milvus/internal/core/unittest/test_indexing.cpp

// Copyright (C) 2019-2020 Zilliz. All rights reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance
// with the License. You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software distributed under the License
// is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express
// or implied. See the License for the specific language governing permissions and limitations under the License

#include <gtest/gtest.h>

#include <iostream>
#include <random>
#include <string>
#include <vector>

#include "faiss/utils/distances.h"
#include "query/SearchBruteForce.h"
#include "segcore/Reduce.h"
#include "index/IndexFactory.h"
#include "knowhere/archive/KnowhereConfig.h"
#include "common/QueryResult.h"
#include "test_utils/indexbuilder_test_utils.h"
#include "test_utils/DataGen.h"
#include "test_utils/Timer.h"

#ifdef BUILD_DISK_ANN
#include "storage/MinioChunkManager.h"
#include "storage/DiskFileManagerImpl.h"

using namespace boost::filesystem;
#endif

using namespace milvus;
using namespace milvus::segcore;

namespace {
template <int DIM>
auto
generate_data(int N) {
    std::vector<float> raw_data;
    std::vector<uint64_t> timestamps;
    std::vector<int64_t> uids;
    std::default_random_engine er(42);
    std::uniform_real_distribution<> distribution(0.0, 1.0);
    std::default_random_engine ei(42);
    for (int i = 0; i < N; ++i) {
        uids.push_back(10 * N + i);
        timestamps.push_back(0);
        // append vec
        std::vector<float> vec(DIM);
        for (auto& x : vec) {
            x = distribution(er);
        }
        raw_data.insert(raw_data.end(), std::begin(vec), std::end(vec));
    }
    return std::make_tuple(raw_data, timestamps, uids);
}
}  // namespace

Status
merge_into(int64_t queries,
           int64_t topk,
           float* distances,
           int64_t* uids,
           const float* new_distances,
           const int64_t* new_uids) {
    for (int64_t qn = 0; qn < queries; ++qn) {
        auto base = qn * topk;
        auto src2_dis = distances + base;
        auto src2_uids = uids + base;

        auto src1_dis = new_distances + base;
        auto src1_uids = new_uids + base;

        std::vector<float> buf_dis(topk);
        std::vector<int64_t> buf_uids(topk);

        auto it1 = 0;
        auto it2 = 0;

        for (auto buf = 0; buf < topk; ++buf) {
            if (src1_dis[it1] <= src2_dis[it2]) {
                buf_dis[buf] = src1_dis[it1];
                buf_uids[buf] = src1_uids[it1];
                ++it1;
            } else {
                buf_dis[buf] = src2_dis[it2];
                buf_uids[buf] = src2_uids[it2];
                ++it2;
            }
        }
        std::copy_n(buf_dis.data(), topk, src2_dis);
        std::copy_n(buf_uids.data(), topk, src2_uids);
    }
    return Status::OK();
}

TEST(Indexing, SmartBruteForce) {
    int64_t N = 1000;
    auto [raw_data, timestamps, uids] = generate_data<DIM>(N);

    constexpr int64_t queries = 3;
    auto total_count = queries * K;

    auto raw = (const float*)raw_data.data();
    EXPECT_NE(raw, nullptr);

    auto query_data = raw;

    std::vector<int64_t> final_uids(total_count, -1);
    std::vector<float> final_dis(total_count, std::numeric_limits<float>::max());

    for (int beg = 0; beg < N; beg += TestChunkSize) {
        std::vector<int64_t> buf_uids(total_count, -1);
        std::vector<float> buf_dis(total_count, std::numeric_limits<float>::max());
        faiss::float_maxheap_array_t buf = {queries, K, buf_uids.data(), buf_dis.data()};
        auto end = beg + TestChunkSize;
        if (end > N) {
            end = N;
        }
        auto nsize = end - beg;
        auto src_data = raw + beg * DIM;

        faiss::knn_L2sqr(query_data, src_data, DIM, queries, nsize, &buf, nullptr);
        for (auto& x : buf_uids) {
            x = uids[x + beg];
        }
        merge_into(queries, K, final_dis.data(), final_uids.data(), buf_dis.data(), buf_uids.data());
    }

    for (int qn = 0; qn < queries; ++qn) {
        for (int kn = 0; kn < K; ++kn) {
            auto index = qn * K + kn;
            std::cout << final_uids[index] << "->" << final_dis[index] << std::endl;
        }
        std::cout << std::endl;
    }
}

