from typing import Optional, Sequence, Any, Tuple, cast, Generator, Union, Dict from chromadb.segment import MetadataReader from chromadb.ingest import Consumer from chromadb.config import System from chromadb.types import Segment from chromadb.db.impl.sqlite import SqliteDB from overrides import override from chromadb.db.base import ( Cursor, ParameterValue, get_sql, ) from chromadb.types import ( Where, WhereDocument, MetadataEmbeddingRecord, EmbeddingRecord, SeqId, Operation, UpdateMetadata, LiteralValue, WhereOperator, ) from uuid import UUID from pypika import Table, Tables from pypika.queries import QueryBuilder import pypika.functions as fn from pypika.terms import Criterion from itertools import islice, groupby from chromadb.config import Component from functools import reduce import sqlite3 import logging logger = logging.getLogger(__name__) class SqliteMetadataSegment(Component, MetadataReader): _consumer: Consumer _db: SqliteDB _id: UUID _topic: Optional[str] _subscription: Optional[UUID] def __init__(self, system: System, segment: Segment): self._db = system.instance(SqliteDB) self._consumer = system.instance(Consumer) self._id = segment["id"] self._topic = segment["topic"] @override def start(self) -> None: if self._topic: seq_id = self.max_seqid() self._subscription = self._consumer.subscribe( self._topic, self._write_metadata, start=seq_id ) @override def stop(self) -> None: if self._subscription: self._consumer.unsubscribe(self._subscription) @override def max_seqid(self) -> SeqId: t = Table("max_seq_id") q = ( self._db.querybuilder() .from_(t) .select(t.seq_id) .where(t.segment_id == ParameterValue(self._db.uuid_to_db(self._id))) ) sql, params = get_sql(q) with self._db.tx() as cur: result = cur.execute(sql, params).fetchone() if result is None: return self._consumer.min_seqid() else: return _decode_seq_id(result[0]) @override def count(self) -> int: embeddings_t = Table("embeddings") q = ( self._db.querybuilder() .from_(embeddings_t) .where( embeddings_t.segment_id == ParameterValue(self._db.uuid_to_db(self._id)) ) .select(fn.Count(embeddings_t.id)) ) sql, params = get_sql(q) with self._db.tx() as cur: result = cur.execute(sql, params).fetchone()[0] return cast(int, result) @override def get_metadata( self, where: Optional[Where] = None, where_document: Optional[WhereDocument] = None, ids: Optional[Sequence[str]] = None, limit: Optional[int] = None, offset: Optional[int] = None, ) -> Sequence[MetadataEmbeddingRecord]: """Query for embedding metadata.""" embeddings_t, metadata_t, fulltext_t = Tables( "embeddings", "embedding_metadata", "embedding_fulltext" ) q = ( ( self._db.querybuilder() .from_(embeddings_t) .left_join(metadata_t) .on(embeddings_t.id == metadata_t.id) ) .select( embeddings_t.id, embeddings_t.embedding_id, embeddings_t.seq_id, metadata_t.key, metadata_t.string_value, metadata_t.int_value, metadata_t.float_value, ) .where( embeddings_t.segment_id == ParameterValue(self._db.uuid_to_db(self._id)) ) .orderby(embeddings_t.id) ) if where: q = q.where(self._where_map_criterion(q, where, embeddings_t, metadata_t)) if where_document: q = q.where( self._where_doc_criterion(q, where_document, embeddings_t, fulltext_t) ) pass # q = self._where_document_query(q, where_document, embeddings_t, fulltext_t) if ids: q = q.where(embeddings_t.embedding_id.isin(ParameterValue(ids))) limit = limit or 2**63 - 1 offset = offset or 0 with self._db.tx() as cur: return list(islice(self._records(cur, q), offset, offset + limit)) def _records( self, cur: Cursor, q: QueryBuilder ) -> Generator[MetadataEmbeddingRecord, None, None]: """Given a cursor and a QueryBuilder, yield a generator of records. Assumes cursor returns rows in ID order.""" sql, params = get_sql(q) cur.execute(sql, params) cur_iterator = iter(cur.fetchone, None) group_iterator = groupby(cur_iterator, lambda r: int(r[0])) for _, group in group_iterator: yield self._record(list(group)) def _record(self, rows: Sequence[Tuple[Any, ...]]) -> MetadataEmbeddingRecord: """Given a list of DB rows with the same ID, construct a MetadataEmbeddingRecord""" _, embedding_id, seq_id = rows[0][:3] metadata = {} for row in rows: key, string_value, int_value, float_value = row[3:] if string_value is not None: metadata[key] = string_value elif int_value is not None: metadata[key] = int_value elif float_value is not None: metadata[key] = float_value return