[Bug]: The impact of the limit parameter on the ranking of results during search

[zmw132]

Is there an existing issue for this?

• I have searched the existing issues

Environment

- Milvus version: 2.5.1
- Deployment mode(standalone or cluster): standalone 
- MQ type(rocksmq, pulsar or kafka):    
- SDK version(e.g. pymilvus v2.0.0rc2): api
- OS(Ubuntu or CentOS): CentOS
- CPU/Memory: 
- GPU: 
- Others:

Current Behavior

Through the search interface, all other contents of the query body remain unchanged, only the limit is changed. Why is the result when limit=3 not among the top three results when limit=10? During this process, there were no operations on the collection

curl --request POST --url "http://xxx:19530/v2/vectordb/entities/search" --header "Authorization: Bearer root:xxx" --header "Content-Type: application/json"
-d '{
"annsField": "vector",
"collectionName": "test_collection",
"data": [[-0.05120890960097313,-0.028760679066181183,-0.008349430747330189,...]],
"dbName": "test_db",
"filter": "",
"limit": 10,
"outputFields": ["_id", "aaa", "bbb", "ccc", "ddd"]
}'

Expected Behavior

No response

Steps To Reproduce

Milvus Log

No response

Anything else?

No response

[dosubot]

Hey @zmw132! I'm here to help you with any bugs, questions, or contributions you have for Milvus. Let me know how I can assist you!

The behavior you're observing, where the top results differ when changing the limit parameter, can be attributed to how Milvus ranks and retrieves search results. The limit parameter specifies the maximum number of entities to return, but it doesn't guarantee that the top results for a smaller limit will always be included in the results for a larger limit. This is because the search algorithm may rank results differently based on the number of results requested.

When you set limit=3, Milvus returns the top three most similar entities based on the similarity scores. However, when you increase the limit to 10, the search algorithm may consider additional factors or perform additional computations that affect the ranking of results. This can lead to a different set of top results, as the algorithm optimizes for the larger set of results.

This behavior is typical in search systems where the ranking can be influenced by the number of results requested, especially in complex vector search scenarios. If consistent top results are critical, you might need to explore additional ranking strategies or configurations that ensure stability across different limits ^[1][2].

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Bug Report | Other}

[yanliang567]

/assign @liliu-z
/unassign

[yanliang567]

/assign @binbinlv
please keep an eye on this issue

[xiaofan-luan]

what is your index parameter and what is the result looks like?

1. maybe some backend handoff/compaction happens, changed the index strucutre.
2. maybe your ef search is too small, so increate limit actual change the backend search parameters.

Suggestion:
try to tune search parameters and see if it's help.
try to wait until flush and compaction happened and see if it's stil reproducible