[BUG] Error doing pagination with hybrid searches

[sonic182]

Describe the bug

When doing pagination with an hybrid query and a search pipeline, I receive the following error: score normalization processor cannot produce final query result, the number of documents after fetch phase [10] is different from number of documents from query phase [12]

Maybe duplicated of opensearch-project/neural-search#280

query and response:

# query
 %{
  "_source" => %{"exclude" => ["vector"], "includes" => []},
  "aggs" => %{
    "group_count" => %{
      "aggs" => %{
        "total" => %{
          "cardinality" => %{
            "field" => "grouping_id",
            "precision_threshold" => 1000
          }
        }
      },
      "filter" => %{"bool" => %{"must" => []}}
    }
  },
  "collapse" => %{"field" => "grouping_id"},
  "from" => 2,
  "post_filter" => %{"bool" => %{"must" => []}},
  "query" => %{
    "hybrid" => %{
      "queries" => [
        %{
          "function_score" => %{
            "boost_mode" => "sum",
            "query" => %{
              "function_score" => %{
                "functions" => [
                  %{
                    "field_value_factor" => %{
                      "field" => "manual_boost",
                      "missing" => 1
                    }
                  },
                  %{
                    "field_value_factor" => %{
                      "field" => "auto_boost",
                      "missing" => 1
                    }
                  }
                ],
                "query" => %{
                  "dis_max" => %{
                    "queries" => [
                      %{
                        "bool" => %{
                          "minimum_should_match" => 1,
                          "should" => [
                            %{
                              "multi_match" => %{
                                "cutoff_frequency" => 0.1,
                                "fields" => ["brand^2", "categories^2",
                                 "all^1", "gtin^1", "mpn^1", "title^3",
                                 "title.autocomplete^1"],
                                "operator" => "and",
                                "query" => "aifon 13",
                                "type" => "best_fields"
                              }
                            }
                          ]
                        }
                      },
                      %{
                        "bool" => %{
                          "minimum_should_match" => 1,
                          "should" => [
                            %{
                              "multi_match" => %{
                                "cutoff_frequency" => 0.1,
                                "fields" => ["brand^2", "categories^2",
                                 "all^1", "gtin^1", "mpn^1", "title^3",
                                 "title.autocomplete^1"],
                                "query" => "aifon 13",
                                "type" => "best_fields"
                              }
                            }
                          ]
                        }
                      }
                    ],
                    "tie_breaker" => 0.3
                  }
                }
              }
            },
            "score_mode" => "max"
          }
        },
        %{
          "knn" => %{
            "vector" => %{
              "k" => 30,
              "vector" => [-0.047876860946416855, 0.05234730988740921,
               0.01735660620033741, 0.03566263988614082, -0.020095694810152054,
               0.005881187971681356, 0.08795266598463058, 0.01124920230358839,
               0.026371607556939125, -0.027546236291527748,
               -0.06709688901901245, -0.030416270717978477, 0.05761145427823067,
               0.04374939948320389, -0.004689693916589022, 0.02921660803258419,
               0.016390908509492874, -0.005048255901783705,
               0.027706122025847435, -0.006353649776428938,
               -0.022267673164606094, 0.05834672600030899,
               -0.024640902876853943, 0.025103332474827766,
               -0.023313814774155617, -0.03295905515551567,
               0.0069340369664132595, -0.018305879086256027,
               -0.02108374983072281, 0.041981711983680725,
               -1.7442446551285684e-4, -0.015122073702514172,
               0.014571044594049454, 0.036353059113025665, 0.004828275181353092,
               -0.012067032046616077, ...]
            }
          }
        }
      ]
    }
  },
  "size" => 2,
  "sort" => [%{"_score" => "desc"}],
  "track_scores" => true
}

# RESPONSE
%{
 "error" => %{
   "caused_by" => %{
     "reason" => "score normalization processor cannot produce final query result, the number of documents after fetch phase [10] is different from number of documents from query phase [12]",
     "type" => "illegal_state_exception"
   },
   "failed_shards" => [],
   "grouped" => true,
   "phase" => "query",
   "reason" => "The phase has failed",
   "root_cause" => [],
   "type" => "search_phase_execution_exception"
 },
 "status" => 500
}

Related component

Search

To Reproduce

1. have an index with a vector field
2. do an hybrid query with "from" and "size" params
3. See error

Expected behavior

Working pagination

Additional Details

Plugins

• icu-analyzer

Screenshots

Host/Environment (please complete the following information):

• OS: docker opensearch 2.16.0

Additional context

[sandeshkr419]

Search Triage: @martin-gaievski - Do you want to a take a look once?