vector-store

vector-store

Covers production use of Qdrant as a vector database: collection lifecycle, batch upserts, hybrid search, payload filtering, snapshot management, and multi-tenant patterns.

1) Collection lifecycle

from qdrant_client import QdrantClient
from qdrant_client.models import (
    Distance, VectorParams, OptimizersConfigDiff,
    HnswConfigDiff, PayloadSchemaType, QuantizationConfig,
    ScalarQuantizationConfig, ScalarType,
)

qdrant = QdrantClient(host="localhost", port=6333)

def create_collection(name: str, dim: int = 1536, on_disk: bool = True):
    qdrant.recreate_collection(
        collection_name=name,
        vectors_config=VectorParams(size=dim, distance=Distance.COSINE, on_disk=on_disk),
        hnsw_config=HnswConfigDiff(m=16, ef_construct=200, full_scan_threshold=10_000),
        optimizers_config=OptimizersConfigDiff(indexing_threshold=20_000),
        on_disk_payload=on_disk,
    )
    # Index frequently-filtered payload fields
    for field in ("domain", "type", "published_at"):
        qdrant.create_payload_index(name, field, PayloadSchemaType.KEYWORD)

def delete_collection(name: str):
    qdrant.delete_collection(name)

def collection_info(name: str) -> dict:
    info = qdrant.get_collection(name)
    return {
        "points": info.points_count,
        "vectors": info.vectors_count,
        "status": info.status,
        "segments": info.segments_count,
    }

2) Batch upsert patterns

from qdrant_client.models import PointStruct, UpdateStatus
import uuid, hashlib

def doc_id_to_uuid(doc_id: str) -> str:
    """Deterministic UUID from string ID — Qdrant accepts UUIDs."""
    return str(uuid.UUID(bytes=hashlib.md5(doc_id.encode()).digest()))

async def upsert_batch(
    collection: str,
    documents: list[dict],   # each: {"id": str, "text": str, "payload": dict}
    embed_fn,                # async (list[str]) -> list[list[float]]
    batch_size: int = 100,
) -> int:
    total = 0
    for i in range(0, len(documents), batch_size):
        batch = documents[i : i + batch_size]
        texts = [d["text"] for d in batch]
        vectors = await embed_fn(texts)
        points = [
            PointStruct(
                id=doc_id_to_uuid(d["id"]),
                vector=vec,
                payload={**d["payload"], "text": d["text"], "_doc_id": d["id"]},
            )
            for d, vec in zip(batch, vectors)
        ]
        result = qdrant.upsert(collection_name=collection, points=points, wait=True)
        assert result.status == UpdateStatus.COMPLETED
        total += len(points)
    return total