enhance: Improve GetVectorById of Sparse Float Vector

internal/core/src/mmap/Column.h

@@ -43,7 +43,7 @@
 namespace milvus {
 
 /*
-* If string field's value all empty, need a string padding to avoid 
+* If string field's value all empty, need a string padding to avoid
 * mmap failing because size_ is zero which causing invalid arguement
 * array has the same problem
 * TODO: remove it when support NULL value
@@ -84,6 +84,7 @@
     }
 
     // mmap mode ctor
+    // User must call Seal to build the view for variable length column.
     ColumnBase(const File& file, size_t size, const FieldMeta& field_meta)
         : type_size_(IsSparseFloatVectorDataType(field_meta.get_data_type())
                          ? 1
@@ -106,12 +107,16 @@
     }
 
     // mmap mode ctor
+    // User must call Seal to build the view for variable length column.
     ColumnBase(const File& file,
                size_t size,
                int dim,
                const DataType& data_type)
-        : type_size_(GetDataTypeSize(data_type, dim)),
-          num_rows_(size / GetDataTypeSize(data_type, dim)),
+        : type_size_(IsSparseFloatVectorDataType(data_type)
+                         ? 1
+                         : GetDataTypeSize(data_type, dim)),
+          num_rows_(
+              IsSparseFloatVectorDataType(data_type) ? 1 : (size / type_size_)),
           size_(size),
           cap_size_(size),
           is_map_anonymous_(false) {
@@ -153,11 +158,18 @@
         column.size_ = 0;
     }
 
+    // Data() points at an addr that contains the elements
     virtual const char*
     Data() const {
         return data_;
     }
 
+    // MmappedData() returns the mmaped address
+    const char*
+    MmappedData() const {
+        return data_;
+    }
+
     size_t
     NumRows() const {
         return num_rows_;
@@ -273,6 +285,7 @@
     // capacity in bytes
     size_t cap_size_{0};
     size_t padding_{0};
+    // type_size_ is not used for sparse float vector column.
     const size_t type_size_{1};
     size_t num_rows_{0};
 
@@ -344,8 +357,7 @@
     }
 };
 
-// mmap not yet supported, thus SparseFloatColumn is not using fields in super
-// class such as ColumnBase::data.
+// when mmap is used, size_, data_ and num_rows_ of ColumnBase are used.
 class SparseFloatColumn : public ColumnBase {
  public:
     // memory mode ctor
@@ -356,7 +368,13 @@
                       size_t size,
                       const FieldMeta& field_meta)
         : ColumnBase(file, size, field_meta) {
-        AssertInfo(false, "SparseFloatColumn mmap mode not supported");
+    }
+    // mmap mode ctor
+    SparseFloatColumn(const File& file,
+                      size_t size,
+                      int dim,
+                      const DataType& data_type)
+        : ColumnBase(file, size, dim, data_type) {
     }
 
     SparseFloatColumn(SparseFloatColumn&& column) noexcept
@@ -372,14 +390,6 @@
         return static_cast<const char*>(static_cast<const void*>(vec_.data()));
     }
 
-    // This is used to advice mmap prefetch, we don't currently support mmap for
-    // sparse float vector thus not implemented for now.
-    size_t
-    ByteSize() const override {
-        throw std::runtime_error(
-            "ByteSize not supported for sparse float column");
-    }
-
     size_t
     Capacity() const override {
         throw std::runtime_error(
@@ -413,6 +423,34 @@
         return dim_;
     }
 