TEST(Indexing, BinaryBruteForce) {
    int64_t N = 100000;
    int64_t num_queries = 10;
    int64_t topk = 5;
    int64_t round_decimal = 3;
    int64_t dim = 8192;
    auto metric_type = knowhere::metric::JACCARD;
    auto result_count = topk * num_queries;
    auto schema = std::make_shared<Schema>();
    auto vec_fid = schema->AddDebugField("vecbin", DataType::VECTOR_BINARY, dim, metric_type);
    auto i64_fid = schema->AddDebugField("age", DataType::INT64);
    auto dataset = DataGen(schema, N, 10);
    auto bin_vec = dataset.get_col<uint8_t>(vec_fid);
    auto query_data = 1024 * dim / 8 + bin_vec.data();
    query::dataset::SearchDataset search_dataset{
        metric_type,  //
        num_queries,  //
        topk,         //
        round_decimal,
        dim,        //
        query_data  //
    };

    auto sub_result = query::BruteForceSearch(search_dataset, bin_vec.data(), N, nullptr);

    SearchResult sr;
    sr.total_nq_ = num_queries;
    sr.unity_topK_ = topk;
    sr.seg_offsets_ = std::move(sub_result.mutable_seg_offsets());
    sr.distances_ = std::move(sub_result.mutable_distances());

    auto json = SearchResultToJson(sr);
    std::cout << json.dump(2);
#ifdef __linux__
    auto ref = json::parse(R"(
[
  [
    [ "1024->0.000000", "48942->0.642000", "18494->0.644000", "68225->0.644000", "93557->0.644000" ],
    [ "1025->0.000000", "73557->0.641000", "53086->0.643000", "9737->0.643000", "62855->0.644000" ],
    [ "1026->0.000000", "62904->0.644000", "46758->0.644000", "57969->0.645000", "98113->0.646000" ],
    [ "1027->0.000000", "92446->0.638000", "96034->0.640000", "92129->0.644000", "45887->0.644000" ],
    [ "1028->0.000000", "22992->0.643000", "73903->0.644000", "19969->0.645000", "65178->0.645000" ],
    [ "1029->0.000000", "19776->0.641000", "15166->0.642000", "85470->0.642000", "16730->0.643000" ],
    [ "1030->0.000000", "55939->0.640000", "84253->0.643000", "31958->0.644000", "11667->0.646000" ],
    [ "1031->0.000000", "89536->0.637000", "61622->0.638000", "9275->0.639000", "91403->0.640000" ],
    [ "1032->0.000000", "69504->0.642000", "23414->0.644000", "48770->0.645000", "23231->0.645000" ],
    [ "1033->0.000000", "33540->0.636000", "25310->0.640000", "18576->0.640000", "73729->0.642000" ]
  ]
]
)");
#else  // for mac
    auto ref = json::parse(R"(
[
  [
    [ "1024->0.000000", "59169->0.645000", "98548->0.646000", "3356->0.646000", "90373->0.647000" ],
    [ "1025->0.000000", "61245->0.638000", "95271->0.639000", "31087->0.639000", "31549->0.640000" ],
    [ "1026->0.000000", "65225->0.648000", "35750->0.648000", "14971->0.649000", "75385->0.649000" ],
    [ "1027->0.000000", "70158->0.640000", "27076->0.640000", "3407->0.641000", "59527->0.641000" ],
    [ "1028->0.000000", "45757->0.645000", "3356->0.645000", "77230->0.646000", "28690->0.647000" ],
    [ "1029->0.000000", "13291->0.642000", "24960->0.643000", "83770->0.643000", "88244->0.643000" ],
    [ "1030->0.000000", "96807->0.641000", "39920->0.643000", "62943->0.644000", "12603->0.644000" ],
    [ "1031->0.000000", "65769->0.648000", "60493->0.648000", "48738->0.648000", "4353->0.648000" ],
    [ "1032->0.000000", "57827->0.637000", "8213->0.638000", "22221->0.639000", "23328->0.640000" ],
    [ "1033->0.000000", "676->0.645000", "91430->0.646000", "85353->0.646000", "6014->0.646000" ]
  ]
]
)");
#endif
    auto json_str = json.dump(2);
    auto ref_str = ref.dump(2);
    ASSERT_EQ(json_str, ref_str);
}

