[Bug]: Search reduce result is unexpected

[yhmo]

Is there an existing issue for this?

• I have searched the existing issues

Describe the bug

Create a collection.
Insert 3 entities with the same ID, but different vectors.
Search with the first vector, set output_fields=["id", "vector"]
Return one entity, the distance is 1, but the vector is not the first vector.

Use this script to test:

from pymilvus import (
    MilvusClient
)

client = MilvusClient(uri="http://localhost:19530")  # standalone

collection_name="demo_collection"
client.drop_collection(collection_name)
client.create_collection(collection_name=collection_name, dimension=4)

client.insert(collection_name=collection_name,data={"id": 5, "vector": [1,2,3,4]})
client.insert(collection_name=collection_name,data={"id": 5, "vector": [2,2,2,2]})
client.insert(collection_name=collection_name,data={"id": 5, "vector": [3,3,3,3]})

print(client.query(collection_name=collection_name, filter="", output_fields=["count(*)"], consistency_level="Strong"))

query_embeddings = [[1,2,3,4]]
result = client.search(
    collection_name=collection_name,
    data=query_embeddings,
    limit=6,
    output_fields=["id", "vector"],
)
print("Search results")
for hits in result:
    for hit in hits:
        print(f"hit: {hit}")

It returns:

data: ["{'count(*)': 3}"] 
Search results
hit: {'id': 5, 'distance': 0.9999999403953552, 'entity': {'vector': [3.0, 3.0, 3.0, 3.0], 'id': 5}}

The 'distance' is 0.9999999, but the 'vector' is [3.0, 3.0, 3.0, 3.0], I suppose it should return [1,2,3,4].

Expected Behavior

No response

Steps/Code To Reproduce behavior

No response

Environment details

- Hardware/Softward conditions (OS, CPU, GPU, Memory):
- Method of installation (Docker, or from source):
- Milvus version (v0.3.1, or v0.4.0): 2.4.9
- Milvus configuration (Settings you made in `server_config.yaml`):

Anything else?

No response

[LacNguyen-Vidoori]

sorry - please pardon the ignorance... Just curious, since I'm trying to put together a somewhat similar test case:

1. did you have Auto_id or some other attribute set as the primary key for this collection, that's why you were able to input same id value for 3 different entities?

2. could you share the config settings that would let you create/load a Collection without an index specified?

3. so it did return all 3 results (since (data: ["{'count(*)': 3}"])), it's just that you had expected [1,2,3,4] to rank 1st? Maybe because of the 'Strong' setting that always return the latest data? (your scripts had [3,3,3,3] inserted last)

sorry again if these questions seem dumb... i'm a noob at this and trying to learn as much as i can. Thank you.