Milvus 与 Qdrant 向量数据库选型对比：千万级向量场景部署、性能与资源实测

# 下载官方 compose 文件
wget https://github.com/milvus-io/milvus/releases/download/v2.4.9/milvus-standalone-docker-compose.yml -O docker-compose-milvus.yml

# 启动
docker compose -f docker-compose-milvus.yml up -d

# 检查
docker ps | grep milvus

docker run -d --name qdrant \
  -p 6333:6333 -p 6334:6334 \
  -v /data/qdrant_storage:/qdrant/storage \
  qdrant/qdrant:v1.9.2

# 检查
curl http://localhost:6333/collections

from pymilvus import (
    connections, Collection, CollectionSchema, FieldSchema, DataType, utility
)
import random, time

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

collection_name = "content_vec_milvus"
if utility.has_collection(collection_name):
    utility.drop_collection(collection_name)

fields = [
    FieldSchema(name="id", dtype=DataType.INT64, is_primary=True, auto_id=False),
    FieldSchema(name="category_id", dtype=DataType.INT64),
    FieldSchema(name="publish_ts", dtype=DataType.INT64),
    FieldSchema(name="embedding", dtype=DataType.FLOAT_VECTOR, dim=768),
]
schema = CollectionSchema(fields=fields, description="content vector")
collection = Collection(name=collection_name, schema=schema)

# 标量索引
collection.create_index("category_id", {"index_type": "INVERTED"})
collection.create_index("publish_ts", {"index_type": "STL_SORT"})

# 向量索引
collection.create_index(
    field_name="embedding",
    index_params={
        "index_type": "HNSW",
        "metric_type": "COSINE",
        "params": {"M": 16, "efConstruction": 200},
    },
)
collection.load()

# 批量写入 10 万条示例
batch_size = 10000
for batch in range(10):
    ids = [batch * batch_size + i for i in range(batch_size)]
    category_ids = [random.randint(1, 50) for _ in range(batch_size)]
    publish_ts = [int(time.time()) - random.randint(0, 86400 * 365) for _ in range(batch_size)]
    vectors = [[random.random() for _ in range(768)] for _ in range(batch_size)]
    collection.insert([ids, category_ids, publish_ts, vectors])
collection.flush()

# 查询
query_vec = [[random.random() for _ in range(768)]]
res = collection.search(
    data=query_vec,
    anns_field="embedding",
    param={"metric_type": "COSINE", "params": {"ef": 128}},
    limit=10,
    expr="category_id == 3 and publish_ts > 1700000000",
    output_fields=["category_id", "publish_ts"],
)
for hits in res:
    for hit in hits:
        print(hit.id, hit.distance, hit.entity.get("category_id"))

from qdrant_client import QdrantClient
from qdrant_client.models import (
    Distance, VectorParams, PointStruct, Filter,
    FieldCondition, MatchValue, Range, HnswConfigDiff
)
import random, time

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

client.recreate_collection(
    collection_name=collection_name,
    vectors_config=VectorParams(size=768, distance=Distance.COSINE),
    hnsw_config=HnswConfigDiff(m=16, ef_construct=200),
)

# payload 索引
client.create_payload_index(collection_name, "category_id", "integer")
client.create_payload_index(collection_name, "publish_ts", "integer")

# 批量写入
batch_size = 1000
points = []
for i in range(10000):
    points.append(PointStruct(
        id=i,
        vector=[random.random() for _ in range(768)],
        payload={
            "category_id": random.randint(1, 50),
            "publish_ts": int(time.time()) - random.randint(0, 86400 * 365),
        },
    ))
    if len(points) == batch_size:
        client.upsert(collection_name=collection_name, points=points)
        points = []
if points:
    client.upsert(collection_name=collection_name, points=points)

# 查询
query_vec = [random.random() for _ in range(768)]
res = client.search(
    collection_name=collection_name,
    query_vector=query_vec,
    query_filter=Filter(
        must=[
            FieldCondition(key="category_id", match=MatchValue(value=3)),
            FieldCondition(key="publish_ts", range=Range(gt=1700000000)),
        ]
    ),
    limit=10,
    search_params={"hnsw_ef": 128},
)
for point in res:
    print(point.id, point.score, point.payload)