TEST(Indexing, Naive) {
    constexpr int N = 10000;
    constexpr int TOPK = 10;

    auto [raw_data, timestamps, uids] = generate_data<DIM>(N);
    milvus::index::CreateIndexInfo create_index_info;
    create_index_info.field_type = DataType::VECTOR_FLOAT;
    create_index_info.metric_type = knowhere::metric::L2;
    create_index_info.index_type = knowhere::IndexEnum::INDEX_FAISS_IVFPQ;
    auto index = milvus::index::IndexFactory::GetInstance().CreateIndex(create_index_info, nullptr);

    auto build_conf = knowhere::Config{
        {knowhere::meta::METRIC_TYPE, knowhere::metric::L2},
        {knowhere::meta::DIM, std::to_string(DIM)},
        {knowhere::indexparam::NLIST, "100"},
        {knowhere::indexparam::M, "4"},
        {knowhere::indexparam::NBITS, "8"},
    };

    auto search_conf = knowhere::Config{
        {knowhere::meta::METRIC_TYPE, knowhere::metric::L2},
        {knowhere::indexparam::NPROBE, 4},
    };

    std::vector<knowhere::DatasetPtr> datasets;
    std::vector<std::vector<float>> ftrashs;
    auto raw = raw_data.data();
    for (int beg = 0; beg < N; beg += TestChunkSize) {
        auto end = beg + TestChunkSize;
        if (end > N) {
            end = N;
        }
        std::vector<float> ft(raw + DIM * beg, raw + DIM * end);

        auto ds = knowhere::GenDataset(end - beg, DIM, ft.data());
        datasets.push_back(ds);
        ftrashs.push_back(std::move(ft));
    }

    for (auto& ds : datasets) {
        index->BuildWithDataset(ds, build_conf);
    }

    auto bitmap = BitsetType(N, false);
    // exclude the first
    for (int i = 0; i < N / 2; ++i) {
        bitmap.set(i);
    }

    BitsetView view = bitmap;
    auto query_ds = knowhere::GenDataset(1, DIM, raw_data.data());

    milvus::SearchInfo searchInfo;
    searchInfo.topk_ = TOPK;
    searchInfo.metric_type_ = knowhere::metric::L2;
    searchInfo.search_params_ = search_conf;
    auto vec_index = dynamic_cast<index::VectorIndex*>(index.get());
    auto result = vec_index->Query(query_ds, searchInfo, view);

    for (int i = 0; i < TOPK; ++i) {
        if (result->seg_offsets_[i] < N / 2) {
            std::cout << "WRONG: ";
        }
        std::cout << result->seg_offsets_[i] << "->" << result->distances_[i] << std::endl;
    }
}

using Param = std::pair<knowhere::IndexType, knowhere::MetricType>;

class IndexTest : public ::testing::TestWithParam<Param> {
 protected:
    void
    SetUp() override {
        knowhere::KnowhereConfig::SetStatisticsLevel(3);
        storage_config_ = get_default_storage_config();

        auto param = GetParam();
        index_type = param.first;
        metric_type = param.second;
        build_conf = generate_build_conf(index_type, metric_type);
        load_conf = generate_load_conf(index_type, metric_type, NB);
        search_conf = generate_search_conf(index_type, metric_type);

        std::map<knowhere::MetricType, bool> is_binary_map = {
            {knowhere::IndexEnum::INDEX_FAISS_IDMAP, false},
            {knowhere::IndexEnum::INDEX_FAISS_IVFPQ, false},
            {knowhere::IndexEnum::INDEX_FAISS_IVFFLAT, false},
            {knowhere::IndexEnum::INDEX_FAISS_IVFSQ8, false},
            {knowhere::IndexEnum::INDEX_FAISS_BIN_IVFFLAT, true},
            {knowhere::IndexEnum::INDEX_FAISS_BIN_IDMAP, true},
            {knowhere::IndexEnum::INDEX_HNSW, false},
            {knowhere::IndexEnum::INDEX_ANNOY, false},
            {knowhere::IndexEnum::INDEX_DISKANN, false},
        };

        is_binary = is_binary_map[index_type];
        if (is_binary) {
            vec_field_data_type = milvus::DataType::VECTOR_BINARY;
        } else {
            vec_field_data_type = milvus::DataType::VECTOR_FLOAT;
        }

        auto dataset = GenDataset(NB, metric_type, is_binary);
        if (!is_binary) {
            xb_data = dataset.get_col<float>(milvus::FieldId(100));
            xb_dataset = knowhere::GenDataset(NB, DIM, xb_data.data());
            xq_dataset = knowhere::GenDataset(NQ, DIM, xb_data.data() + DIM * query_offset);
        } else {
            xb_bin_data = dataset.get_col<uint8_t>(milvus::FieldId(100));
            xb_dataset = knowhere::GenDataset(NB, DIM, xb_bin_data.data());
            xq_dataset = knowhere::GenDataset(NQ, DIM, xb_bin_data.data() + DIM * query_offset);
        }
    }

    void
    TearDown() override {
    }

 protected:
    std::string index_type, metric_type;
    bool is_binary;
    milvus::Config build_conf;
    milvus::Config load_conf;
    milvus::Config search_conf;
    milvus::DataType vec_field_data_type;
    knowhere::DatasetPtr xb_dataset;
    std::vector<float> xb_data;
    std::vector<uint8_t> xb_bin_data;
    knowhere::DatasetPtr xq_dataset;
    int64_t query_offset = 100;
    int64_t NB = 10000;
    StorageConfig storage_config_;
};