MetadataEmbeddingRecord( id=embedding_id, seq_id=_decode_seq_id(seq_id), metadata=metadata or None, ) def _insert_record( self, cur: Cursor, record: EmbeddingRecord, upsert: bool ) -> None: """Add or update a single EmbeddingRecord into the DB""" t = Table("embeddings") q = ( self._db.querybuilder() .into(t) .columns(t.segment_id, t.embedding_id, t.seq_id) .where(t.segment_id == ParameterValue(self._db.uuid_to_db(self._id))) .where(t.embedding_id == ParameterValue(record["id"])) ).insert( ParameterValue(self._db.uuid_to_db(self._id)), ParameterValue(record["id"]), ParameterValue(_encode_seq_id(record["seq_id"])), ) sql, params = get_sql(q) sql = sql + "RETURNING id" try: id = cur.execute(sql, params).fetchone()[0] except sqlite3.IntegrityError: # Can't use INSERT OR REPLACE here because it changes the primary key. if upsert: return self._update_record(cur, record) else: logger.warning(f"Insert of existing embedding ID: {record['id']}") if record["metadata"]: self._update_metadata(cur, id, record["metadata"]) def _update_metadata(self, cur: Cursor, id: int, metadata: UpdateMetadata) -> None: """Update the metadata for a single EmbeddingRecord""" t = Table("embedding_metadata") to_delete = [k for k, v in metadata.items() if v is None] if to_delete: q = ( self._db.querybuilder() .from_(t) .where(t.id == ParameterValue(id)) .where(t.key.isin(ParameterValue(to_delete))) .delete() ) sql, params = get_sql(q) cur.execute(sql, params) if "document" in metadata: t = Table("embedding_fulltext") q = ( self._db.querybuilder() .from_(t) .where(t.id == ParameterValue(id)) .delete() ) sql, params = get_sql(q) cur.execute(sql, params) self._insert_metadata(cur, id, metadata) def _insert_metadata(self, cur: Cursor, id: int, metadata: UpdateMetadata) -> None: """Insert or update each metadata row for a single embedding record""" t = Table("embedding_metadata") q = ( self._db.querybuilder() .into(t) .columns(t.id, t.key, t.string_value, t.int_value, t.float_value) ) for key, value in metadata.items(): if isinstance(value, str): q = q.insert( ParameterValue(id), ParameterValue(key), ParameterValue(value), None, None, ) elif isinstance(value, int): q = q.insert( ParameterValue(id), ParameterValue(key), None, ParameterValue(value), None, ) elif isinstance(value, float): q = q.insert( ParameterValue(id), ParameterValue(key), None, None, ParameterValue(value), ) sql, params = get_sql(q) sql = sql.replace("INSERT", "INSERT OR REPLACE") if sql: cur.execute(sql, params) if "document" in metadata: t = Table("embedding_fulltext") q = ( self._db.querybuilder() .into(t) .columns(t.id, t.string_value) .insert(ParameterValue(id), ParameterValue(metadata["document"])) ) sql, params = get_sql(q) cur.execute(sql, params) def _delete_record(self, cur: Cursor, record: EmbeddingRecord) -> None: """Delete a single EmbeddingRecord from the DB""" t = Table("embeddings") q = ( self._db.querybuilder() .from_(t) .where(t.segment_id == ParameterValue(self._db.uuid_to_db(self._id))) .where(t.embedding_id == ParameterValue(record["id"])) .delete() ) sql, params = get_sql(q) sql = sql + " RETURNING id" result = cur.execute(sql, params).fetchone() if result is None: logger.warning(f"Delete of nonexisting embedding ID: {record['id']}") else: id = result[0] # Manually delete metadata; cannot use cascade because # that triggers on replace metadata_t = Table("embedding_metadata") q = ( self._db.querybuilder() .from_(metadata_t) .where(metadata_t.id == ParameterValue(id)) .delete() ) sql, params = get_sql(q) cur.execute(sql, params) def _update_record(self, cur: Cursor, record: EmbeddingRecord) -> None: """Update a single EmbeddingRecord in the DB""" t = Table("embeddings") q = ( self._db.querybuilder() .update(t) .set(t.seq_id, ParameterValue(_encode_seq_id(record["seq_id"]))) .where(t.segment_id == ParameterValue(self._db.uuid_to_db(self._id))) .where(t.embedding_id == ParameterValue(record["id"])) ) sql, params = get_sql(q) sql = sql + " RETURNING id" result = cur.execute(sql, params).fetchone() if result is None: logger.warning(f"Update of nonexisting embedding ID: {record['id']}") else: id = result[0] if record["metadata"]: self._update_metadata(cur, id, record["metadata"]) def _write_metadata(self, records: Sequence[EmbeddingRecord]) -> None: """Write embedding metadata to the database. Care should be taken to ensure records are append-only (that is, that seq-ids should