBase } from '../../../storage';
import { IDBConnection, type IDBConnectionOptions } from '../db';
import { DataStruct } from './data-struct';
import { backoffRetry, exhaustMapWithTrailing, fromPromise } from './utils';
export class IndexedDBIndexerStorage extends IndexerStorageBase {
static readonly identifier = 'IndexedDBIndexerStorage';
readonly connection = new IDBConnection(this.options);
override isReadonly = false;
private readonly data = new DataStruct();
private readonly tableUpdate$ = new Subject<string>();
/**
* The write operations of IndexedDBIndexerStorage are first cached in pendingUpdates,
* and then committed to IndexedDB in a batch through the refresh method.
*/
private readonly pendingUpdates: Record<
keyof IndexerSchema,
{
deleteByQueries: Query<any>[];
deletes: string[];
inserts: IndexerDocument[];
updates: IndexerDocument[];
}
> = {
doc: { deleteByQueries: [], deletes: [], inserts: [], updates: [] },
block: { deleteByQueries: [], deletes: [], inserts: [], updates: [] },
};
get channel() {
return this.connection.inner.channel;
}
get database() {
return this.connection.inner.db;
}
constructor(private readonly options: IDBConnectionOptions) {
super();
}
override async search<
T extends keyof IndexerSchema,
const O extends SearchOptions<T>,
>(table: T, query: Query<T>, options?: O): Promise<SearchResult<T, O>> {
const trx = await this.data.readonly(this.database);
return this.data.search(trx, table, query, options);
}
override async aggregate<
T extends keyof IndexerSchema,
const O extends AggregateOptions<T>,
>(
table: T,
query: Query<T>,
field: keyof IndexerSchema[T],
options?: O
): Promise<AggregateResult<T, O>> {
const trx = await this.data.readonly(this.database);
return this.data.aggregate(trx, table, query, field as string, options);
}
override search$<
T extends keyof IndexerSchema,
const O extends SearchOptions<T>,
>(table: T, query: Query<T>, options?: O): Observable<SearchResult<T, O>> {
return merge(of(1), this.watchTableUpdated(table)).pipe(
throttleTime(3000, undefined, { leading: true, trailing: true }),
exhaustMapWithTrailing(() => {
return fromPromise(async () => {
try {
const trx = await this.data.readonly(this.database);
return await this.data.search(trx, table, query, options);
} catch (error) {
console.error('search error', error);
throw error;
}
}).pipe(backoffRetry());
})
);
}
override aggregate$<
T extends keyof IndexerSchema,
const O extends AggregateOptions<T>,
>(
table: T,
query: Query<T>,
field: keyof IndexerSchema[T],
options?: O
): Observable<AggregateResult<T, O>> {
return merge(of(1), this.watchTableUpdated(table)).pipe(
throttleTime(3000, undefined, { leading: true, trailing: true }),
exhaustMapWithTrailing(() => {
return fromPromise(async () => {
try {
const trx = await this.data.readonly(this.database);
return await this.data.aggregate(
trx,
table,
query,
field as string,
options
);
} catch (error) {
console.error('aggregate error', error);
throw error;
}
}).pipe(backoffRetry());
})
);
}
override async deleteByQuery<T extends keyof IndexerSchema>(
table: T,
query: Query<T>
): Promise<void> {
this.pendingUpdates[table].deleteByQueries.push(query);
}
override insert<T extends keyof IndexerSchema>(
table: T,
document: IndexerDocument
): Promise<void> {
this.pendingUpdates[table].inserts.push(document);
return Promise.resolve();
}
override delete<T extends keyof IndexerSchema>(
table: T,
id: string
): Promise<void> {
this.pendingUpdates[table].deletes.push(id);
return Promise.resolve();
}
override update<T extends keyof IndexerSchema>(
table: T,
document: IndexerDocument
): Promise<void> {
this.pendingUpdates[table].updates.push(document);
return Promise.resolve();
}