INSTANTIATE_TEST_CASE_P(
    IndexTypeParameters,
    IndexTest,
    ::testing::Values(std::pair(knowhere::IndexEnum::INDEX_FAISS_IDMAP, knowhere::metric::L2),
                      std::pair(knowhere::IndexEnum::INDEX_FAISS_IVFPQ, knowhere::metric::L2),
                      std::pair(knowhere::IndexEnum::INDEX_FAISS_IVFFLAT, knowhere::metric::L2),
                      std::pair(knowhere::IndexEnum::INDEX_FAISS_IVFSQ8, knowhere::metric::L2),
                      std::pair(knowhere::IndexEnum::INDEX_FAISS_BIN_IVFFLAT, knowhere::metric::JACCARD),
                      std::pair(knowhere::IndexEnum::INDEX_FAISS_BIN_IVFFLAT, knowhere::metric::TANIMOTO),
                      std::pair(knowhere::IndexEnum::INDEX_FAISS_BIN_IDMAP, knowhere::metric::JACCARD),
                      std::pair(knowhere::IndexEnum::INDEX_HNSW, knowhere::metric::L2),
                      // ci ut not start minio, so not run ut about diskann index for now
                      //#ifdef BUILD_DISK_ANN
                      //                      std::pair(knowhere::IndexEnum::INDEX_DISKANN, knowhere::metric::L2),
                      //#endif
                      std::pair(knowhere::IndexEnum::INDEX_ANNOY, knowhere::metric::L2)));

TEST_P(IndexTest, BuildAndQuery) {
    milvus::index::CreateIndexInfo create_index_info;
    create_index_info.index_type = index_type;
    create_index_info.metric_type = metric_type;
    create_index_info.field_type = vec_field_data_type;
    index::IndexBasePtr index;
    if (index_type == knowhere::IndexEnum::INDEX_DISKANN) {
#ifdef BUILD_DISK_ANN
        milvus::storage::FieldDataMeta field_data_meta{1, 2, 3, 100};
        milvus::storage::IndexMeta index_meta{3, 100, 1000, 1};
        auto file_manager =
            std::make_shared<milvus::storage::DiskFileManagerImpl>(field_data_meta, index_meta, storage_config_);
        index = milvus::index::IndexFactory::GetInstance().CreateIndex(create_index_info, file_manager);
#endif
    } else {
        index = milvus::index::IndexFactory::GetInstance().CreateIndex(create_index_info, nullptr);
    }
    ASSERT_NO_THROW(index->BuildWithDataset(xb_dataset, build_conf));
    milvus::index::IndexBasePtr new_index;
    milvus::index::VectorIndex* vec_index = nullptr;

    if (index_type == knowhere::IndexEnum::INDEX_DISKANN) {
#ifdef BUILD_DISK_ANN
        // TODO ::diskann.query need load first, ugly
        auto binary_set = index->Serialize(milvus::Config{});
        index.reset();
        milvus::storage::FieldDataMeta field_data_meta{1, 2, 3, 100};
        milvus::storage::IndexMeta index_meta{3, 100, 1000, 1};
        auto file_manager =
            std::make_shared<milvus::storage::DiskFileManagerImpl>(field_data_meta, index_meta, storage_config_);
        new_index = milvus::index::IndexFactory::GetInstance().CreateIndex(create_index_info, file_manager);

        vec_index = dynamic_cast<milvus::index::VectorIndex*>(new_index.get());

        std::vector<std::string> index_files;
        for (auto& binary : binary_set.binary_map_) {
            index_files.emplace_back(binary.first);
        }
        load_conf["index_files"] = index_files;
        vec_index->Load(binary_set, load_conf);
        EXPECT_EQ(vec_index->Count(), NB);
#endif
    } else {
        vec_index = dynamic_cast<milvus::index::VectorIndex*>(index.get());
    }
    EXPECT_EQ(vec_index->GetDim(), DIM);
    EXPECT_EQ(vec_index->Count(), NB);

    milvus::SearchInfo search_info;
    search_info.topk_ = K;
    search_info.metric_type_ = metric_type;
    search_info.search_params_ = search_conf;
    auto result = vec_index->Query(xq_dataset, search_info, nullptr);
    EXPECT_EQ(result->total_nq_, NQ);
    EXPECT_EQ(result->unity_topK_, K);
    EXPECT_EQ(result->distances_.size(), NQ * K);
    EXPECT_EQ(result->seg_offsets_.size(), NQ * K);
    if (!is_binary) {
        EXPECT_EQ(result->seg_offsets_[0], query_offset);
    }
}