increase monotonically)""" with self._db.tx() as cur: for record in records: q = ( self._db.querybuilder() .into(Table("max_seq_id")) .columns("segment_id", "seq_id") .insert( ParameterValue(self._db.uuid_to_db(self._id)), ParameterValue(_encode_seq_id(record["seq_id"])), ) ) sql, params = get_sql(q) sql = sql.replace("INSERT", "INSERT OR REPLACE") cur.execute(sql, params) if record["operation"] == Operation.ADD: self._insert_record(cur, record, False) elif record["operation"] == Operation.UPSERT: self._insert_record(cur, record, True) elif record["operation"] == Operation.DELETE: self._delete_record(cur, record) elif record["operation"] == Operation.UPDATE: self._update_record(cur, record) def _where_map_criterion( self, q: QueryBuilder, where: Where, embeddings_t: Table, metadata_t: Table ) -> Criterion: clause: list[Criterion] = [] for k, v in where.items(): if k == "$and": criteria = [ self._where_map_criterion(q, w, embeddings_t, metadata_t) for w in cast(Sequence[Where], v) ] clause.append(reduce(lambda x, y: x & y, criteria)) elif k == "$or": criteria = [ self._where_map_criterion(q, w, embeddings_t, metadata_t) for w in cast(Sequence[Where], v) ] clause.append(reduce(lambda x, y: x | y, criteria)) else: expr = cast(Union[LiteralValue, Dict[WhereOperator, LiteralValue]], v) sq = ( self._db.querybuilder() .from_(metadata_t) .select(metadata_t.id) .where(metadata_t.key == ParameterValue(k)) .where(_where_clause(expr, metadata_t)) ) clause.append(embeddings_t.id.isin(sq)) return reduce(lambda x, y: x & y, clause) def _where_doc_criterion( self, q: QueryBuilder, where: WhereDocument, embeddings_t: Table, fulltext_t: Table, ) -> Criterion: for k, v in where.items(): if k == "$and": criteria = [ self._where_doc_criterion(q, w, embeddings_t, fulltext_t) for w in cast(Sequence[WhereDocument], v) ] return reduce(lambda x, y: x & y, criteria) elif k == "$or": criteria = [ self._where_doc_criterion(q, w, embeddings_t, fulltext_t) for w in cast(Sequence[WhereDocument], v) ] return reduce(lambda x, y: x | y, criteria) elif k == "$contains": search_term = f"%{v}%" sq = ( self._db.querybuilder() .from_(fulltext_t) .select(fulltext_t.id) .where(fulltext_t.string_value.like(ParameterValue(search_term))) ) return embeddings_t.id.isin(sq) else: raise ValueError(f"Unknown where_doc operator {k}") raise ValueError("Empty where_doc") def _encode_seq_id(seq_id: SeqId) -> bytes: """Encode a SeqID into a byte array""" if seq_id.bit_length() < 64: return int.to_bytes(seq_id, 8, "big") elif seq_id.bit_length() < 192: return int.to_bytes(seq_id, 24, "big") else: raise ValueError(f"Unsupported SeqID: {seq_id}") def _decode_seq_id(seq_id_bytes: bytes) -> SeqId: """Decode a byte array into a SeqID""" if len(seq_id_bytes) == 8: return int.from_bytes(seq_id_bytes, "big") elif len(seq_id_bytes) == 24: return int.from_bytes(seq_id_bytes, "big") else: raise ValueError(f"Unknown SeqID type with length {len(seq_id_bytes)}") def _where_clause( expr: Union[LiteralValue, Dict[WhereOperator, LiteralValue]], table: Table, ) -> Criterion: """Given a field name, an expression, and a table, construct a Pypika Criterion""" # Literal value case if isinstance(expr, (str, int, float)): return _where_clause({"$eq": expr}, table) # Operator dict case operator, value = next(iter(expr.items())) return _value_criterion(value, operator, table) def _value_criterion(value: LiteralValue, op: WhereOperator, table: Table) -> Criterion: """Return a criterion to compare a value with the appropriate columns given its type and the operation type.""" if isinstance(value, str): cols = [table.string_value] elif isinstance(value, int) and op in ("$eq", "$ne"): cols = [table.int_value] elif isinstance(value, float) and op in ("$eq", "$ne"): cols = [table.float_value] else: cols = [table.int_value, table.float_value] if op == "$eq": col_exprs = [col == ParameterValue(value) for col in cols] elif op == "$ne": col_exprs = [col != ParameterValue(value) for col in cols] elif op == "$gt": col_exprs = [col > ParameterValue(value) for col in cols] elif op == "$gte": col_exprs = [col >= ParameterValue(value) for col in cols] elif op == "$lt": col_exprs = [col < ParameterValue(value) for col in cols] elif op == "$lte": col_exprs = [col <= ParameterValue(value) for col in cols] if op == "$ne": return reduce(lambda x, y: x & y, col_exprs) else: return reduce(lambda x, y: x | y, col_exprs)