[Bug]: [Nightly] Hybrid search results are different from expected

[NicoYuan1986]

Is there an existing issue for this?

• I have searched the existing issues

Environment

- Milvus version: 34e5f99
- Deployment mode(standalone or cluster):cluster & standalone
- MQ type(rocksmq, pulsar or kafka):    
- SDK version(e.g. pymilvus v2.0.0rc2):
- OS(Ubuntu or CentOS): 
- CPU/Memory: 
- GPU: 
- Others:

Current Behavior

Hybrid search results are different from expected.

[pytest : test] ____ TestCollectionHybridSearchValid.test_hybrid_search_min_limit[IP-int64] ____
[pytest : test] [gw4] linux -- Python 3.8.17 /usr/local/bin/python3
[pytest : test] 
[pytest : test] self = <test_search.TestCollectionHybridSearchValid object at 0x7fd4b73f7820>
[pytest : test] primary_field = 'int64', metric_type = 'IP'
[pytest : test] 
[pytest : test]     @pytest.mark.tags(CaseLabel.L2)
[pytest : test]     @pytest.mark.parametrize("primary_field", [ct.default_int64_field_name, ct.default_string_field_name])
[pytest : test]     def test_hybrid_search_min_limit(self, primary_field, metric_type):
[pytest : test]         """
[pytest : test]         target: test hybrid search with minimum limit params
[pytest : test]         method: create connection, collection, insert and search
[pytest : test]         expected: hybrid search successfully with limit(topK)
[pytest : test]         """
[pytest : test]         # 1. initialize collection with data
[pytest : test]         dim = 99
[pytest : test]         multiple_dim_array = [dim + dim, dim - 10]
[pytest : test]         collection_w, _, _, insert_ids, time_stamp = \
[pytest : test]             self.init_collection_general(prefix, True, dim=dim, is_index=False, primary_field=primary_field,
[pytest : test]                                          enable_dynamic_field=False, multiple_dim_array=multiple_dim_array)[0:5]
[pytest : test]         # 2. extract vector field name
[pytest : test]         vector_name_list = cf.extract_vector_field_name_list(collection_w)
[pytest : test]         flat_index = {"index_type": "FLAT", "params": {}, "metric_type": metric_type}
[pytest : test]         for vector_name in vector_name_list:
[pytest : test]             collection_w.create_index(vector_name, flat_index)
[pytest : test]         collection_w.create_index(ct.default_float_vec_field_name, flat_index)
[pytest : test]         collection_w.load()
[pytest : test]         # 3. prepare search params
[pytest : test]         req_list = []
[pytest : test]         id_list = []
[pytest : test]         for i in range(len(vector_name_list)):
[pytest : test]             vectors = [[random.random() for _ in range(multiple_dim_array[i])] for _ in range(1)]
[pytest : test]             search_params = {"metric_type": metric_type, "offset": 0}
[pytest : test]             search_param = {
[pytest : test]                 "data": vectors,
[pytest : test]                 "anns_field": vector_name_list[i],
[pytest : test]                 "param": search_params,
[pytest : test]                 "limit": min_dim,
[pytest : test]                 "expr": default_search_exp}
[pytest : test]             req = AnnSearchRequest(**search_param)
[pytest : test]             req_list.append(req)
[pytest : test]             search_res = collection_w.search(vectors[:1], vector_name_list[i],
[pytest : test]                                              search_params, min_dim,
[pytest : test]                                              default_search_exp,
[pytest : test]                                              check_task=CheckTasks.check_search_results,
[pytest : test]                                              check_items={"nq": 1,
[pytest : test]                                                           "ids": insert_ids,
[pytest : test]                                                           "limit": min_dim})[0]
[pytest : test]             id_list.extend(search_res[0].ids)
[pytest : test]         # 4. hybrid search
[pytest : test]         hybrid_search = collection_w.hybrid_search(req_list, WeightedRanker(0.1, 0.9), default_limit)[0]
[pytest : test]         assert len(hybrid_search) == 1
[pytest : test] >       assert len(hybrid_search[0].ids) == len(list(set(id_list)))
[pytest : test] E       AssertionError: assert 10 == 4
[pytest : test] E        +  where 10 = len([980, 1430, 1240, 1247, 1360, 1516, ...])
[pytest : test] E        +    where [980, 1430, 1240, 1247, 1360, 1516, ...] = ['id: 980, distance: 0.8404409289360046, entity: {}', 'id: 1430, distance: 0.83946293592453, entity: {}', 'id: 1240, distance: 0.8389506936073303, entity: {}', 'id: 1247, distance: 0.8389191627502441, entity: {}', 'id: 1360, distance: 0.8387582898139954, entity: {}', 'id: 1516, distance: 0.8387090563774109, entity: {}', 'id: 1932, distance: 0.8386892676353455, entity: {}', 'id: 184, distance: 0.8385480642318726, entity: {}', 'id: 1825, distance: 0.8385159969329834, entity: {}', 'id: 653, distance: 0.8384677171707153, entity: {}'].ids
[pytest : test] E        +  and   4 = len([1430, 427, 980, 1422])
[pytest : test] E        +    where [1430, 427, 980, 1422] = list({427, 980, 1422, 1430})
[pytest : test] E        +      where {427, 980, 1422, 1430} = set([427, 1422, 980, 1430])

Expected Behavior

pass

Steps To Reproduce

No response

Milvus Log

https://jenkins.milvus.io:18080/blue/organizations/jenkins/Milvus%20Nightly%20CI(new)/detail/2.4/21/pipeline/118/

case:
test_hybrid_search_overall_limit_larger_sum_each_limit
test_hybrid_search_min_limit

Anything else?

No response

[NicoYuan1986]

@binbinlv Can you help look at the error? it is case error or kernel error ?

[binbinlv]

@binbinlv Can you help look at the error? it is case error or kernel error ?

Here I think is a bug, the hybrid search returns more than the sum of limit of each search request.

here the limit of search request is 2, and 2 search requests, so the sum limit is 4, even if the limit of hybrid search is 10, it should only return 4.

[binbinlv]

@czs007

could you please have a look here please? Thanks.

/assign @czs007

[czs007]

working

[yanliang567]

/assign @NicoYuan1986
please help to verify the fix above.

[NicoYuan1986]

fixed. master-20241106-8275e40f