override async refresh<T extends keyof IndexerSchema>(
table: T
): Promise<void> {
const trx = await this.data.readwrite(this.database);
const tables = table ? [table] : (['doc', 'block'] as const);
for (const table of tables) {
await this.data.batchWrite(
trx,
table,
this.pendingUpdates[table].deleteByQueries,
this.pendingUpdates[table].deletes,
this.pendingUpdates[table].inserts,
this.pendingUpdates[table].updates
);
this.pendingUpdates[table] = {
deleteByQueries: [],
deletes: [],
inserts: [],
updates: [],
};
}
this.emitTableUpdated(table);
}
private watchTableUpdated(table: keyof IndexerSchema) {
return new Observable(subscriber => {
const listener = (ev: MessageEvent) => {
if (ev.data.type === 'indexer-updated' && ev.data.table === table) {
subscriber.next(1);
}
};
const subscription = this.tableUpdate$.subscribe(updatedTable => {
if (updatedTable === table) {
subscriber.next(1);
}
});
this.channel.addEventListener('message', listener);
return () => {
this.channel.removeEventListener('message', listener);
subscription.unsubscribe();
};
});
}
emitTableUpdated(table: keyof IndexerSchema) {
this.tableUpdate$.next(table);
this.channel.postMessage({ type: 'indexer-updated', table });
}
}
@@ -0,0 +1,496 @@
import { bm25 } from './bm25';
import type {
DataStructROTransaction,
DataStructRWTransaction,
} from './data-struct';
import { Match } from './match';
import { GeneralTokenizer, type Token } from './tokenizer';
export interface InvertedIndex {
fieldKey: string;
match(trx: DataStructROTransaction, term: string): Promise<Match>;
all(trx: DataStructROTransaction): Promise<Match>;
insert(
trx: DataStructRWTransaction,
id: number,
terms: string[]
): Promise<void>;
}
export class StringInvertedIndex implements InvertedIndex {
constructor(
readonly table: string,
readonly fieldKey: string
) {}
async match(trx: DataStructROTransaction, term: string): Promise<Match> {
const objs = await trx
.objectStore('invertedIndex')
.index('key')
.getAll([
this.table,
InvertedIndexKey.forString(this.fieldKey, term).buffer(),
]);
const match = new Match();
for (const obj of objs) {
match.addScore(obj.nid, 1);
}
return match;
}
async all(trx: DataStructROTransaction): Promise<Match> {
const objs = await trx
.objectStore('invertedIndex')
.index('key')
.getAll(
IDBKeyRange.bound(
[this.table, InvertedIndexKey.forPrefix(this.fieldKey).buffer()],
[
this.table,
InvertedIndexKey.forPrefix(this.fieldKey).add1().buffer(),
]
)
);
const set = new Set<number>();
for (const obj of objs) {
set.add(obj.nid);
}
const match = new Match();
for (const nid of set) {
match.addScore(nid, 1);
}
return match;
}
async insert(trx: DataStructRWTransaction, id: number, terms: string[]) {
for (const term of terms) {
await trx.objectStore('invertedIndex').put({
table: this.table,
key: InvertedIndexKey.forString(this.fieldKey, term).buffer(),
nid: id,
});
}
}
}
export class IntegerInvertedIndex implements InvertedIndex {
constructor(
readonly table: string,
readonly fieldKey: string
) {}
async match(trx: DataStructROTransaction, term: string): Promise<Match> {
const objs = await trx
.objectStore('invertedIndex')
.index('key')
.getAll([
this.table,
InvertedIndexKey.forInt64(this.fieldKey, BigInt(term)).buffer(),
]);
const match = new Match();
for (const obj of objs) {
match.addScore(obj.nid, 1);
}
return match;
}
// eslint-disable-next-line sonarjs/no-identical-functions
async all(trx: DataStructROTransaction): Promise<Match> {
const objs = await trx
.objectStore('invertedIndex')
.index('key')
.getAll(
IDBKeyRange.bound(
[this.table, InvertedIndexKey.forPrefix(this.fieldKey).buffer()],
[
this.table,
InvertedIndexKey.forPrefix(this.fieldKey).add1().buffer(),
]
)
);
const set = new Set<number>();
for (const obj of objs) {
