Merge branch 'master' into master

dbms/src/AggregateFunctions/AggregateFunctionFactory.cpp

@@ -13,6 +13,7 @@
 #include <Common/typeid_cast.h>
 
 #include <Poco/String.h>
+#include <DataTypes/DataTypeWithDictionary.h>
 
 
 namespace DB
@@ -41,6 +42,20 @@ void AggregateFunctionFactory::registerFunction(const String & name, Creator cre
             ErrorCodes::LOGICAL_ERROR);
 }
 
+static DataTypes convertTypesWithDictionaryToNested(const DataTypes & types)
+{
+    DataTypes res_types;
+    res_types.reserve(types.size());
+    for (const auto & type : types)
+    {
+        if (auto * type_with_dict = typeid_cast<const DataTypeWithDictionary *>(type.get()))
+            res_types.push_back(type_with_dict->getDictionaryType());
+        else
+            res_types.push_back(type);
+    }
+
+    return res_types;
+}
 
 AggregateFunctionPtr AggregateFunctionFactory::get(
     const String & name,
@@ -48,6 +63,8 @@ AggregateFunctionPtr AggregateFunctionFactory::get(
     const Array & parameters,
     int recursion_level) const
 {
+    auto type_without_dictionary = convertTypesWithDictionaryToNested(argument_types);
+
     /// If one of types is Nullable, we apply aggregate function combinator "Null".
 
     if (std::any_of(argument_types.begin(), argument_types.end(),
@@ -57,7 +74,7 @@ AggregateFunctionPtr AggregateFunctionFactory::get(
         if (!combinator)
             throw Exception("Logical error: cannot find aggregate function combinator to apply a function to Nullable arguments.", ErrorCodes::LOGICAL_ERROR);
 
-        DataTypes nested_types = combinator->transformArguments(argument_types);
+        DataTypes nested_types = combinator->transformArguments(type_without_dictionary);
 
         AggregateFunctionPtr nested_function;
 
@@ -70,7 +87,7 @@ AggregateFunctionPtr AggregateFunctionFactory::get(
         return combinator->transformAggregateFunction(nested_function, argument_types, parameters);
     }
 
-    auto res = getImpl(name, argument_types, parameters, recursion_level);
+    auto res = getImpl(name, type_without_dictionary, parameters, recursion_level);
     if (!res)
         throw Exception("Logical error: AggregateFunctionFactory returned nullptr", ErrorCodes::LOGICAL_ERROR);
     return res;

dbms/src/Columns/ColumnAggregateFunction.cpp

@@ -4,6 +4,7 @@
 #include <IO/WriteBufferFromArena.h>
 #include <Common/SipHash.h>
 #include <Common/typeid_cast.h>
+#include <Columns/ColumnsCommon.h>
 
 namespace DB
 {
@@ -161,6 +162,25 @@ ColumnPtr ColumnAggregateFunction::permute(const Permutation & perm, size_t limi
     return std::move(res);
 }
 
+ColumnPtr ColumnAggregateFunction::index(const IColumn & indexes, size_t limit) const
+{
+    return selectIndexImpl(*this, indexes, limit);
+}
+
+template <typename Type>
+ColumnPtr ColumnAggregateFunction::indexImpl(const PaddedPODArray<Type> & indexes, size_t limit) const
+{
+    auto res = createView();
+
+    res->getData().resize(limit);
+    for (size_t i = 0; i < limit; ++i)
+        res->getData()[i] = getData()[indexes[i]];
+
+    return std::move(res);
+}
+
+INSTANTIATE_INDEX_IMPL(ColumnAggregateFunction);
+
 /// Is required to support operations with Set
 void ColumnAggregateFunction::updateHashWithValue(size_t n, SipHash & hash) const
 {

dbms/src/Columns/ColumnAggregateFunction.h

@@ -156,6 +156,11 @@ public:
 
     ColumnPtr permute(const Permutation & perm, size_t limit) const override;
 
+    ColumnPtr index(const IColumn & indexes, size_t limit) const override;
+
+    template <typename Type>
+    ColumnPtr indexImpl(const PaddedPODArray<Type> & indexes, size_t limit) const;
+
     ColumnPtr replicate(const Offsets & offsets) const override;
 
     MutableColumns scatter(ColumnIndex num_columns, const Selector & selector) const override;

dbms/src/Columns/ColumnArray.cpp

@@ -626,6 +626,44 @@ ColumnPtr ColumnArray::permute(const Permutation & perm, size_t limit) const
     return std::move(res);
 }
 