+    void
+    Seal(std::vector<uint64_t> indices) {
+        AssertInfo(!indices.empty(),
+                   "indices should not be empty, Seal() of "
+                   "SparseFloatColumn must be called only "
+                   "at mmap mode");
+        AssertInfo(data_,
+                   "data_ should not be nullptr, Seal() of "
+                   "SparseFloatColumn must be called only "
+                   "at mmap mode");
+        num_rows_ = indices.size();
+        // so that indices[num_rows_] - indices[num_rows_ - 1] is the size of
+        // the last row.
+        indices.push_back(size_);
+        for (size_t i = 0; i < num_rows_; i++) {
+            auto vec_size = indices[i + 1] - indices[i];
+            AssertInfo(
+                vec_size % knowhere::sparse::SparseRow<float>::element_size() ==
+                    0,
+                "Incorrect sparse vector size: {}",
+                vec_size);
+            vec_.emplace_back(
+                vec_size / knowhere::sparse::SparseRow<float>::element_size(),
+                (uint8_t*)(data_) + indices[i],
+                false);
+        }
+    }
+
  private:
     int64_t dim_ = 0;
     std::vector<knowhere::sparse::SparseRow<float>> vec_;

internal/core/src/mmap/Utils.h

@@ -81,10 +81,18 @@
                 break;
             }
             case DataType::VECTOR_SPARSE_FLOAT: {
-                // TODO(SPARSE): this is for mmap to write data to disk so that
-                // the file can be mmaped into memory.
-                throw std::runtime_error(
-                    "WriteFieldData for VECTOR_SPARSE_FLOAT not implemented");
+                for (size_t i = 0; i < data->get_num_rows(); ++i) {
+                    auto vec =
+                        static_cast<const knowhere::sparse::SparseRow<float>*>(
+                            data->RawValue(i));
+                    ssize_t written =
+                        file.Write(vec->data(), vec->data_byte_size());
+                    if (written < vec->data_byte_size()) {
+                        break;
+                    }
+                    total_written += written;
+                }
+                break;
             }
             default:
                 PanicInfo(DataTypeInvalid,

internal/core/src/segcore/SegmentSealedImpl.cpp

@@ -585,7 +585,13 @@
                 column = std::move(arr_column);
                 break;
             }
-            // TODO(SPARSE) support mmap
+            case milvus::DataType::VECTOR_SPARSE_FLOAT: {
+                auto sparse_column = std::make_shared<SparseFloatColumn>(
+                    file, total_written, field_meta);
+                sparse_column->Seal(std::move(indices));
+                column = std::move(sparse_column);
+                break;
+            }
             default: {
                 PanicInfo(DataTypeInvalid,
                           fmt::format("unsupported data type {}", data_type));
@@ -850,7 +856,7 @@
     auto metric_type = vec_index->GetMetricType();
     auto has_raw_data = vec_index->HasRawData();
 
-    if (has_raw_data) {
+    if (has_raw_data && !TEST_skip_index_for_retrieve_) {
         // If index has raw data, get vector from memory.
         auto ids_ds = GenIdsDataset(count, ids);
         if (field_meta.get_data_type() == DataType::VECTOR_SPARSE_FLOAT) {
@@ -865,9 +871,6 @@
         }
     }
 
-    AssertInfo(field_meta.get_data_type() != DataType::VECTOR_SPARSE_FLOAT,
-               "index of sparse float vector is guaranteed to have raw data");
-
     // If index doesn't have raw data, get vector from chunk cache.
     auto cc = storage::ChunkCacheSingleton::GetInstance().GetChunkCache();
 