set.add(obj.nid);
}
const match = new Match();
for (const nid of set) {
match.addScore(nid, 1);
}
return match;
}
async insert(trx: DataStructRWTransaction, id: number, terms: string[]) {
for (const term of terms) {
await trx.objectStore('invertedIndex').put({
table: this.table,
key: InvertedIndexKey.forInt64(this.fieldKey, BigInt(term)).buffer(),
nid: id,
});
}
}
}
export class BooleanInvertedIndex implements InvertedIndex {
constructor(
readonly table: string,
readonly fieldKey: string
) {}
// eslint-disable-next-line sonarjs/no-identical-functions
async all(trx: DataStructROTransaction): Promise<Match> {
const objs = await trx
.objectStore('invertedIndex')
.index('key')
.getAll(
IDBKeyRange.bound(
[this.table, InvertedIndexKey.forPrefix(this.fieldKey).buffer()],
[
this.table,
InvertedIndexKey.forPrefix(this.fieldKey).add1().buffer(),
]
)
);
const set = new Set<number>();
for (const obj of objs) {
set.add(obj.nid);
}
const match = new Match();
for (const nid of set) {
match.addScore(nid, 1);
}
return match;
}
async match(trx: DataStructROTransaction, term: string): Promise<Match> {
const objs = await trx
.objectStore('invertedIndex')
.index('key')
.getAll([
this.table,
InvertedIndexKey.forBoolean(this.fieldKey, term === 'true').buffer(),
]);
const match = new Match();
for (const obj of objs) {
match.addScore(obj.nid, 1);
}
return match;
}
async insert(trx: DataStructRWTransaction, id: number, terms: string[]) {
for (const term of terms) {
await trx.objectStore('invertedIndex').put({
table: this.table,
key: InvertedIndexKey.forBoolean(
this.fieldKey,
term === 'true'
).buffer(),
nid: id,
});
}
}
}
export class FullTextInvertedIndex implements InvertedIndex {
constructor(
readonly table: string,
readonly fieldKey: string
) {}
async match(trx: DataStructROTransaction, term: string): Promise<Match> {
const queryTokens = new GeneralTokenizer().tokenize(term);
const matched = new Map<
number,
Map<
number, // index
{
score: number;
ranges: [number, number][];
}
>
>();
const avgFieldLength =
(
await trx
.objectStore('indexerMetadata')
.get(`full-text:avg-field-length:${this.table}:${this.fieldKey}`)
)?.value ?? 0;
for (const token of queryTokens) {
const key = InvertedIndexKey.forString(this.fieldKey, token.term);
const objs = await trx
.objectStore('invertedIndex')
.index('key')
.getAll(
IDBKeyRange.bound(
[this.table, key.buffer()],
[this.table, key.add1().buffer()],
false,
true
)
);
const submatched: {
nid: number;
score: number;
position: {
index: number;
ranges: [number, number][];
};
}[] = [];
for (const obj of objs) {
const key = InvertedIndexKey.fromBuffer(obj.key);
const originTokenTerm = key.asString();
const matchLength = token.term.length;
const position = obj.pos ?? {
i: 0,
l: 0,
rs: [],
};
const termFreq = position.rs.length;
const totalCount = objs.length;
const fieldLength = position.l;
const score =
bm25(termFreq, 1, totalCount, fieldLength, avgFieldLength) *
(matchLength / originTokenTerm.length);
const match = {
score,
positions: new Map(),
};
const ranges = match.positions.get(position.i) || [];
ranges.push(
...position.rs.map(([start, _end]) => [start, start + matchLength])
);
match.positions.set(position.i, ranges);
submatched.push({
nid: obj.nid,
score,
position: {
index: position.i,
ranges: position.rs.map(([start, _end]) => [
start,
start + matchLength,
]),
},
});
}
// normalize score
const maxScore = submatched.reduce((acc, s) => Math.max(acc, s.score), 0);
const minScore = submatched.reduce((acc, s) => Math.min(acc, s.score), 0);
for (const { nid, score, position } of submatched) {
const normalizedScore =
maxScore === minScore
? score
: (score - minScore) / (maxScore - minScore);
const match =
matched.get(nid) ??
