向量数据库选型：Milvus 2.4与Qdrant 1.9在图像检索场景下的实测对比

# docker-compose.yml (Milvus 2.4.1关键配置)
version: '3.5'
services:
  etcd:
    container_name: milvus-etcd
    image: quay.io/coreos/etcd:v3.5.5
    environment:
      - ETCD_AUTO_COMPACTION_MODE=revision
      - ETCD_AUTO_COMPACTION_RETENTION=1000
      - ETCD_QUOTA_BACKEND_BYTES=4294967296
    volumes:
      - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/etcd:/etcd
    command: etcd -advertise-client-urls=http://etcd:2379 -listen-client-urls http://0.0.0.0:2379 --data-dir /etcd

  minio:
    container_name: milvus-minio
    image: minio/minio:RELEASE.2023-03-20T20-16-18Z
    environment:
      MINIO_ACCESS_KEY: minioadmin
      MINIO_SECRET_KEY: minioadmin
    command: minio server /minio_data --console-address ":9001"
    volumes:
      - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/minio:/minio_data

  milvus:
    container_name: milvus-standalone
    image: milvusdb/milvus:v2.4.1
    command: ["milvus", "run", "standalone"]
    environment:
      ETCD_ENDPOINTS: etcd:2379
      MINIO_ADDRESS: minio:9000
    ports:
      - "19530:19530"
      - "9091:9091"
    volumes:
      - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/milvus:/var/lib/milvus
    depends_on:
      - etcd
      - minio

# pymilvus 2.4.3 建索引与写入
from pymilvus import (
    connections, Collection, CollectionSchema,
    FieldSchema, DataType, utility
)

connections.connect(host="localhost", port="19530")

fields = [
    FieldSchema(name="id", dtype=DataType.INT64, is_primary=True),
    FieldSchema(name="embedding", dtype=DataType.FLOAT_VECTOR, dim=256)
]
schema = CollectionSchema(fields, description="product image embedding")
col = Collection("product_embedding", schema)

# 创建HNSW索引，注意metric_type必须是IP，因为我们做了归一化
index_params = {
    "index_type": "HNSW",
    "metric_type": "IP",
    "params": {"M": 16, "efConstruction": 200}
}
col.create_index("embedding", index_params)

# 批量写入500万条
import random
import time

batch_size = 10000
total = 5000000
start = time.time()
for i in range(0, total, batch_size):
    ids = list(range(i, i + batch_size))
    vectors = [[random.random() for _ in range(256)] for _ in range(batch_size)]
    col.insert([ids, vectors])
    if i % 500000 == 0:
        print(f"inserted {i} rows, elapsed {time.time()-start:.2f}s")
col.flush()
print(f"Total insert time: {time.time()-start:.2f}s")

# Qdrant 1.9.2 Docker部署
docker run -d --name qdrant \
  -p 6333:6333 -p 6334:6334 \
  -v $(pwd)/qdrant_storage:/qdrant/storage \
  qdrant/qdrant:v1.9.2

# qdrant-client 1.9.1 批量写入
from qdrant_client import QdrantClient
from qdrant_client.models import VectorParams, Distance, PointStruct
import random, time

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

# 创建collection，设置HNSW参数
client.recreate_collection(
    collection_name="product_embedding",
    vectors_config=VectorParams(size=256, distance=Distance.DOT),
    hnsw_config={"m": 16, "ef_construct": 200}
)

# 批量写入，Qdrant的upsert支持分批
batch_size = 10000
total = 5000000
start = time.time()
for i in range(0, total, batch_size):
    points = [
        PointStruct(
            id=j,
            vector=[random.random() for _ in range(256)]
        ) for j in range(i, i + batch_size)
    ]
    client.upsert(
        collection_name="product_embedding",
        points=points,
        wait=False  # 非阻塞写入，性能更高
    )
    if i % 500000 == 0:
        print(f"upserted {i} rows, elapsed {time.time()-start:.2f}s")

# 等待所有写入完成
client.update_collection(collection_name="product_embedding")
print(f"Total upsert time: {time.time()-start:.2f}s")