@@ -898,23 +901,50 @@
         path_to_column[data_path] = column;
     }
 
-    // assign to data array
-    auto row_bytes = field_meta.get_sizeof();
-    auto buf = std::vector<char>(count * row_bytes);
-    for (auto i = 0; i < count; i++) {
-        AssertInfo(id_to_data_path.count(ids[i]) != 0, "id not found");
-        const auto& [data_path, offset_in_binlog] = id_to_data_path.at(ids[i]);
-        AssertInfo(path_to_column.count(data_path) != 0, "column not found");
-        const auto& column = path_to_column.at(data_path);
-        AssertInfo(offset_in_binlog * row_bytes < column->ByteSize(),
-                   "column idx out of range, idx: {}, size: {}, data_path: {}",
-                   offset_in_binlog * row_bytes,
-                   column->ByteSize(),
-                   data_path);
-        auto vector = &column->Data()[offset_in_binlog * row_bytes];
-        std::memcpy(buf.data() + i * row_bytes, vector, row_bytes);
+    if (field_meta.get_data_type() == DataType::VECTOR_SPARSE_FLOAT) {
+        auto buf = std::vector<knowhere::sparse::SparseRow<float>>(count);
+        for (auto i = 0; i < count; ++i) {
+            const auto& [data_path, offset_in_binlog] =
+                id_to_data_path.at(ids[i]);
+            const auto& column = path_to_column.at(data_path);
+            AssertInfo(
+                offset_in_binlog < column->NumRows(),
+                "column idx out of range, idx: {}, size: {}, data_path: {}",
+                offset_in_binlog,
+                column->NumRows(),
+                data_path);
+            auto sparse_column =
+                std::dynamic_pointer_cast<SparseFloatColumn>(column);
+            AssertInfo(sparse_column, "incorrect column created");
+            buf[i] = static_cast<const knowhere::sparse::SparseRow<float>*>(
+                static_cast<const void*>(
+                    sparse_column->Data()))[offset_in_binlog];
+        }
+        return segcore::CreateVectorDataArrayFrom(
+            buf.data(), count, field_meta);
+    } else {
+        // assign to data array
+        auto row_bytes = field_meta.get_sizeof();
+        auto buf = std::vector<char>(count * row_bytes);
+        for (auto i = 0; i < count; ++i) {
+            AssertInfo(id_to_data_path.count(ids[i]) != 0, "id not found");
+            const auto& [data_path, offset_in_binlog] =
+                id_to_data_path.at(ids[i]);
+            AssertInfo(path_to_column.count(data_path) != 0,
+                       "column not found");
+            const auto& column = path_to_column.at(data_path);
+            AssertInfo(
+                offset_in_binlog * row_bytes < column->ByteSize(),
+                "column idx out of range, idx: {}, size: {}, data_path: {}",
+                offset_in_binlog * row_bytes,
+                column->ByteSize(),
+                data_path);
+            auto vector = &column->Data()[offset_in_binlog * row_bytes];
+            std::memcpy(buf.data() + i * row_bytes, vector, row_bytes);
+        }
+        return segcore::CreateVectorDataArrayFrom(
+            buf.data(), count, field_meta);
     }
-    return segcore::CreateVectorDataArrayFrom(buf.data(), count, field_meta);
 }
 
 void
@@ -997,7 +1027,8 @@
 SegmentSealedImpl::SegmentSealedImpl(SchemaPtr schema,
                                      IndexMetaPtr index_meta,
                                      const SegcoreConfig& segcore_config,
-                                     int64_t segment_id)
+                                     int64_t segment_id,
+                                     bool TEST_skip_index_for_retrieve)
     : segcore_config_(segcore_config),
       field_data_ready_bitset_(schema->size()),
       index_ready_bitset_(schema->size()),
@@ -1006,7 +1037,8 @@
       insert_record_(*schema, MAX_ROW_COUNT),
       schema_(schema),
       id_(segment_id),
-      col_index_meta_(index_meta) {
+      col_index_meta_(index_meta),
+      TEST_skip_index_for_retrieve_(TEST_skip_index_for_retrieve) {
 }
 
 SegmentSealedImpl::~SegmentSealedImpl() {

internal/core/src/segcore/SegmentSealedImpl.h

@@ -42,7 +42,8 @@ class SegmentSealedImpl : public SegmentSealed {
     explicit SegmentSealedImpl(SchemaPtr schema,
                                IndexMetaPtr index_meta,
                                const SegcoreConfig& segcore_config,
-                               int64_t segment_id);
+                               int64_t segment_id,
+                               bool TEST_skip_index_for_retrieve = false);
     ~SegmentSealedImpl() override;
     void
     LoadIndex(const LoadIndexInfo& info) override;
@@ -312,16 +313,24 @@ class SegmentSealedImpl : public SegmentSealed {
         vec_binlog_config_;
 