new Map<
number, // index
{
score: number;
ranges: [number, number][];
}
>();
const item = match.get(position.index) || {
score: 0,
ranges: [],
};
item.score += normalizedScore;
item.ranges.push(...position.ranges);
match.set(position.index, item);
matched.set(nid, match);
}
}
const match = new Match();
for (const [nid, items] of matched) {
if (items.size === 0) {
break;
}
let highestScore = -1;
let highestIndex = -1;
let highestRanges: [number, number][] = [];
for (const [index, { score, ranges }] of items) {
if (score > highestScore) {
highestScore = score;
highestIndex = index;
highestRanges = ranges;
}
}
match.addScore(nid, highestScore);
match.addHighlighter(nid, this.fieldKey, highestIndex, highestRanges);
}
return match;
}
// eslint-disable-next-line sonarjs/no-identical-functions
async all(trx: DataStructROTransaction): Promise<Match> {
const objs = await trx
.objectStore('invertedIndex')
.index('key')
.getAll(
IDBKeyRange.bound(
[this.table, InvertedIndexKey.forPrefix(this.fieldKey).buffer()],
[
this.table,
InvertedIndexKey.forPrefix(this.fieldKey).add1().buffer(),
]
)
);
const set = new Set<number>();
for (const obj of objs) {
set.add(obj.nid);
}
const match = new Match();
for (const nid of set) {
match.addScore(nid, 1);
}
return match;
}
async insert(trx: DataStructRWTransaction, id: number, terms: string[]) {
for (let i = 0; i < terms.length; i++) {
const tokenMap = new Map<string, Token[]>();
const originString = terms[i];
const tokens = new GeneralTokenizer().tokenize(originString);
for (const token of tokens) {
const tokens = tokenMap.get(token.term) || [];
tokens.push(token);
tokenMap.set(token.term, tokens);
}
for (const [term, tokens] of tokenMap) {
await trx.objectStore('invertedIndex').put({
table: this.table,
key: InvertedIndexKey.forString(this.fieldKey, term).buffer(),
nid: id,
pos: {
l: originString.length,
i: i,
rs: tokens.map(token => [token.start, token.end]),
},
});
}
const indexerMetadataStore = trx.objectStore('indexerMetadata');
// update avg-field-length
const totalCount =
(
await indexerMetadataStore.get(
`full-text:field-count:${this.table}:${this.fieldKey}`
)
)?.value ?? 0;
const avgFieldLength =
(
await indexerMetadataStore.get(
`full-text:avg-field-length:${this.table}:${this.fieldKey}`
)
)?.value ?? 0;
await indexerMetadataStore.put({
key: `full-text:field-count:${this.table}:${this.fieldKey}`,
value: totalCount + 1,
});
await indexerMetadataStore.put({
key: `full-text:avg-field-length:${this.table}:${this.fieldKey}`,
value:
avgFieldLength +
(terms.reduce((acc, term) => acc + term.length, 0) - avgFieldLength) /
(totalCount + 1),
});
}
}
}
export class InvertedIndexKey {
constructor(
readonly field: Uint8Array,
readonly value: Uint8Array,
readonly gap: Uint8Array = new Uint8Array([58])
) {}
asString() {
return new TextDecoder().decode(this.value);
}
asInt64() {
return new DataView(this.value.buffer).getBigInt64(
0,
false
); /* big-endian */
}
add1() {
if (this.value.byteLength > 0) {
const bytes = new Uint8Array(this.value.slice(0));
let carry = 1;
for (let i = bytes.length - 1; i >= 0 && carry > 0; i--) {
const sum = bytes[i] + carry;
bytes[i] = sum % 256;
carry = sum >> 8;
}
return new InvertedIndexKey(this.field, bytes);
} else {
return new InvertedIndexKey(
this.field,
new Uint8Array(0),
new Uint8Array([59])
);
}
}
static forPrefix(field: string) {
return new InvertedIndexKey(
new TextEncoder().encode(field),
new Uint8Array(0)
);
}
static forString(field: string, value: string) {
return new InvertedIndexKey(
new TextEncoder().encode(field),
new TextEncoder().encode(value)
);
}
static forBoolean(field: string, value: boolean) {
const bytes = new Uint8Array(1);
bytes.set([value ? 1 : 0]);
return new InvertedIndexKey(new TextEncoder().encode(field), bytes);
}
static forInt64(field: string, value: bigint) {
const bytes = new Uint8Array(8);
new DataView(bytes.buffer).setBigInt64(0, value, false); /* big-endian */
return new InvertedIndexKey(new TextEncoder().encode(field), bytes);
}
buffer() {
const tmp = new Uint8Array(
this.field.byteLength + (this.value?.byteLength ?? 0) + 1
);
tmp.set(new Uint8Array(this.field), 0);
tmp.set(new Uint8Array(this.gap), this.field.byteLength);
if (this.value.byteLength > 0) {
tmp.set(new Uint8Array(this.value), this.field.byteLength + 1);
}
return tmp.buffer;
}
static fromBuffer(buffer: ArrayBuffer) {
const array = new Uint8Array(buffer);
const fieldLength = array.indexOf(58);
const field = array.slice(0, fieldLength);
const value = array.slice(fieldLength + 1);
return new InvertedIndexKey(field, value);
}
}
@@ -0,0 +1,105 @@
export class Match {
scores = new Map<number, number>();
/**
* nid -> field -> index(multi value field) -> [start, end][]
*/
highlighters = new Map<
number,
Map<string, Map<number, [number, number][]>>
>();
constructor() {}
size() {
return this.scores.size;
}
getScore(id: number) {
return this.scores.get(id) ?? 0;
}
addScore(id: number, score: number) {
const currentScore = this.scores.get(id) || 0;
this.scores.set(id, currentScore + score);
}
getHighlighters(id: number, field: string) {
return this.highlighters.get(id)?.get(field);
}
addHighlighter(
id: number,
field: string,
index: number,
newRanges: [number, number][]
) {
const fields =
this.highlighters.get(id) ||
new Map<string, Map<number, [number, number][]>>();
const values = fields.get(field) || new Map<number, [number, number][]>();
const ranges = values.get(index) || [];
ranges.push(...newRanges);
values.set(index, ranges);
fields.set(field, values);
this.highlighters.set(id, fields);
}
and(other: Match) {
const newWeight = new Match();
for (const [id, score] of this.scores) {
if (other.scores.has(id)) {
newWeight.addScore(id, score + (other.scores.get(id) ?? 0));
newWeight.copyExtData(this, id);
newWeight.copyExtData(other, id);
}
}
return newWeight;
}
or(other: Match) {
const newWeight = new Match();
for (const [id, score] of this.scores) {
newWeight.addScore(id, score);
newWeight.copyExtData(this, id);
}
for (const [id, score] of other.scores) {
newWeight.addScore(id, score);
newWeight.copyExtData(other, id);
}
return newWeight;
}
exclude(other: Match) {
const newWeight = new Match();
for (const [id, score] of this.scores) {
if (!other.scores.has(id)) {
newWeight.addScore(id, score);
newWeight.copyExtData(this, id);
}
}
return newWeight;
}
boost(boost: number) {
const newWeight = new Match();
for (const [id, score] of this.scores) {
newWeight.addScore(id, score * boost);
newWeight.copyExtData(this, id);
}
return newWeight;
}
toArray() {
return Array.from(this.scores.entries())
.sort((a, b) => b[1] - a[1])
.map(e => e[0]);
}
private copyExtData(from: Match, id: number) {
for (const [field, values] of from.highlighters.get(id) ?? []) {
for (const [index, ranges] of values) {
this.addHighlighter(id, field, index, ranges);
}
}
}
}
@@ -0,0 +1,162 @@
import Graphemer from 'graphemer';
export interface Tokenizer {
tokenize(text: string): Token[];
}
export interface Token {