     SegmentStats stats_{};
+
+    // for sparse vector unit test only! Once a type of sparse index that
+    // doesn't has raw data is added, this should be removed.
+    bool TEST_skip_index_for_retrieve_ = false;
 };
 
 inline SegmentSealedUPtr
 CreateSealedSegment(
     SchemaPtr schema,
     IndexMetaPtr index_meta = nullptr,
     int64_t segment_id = -1,
-    const SegcoreConfig& segcore_config = SegcoreConfig::default_config()) {
-    return std::make_unique<SegmentSealedImpl>(
-        schema, index_meta, segcore_config, segment_id);
+    const SegcoreConfig& segcore_config = SegcoreConfig::default_config(),
+    bool TEST_skip_index_for_retrieve = false) {
+    return std::make_unique<SegmentSealedImpl>(schema,
+                                               index_meta,
+                                               segcore_config,
+                                               segment_id,
+                                               TEST_skip_index_for_retrieve);
 }
 
 }  // namespace milvus::segcore

internal/core/src/storage/ChunkCache.cpp

@@ -64,10 +64,10 @@
     }
 
     auto column = it->second;
-    auto ok =
-        madvise(reinterpret_cast<void*>(const_cast<char*>(column->Data())),
-                column->ByteSize(),
-                read_ahead_policy_);
+    auto ok = madvise(
+        reinterpret_cast<void*>(const_cast<char*>(column->MmappedData())),
+        column->ByteSize(),
+        read_ahead_policy_);
     AssertInfo(ok == 0,
                "failed to madvise to the data file {}, err: {}",
                path,
@@ -100,7 +100,18 @@
 
     std::shared_ptr<ColumnBase> column{};
 
-    if (IsVariableDataType(data_type)) {
+    if (IsSparseFloatVectorDataType(data_type)) {
+        std::vector<uint64_t> indices{};
+        uint64_t offset = 0;
+        for (auto i = 0; i < field_data->get_num_rows(); ++i) {
+            indices.push_back(offset);
+            offset += field_data->Size(i);
+        }
+        auto sparse_column = std::make_shared<SparseFloatColumn>(
+            file, data_size, dim, data_type);
+        sparse_column->Seal(std::move(indices));
+        column = std::move(sparse_column);
+    } else if (IsVariableDataType(data_type)) {
         AssertInfo(
             false, "TODO: unimplemented for variable data type: {}", data_type);
     } else {

internal/core/src/storage/Util.cpp

@@ -533,13 +533,16 @@ EncodeAndUploadIndexSlice2(std::shared_ptr<milvus_storage::Space> space,
 
 std::pair<std::string, size_t>
 EncodeAndUploadFieldSlice(ChunkManager* chunk_manager,
-                          uint8_t* buf,
+                          void* buf,
                           int64_t element_count,
                           FieldDataMeta field_data_meta,
                           const FieldMeta& field_meta,
                           std::string object_key) {
-    auto field_data =
-        CreateFieldData(field_meta.get_data_type(), field_meta.get_dim(), 0);
+    // dim should not be used for sparse float vector field
+    auto dim = IsSparseFloatVectorDataType(field_meta.get_data_type())
+                   ? -1
+                   : field_meta.get_dim();
+    auto field_data = CreateFieldData(field_meta.get_data_type(), dim, 0);
     field_data->FillFieldData(buf, element_count);
     auto insertData = std::make_shared<InsertData>(field_data);
     insertData->SetFieldDataMeta(field_data_meta);

internal/core/src/storage/Util.h

@@ -110,7 +110,7 @@ EncodeAndUploadIndexSlice2(std::shared_ptr<milvus_storage::Space> space,
                            std::string object_key);
 std::pair<std::string, size_t>
 EncodeAndUploadFieldSlice(ChunkManager* chunk_manager,
-                          uint8_t* buf,
+                          void* buf,
                           int64_t element_count,
                           FieldDataMeta field_data_meta,
                           const FieldMeta& field_meta,