term: string;
start: number;
end: number;
}
export class SimpleTokenizer implements Tokenizer {
tokenize(text: string): Token[] {
const tokens: Token[] = [];
let start = 0;
let end = 0;
let inWord = false;
for (let i = 0; i < text.length; i++) {
const c = text[i];
if (c.match(/[\n\r\p{Z}\p{P}]/u)) {
if (inWord) {
end = i;
tokens.push({
term: text.substring(start, end).toLowerCase(),
start,
end,
});
inWord = false;
}
} else {
if (!inWord) {
start = i;
end = i;
inWord = true;
}
}
}
if (inWord) {
tokens.push({
term: text.substring(start).toLowerCase(),
start,
end: text.length,
});
}
return tokens;
}
}
export class NGramTokenizer implements Tokenizer {
constructor(private readonly n: number) {}
tokenize(text: string): Token[] {
const splitted: Token[] = [];
for (let i = 0; i < text.length; ) {
const nextBreak = Graphemer.nextBreak(text, i);
const c = text.substring(i, nextBreak);
splitted.push({
term: c,
start: i,
end: nextBreak,
});
i = nextBreak;
}
const tokens: Token[] = [];
for (let i = 0; i < splitted.length - this.n + 1; i++) {
tokens.push(
splitted.slice(i, i + this.n).reduce(
(acc, t) => ({
term: acc.term + t.term,
start: Math.min(acc.start, t.start),
end: Math.max(acc.end, t.end),
}),
{ term: '', start: Infinity, end: -Infinity }
)
);
}
return tokens;
}
}
export class GeneralTokenizer implements Tokenizer {
constructor() {}
tokenizeWord(word: string, lang: string): Token[] {
if (lang === 'en') {
return [{ term: word.toLowerCase(), start: 0, end: word.length }];
} else if (lang === 'cjk') {
if (word.length < 3) {
return [{ term: word, start: 0, end: word.length }];
}
return new NGramTokenizer(2).tokenize(word);
} else if (lang === 'emoji') {
return new NGramTokenizer(1).tokenize(word);
} else if (lang === '-') {
return [];
}
throw new Error('Not implemented');
}
testLang(c: string): string {
if (c.match(/[\p{Emoji}]/u)) {
return 'emoji';
} else if (c.match(/[\p{sc=Han}\p{scx=Hira}\p{scx=Kana}\p{sc=Hang}]/u)) {
return 'cjk';
} else if (c.match(/[\n\r\p{Z}\p{P}]/u)) {
return '-';
} else {
return 'en';
}
}
tokenize(text: string): Token[] {
const tokens: Token[] = [];
let start = 0;
let end = 0;
let lang: string | null = null;
for (let i = 0; i < text.length; ) {
const nextBreak = Graphemer.nextBreak(text, i);
const c = text.substring(i, nextBreak);
const l = this.testLang(c);
if (lang !== l) {
if (lang !== null) {
end = i;
tokens.push(
...this.tokenizeWord(text.substring(start, end), lang).map(
token => ({
...token,
start: token.start + start,
end: token.end + start,
})
)
);
}
start = i;
end = i;
lang = l;
}
i = nextBreak;
}
if (lang !== null) {
tokens.push(
...this.tokenizeWord(text.substring(start, text.length), lang).map(
token => ({
...token,
start: token.start + start,
end: token.end + start,
})
)
);
}
return tokens;
}
}
@@ -0,0 +1,104 @@
import {
asyncScheduler,
defer,
exhaustMap,
finalize,
Observable,
type ObservableInput,
type OperatorFunction,
retry,
scheduled,
Subject,
throttle,
throwError,
timer,
} from 'rxjs';
import { MANUALLY_STOP } from '../../../utils/throw-if-aborted';
/**
* Like exhaustMap, but also includes the trailing value emitted from the source observable while waiting for the preceding inner observable to complete
*
* Original code adapted from https://github.com/ReactiveX/rxjs/issues/5004
* @param {function<T, K>(value: T, ?index: number): ObservableInput<K>} project - A function that, when applied to an item emitted by the
* source Observable, returns a projected Observable.
*/
export function exhaustMapWithTrailing<T, R>(
project: (value: T, index: number) => ObservableInput<R>
): OperatorFunction<T, R> {
return (source$): Observable<R> =>
defer(() => {
const release$ = new Subject<void>();
return source$.pipe(
throttle(() => release$, {
leading: true,
trailing: true,
}),
exhaustMap((value, index) =>
scheduled(project(value, index), asyncScheduler).pipe(
finalize(() => {
release$.next();
})
)
)
);
});
}
/**
* Convert a promise to an observable.
*
* like `from` but support `AbortSignal`.
*/
export function fromPromise<T>(
promise: Promise<T> | ((signal: AbortSignal) => Promise<T>)
): Observable<T> {
return new Observable(subscriber => {
const abortController = new AbortController();
const rawPromise =
promise instanceof Function ? promise(abortController.signal) : promise;
rawPromise
.then(value => {
subscriber.next(value);
subscriber.complete();
})
.catch(error => {
subscriber.error(error);
});
return () => abortController.abort(MANUALLY_STOP);
});
}
/**
* An operator that retries the source observable when an error occurs.
*
* https://en.wikipedia.org/wiki/Exponential_backoff
*/
export function backoffRetry<T>({
when,
count = 3,
delay = 200,
maxDelay = 15000,
}: {
when?: (err: any) => boolean;
count?: number;
delay?: number;
maxDelay?: number;
} = {}) {
return (obs$: Observable<T>) =>
obs$.pipe(
retry({
count,
delay: (err, retryIndex) => {
if (when && !when(err)) {
return throwError(() => err);
}
const d = Math.pow(2, retryIndex - 1) * delay;
return timer(Math.min(d, maxDelay));
},
})
);
}
@@ -37,6 +37,11 @@ Table(PeerClocks)
|------|-------|-----------|-----------|
| str | str | Date | Date |
Table(IndexerSync)
| docId | clock |
|-------|-------|
| str | Date |
Table(BlobSync)
| peer | key | uploadedAt |
|------|-----|------------|
@@ -124,6 +129,43 @@ export interface DocStorageSchema extends DBSchema {
lock: Date;
};
};
indexerSync: {
key: string;
value: {
docId: string;
indexedClock: Date;
};
};
indexerMetadata: {
key: string;
value: {
key: string;
value: any;
};
};
indexerRecords: {
key: number;
value: {
table: string;
id: string;
data: Map<string, string[]>;
};
indexes: { table: string; id: [string, string] };
};
invertedIndex: {
key: number;
value: {
table: string;
nid: number;
pos?: {
i: number /* index */;
l: number /* length */;
rs: [number, number][] /* ranges: [start, end] */;
};
key: ArrayBuffer;
};
indexes: { key: [string, ArrayBuffer]; nid: number };
};
}
const migrate: OpenDBCallbacks<DocStorageSchema>['upgrade'] = (
@@ -199,11 +241,36 @@ const initBlobSync: Migrate = db => {
blobSync.createIndex('peer', 'peer', { unique: false });
};
const initIndexer: Migrate = db => {
db.createObjectStore('indexerMetadata', {
keyPath: 'key',
});
const indexRecordsStore = db.createObjectStore('indexerRecords', {
autoIncrement: true,
});
indexRecordsStore.createIndex('table', 'table', {
unique: false,
});
indexRecordsStore.createIndex('id', ['table', 'id'], {
unique: true,
});
const invertedIndexStore = db.createObjectStore('invertedIndex', {
autoIncrement: true,
});
invertedIndexStore.createIndex('key', ['table', 'key'], {
unique: false,
});
invertedIndexStore.createIndex('nid', 'nid', { unique: false });
db.createObjectStore('indexerSync', {
keyPath: 'docId',
autoIncrement: false,
});
};
// END REGION
// 1. all schema changed should be put in migrations
// 2. order matters
const migrations: Migrate[] = [init, initBlobSync];
const migrations: Migrate[] = [init, initBlobSync, initIndexer];
export const migrator = {
version: migrations.length,