+ColumnPtr ColumnArray::index(const IColumn & indexes, size_t limit) const
+{
+    return selectIndexImpl(*this, indexes, limit);
+}
+
+template <typename T>
+ColumnPtr ColumnArray::indexImpl(const PaddedPODArray<T> & indexes, size_t limit) const
+{
+    if (limit == 0)
+        return ColumnArray::create(data);
+
+    /// Convert indexes to UInt64 in case of overflow.
+    auto nested_indexes_column = ColumnUInt64::create();
+    PaddedPODArray<UInt64> & nested_indexes = nested_indexes_column->getData();
+    nested_indexes.reserve(getOffsets().back());
+
+    auto res = ColumnArray::create(data->cloneEmpty());
+
+    Offsets & res_offsets = res->getOffsets();
+    res_offsets.resize(limit);
+    size_t current_offset = 0;
+
+    for (size_t i = 0; i < limit; ++i)
+    {
+        for (size_t j = 0; j < sizeAt(indexes[i]); ++j)
+            nested_indexes.push_back(offsetAt(indexes[i]) + j);
+        current_offset += sizeAt(indexes[i]);
+        res_offsets[i] = current_offset;
+    }
+
+    if (current_offset != 0)
+        res->data = data->index(*nested_indexes_column, current_offset);
+
+    return std::move(res);
+}
+
+INSTANTIATE_INDEX_IMPL(ColumnArray);
+
 void ColumnArray::getPermutation(bool reverse, size_t limit, int nan_direction_hint, Permutation & res) const
 {
     size_t s = size();

dbms/src/Columns/ColumnArray.h

@@ -71,6 +71,8 @@ public:
     void popBack(size_t n) override;
     ColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;
     ColumnPtr permute(const Permutation & perm, size_t limit) const override;
+    ColumnPtr index(const IColumn & indexes, size_t limit) const override;
+    template <typename Type> ColumnPtr indexImpl(const PaddedPODArray<Type> & indexes, size_t limit) const;
     int compareAt(size_t n, size_t m, const IColumn & rhs_, int nan_direction_hint) const override;
     void getPermutation(bool reverse, size_t limit, int nan_direction_hint, Permutation & res) const override;
     void reserve(size_t n) override;

dbms/src/Columns/ColumnConst.cpp

@@ -63,6 +63,18 @@ ColumnPtr ColumnConst::permute(const Permutation & perm, size_t limit) const
     return ColumnConst::create(data, limit);
 }
 
+ColumnPtr ColumnConst::index(const IColumn & indexes, size_t limit) const
+{
+    if (limit == 0)
+        limit = indexes.size();
+
+    if (indexes.size() < limit)
+        throw Exception("Size of indexes (" + toString(indexes.size()) + ") is less than required (" + toString(limit) + ")",
+                        ErrorCodes::SIZES_OF_COLUMNS_DOESNT_MATCH);
+
+    return ColumnConst::create(data, limit);
+}
+
 MutableColumns ColumnConst::scatter(ColumnIndex num_columns, const Selector & selector) const
 {
     if (s != selector.size())

dbms/src/Columns/ColumnConst.h

@@ -153,6 +153,7 @@ public:
     ColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;
     ColumnPtr replicate(const Offsets & offsets) const override;
     ColumnPtr permute(const Permutation & perm, size_t limit) const override;
+    ColumnPtr index(const IColumn & indexes, size_t limit) const override;
     void getPermutation(bool reverse, size_t limit, int nan_direction_hint, Permutation & res) const override;
 
     size_t byteSize() const override

dbms/src/Columns/ColumnFixedString.cpp

@@ -1,4 +1,5 @@
 #include <Columns/ColumnFixedString.h>
+#include <Columns/ColumnsCommon.h>
 
 #include <Common/Arena.h>
 #include <Common/SipHash.h>
@@ -258,6 +259,32 @@ ColumnPtr ColumnFixedString::permute(const Permutation & perm, size_t limit) con
     return std::move(res);
 }
 
+
+ColumnPtr ColumnFixedString::index(const IColumn & indexes, size_t limit) const
+{
+    return selectIndexImpl(*this, indexes, limit);
+}
+
+
+template <typename Type>
+ColumnPtr ColumnFixedString::indexImpl(const PaddedPODArray<Type> & indexes, size_t limit) const
+{
+    if (limit == 0)
+        return ColumnFixedString::create(n);
+
+    auto res = ColumnFixedString::create(n);
+
+    Chars_t & res_chars = res->chars;
+
+    res_chars.resize(n * limit);
+
+    size_t offset = 0;
+    for (size_t i = 0; i < limit; ++i, offset += n)
+        memcpySmallAllowReadWriteOverflow15(&res_chars[offset], &chars[indexes[i] * n], n);
+
+    return std::move(res);
+}
+
 ColumnPtr ColumnFixedString::replicate(const Offsets & offsets) const
 {
     size_t col_size = size();

dbms/src/Columns/ColumnFixedString.h

@@ -108,6 +108,11 @@ public:
 
     ColumnPtr permute(const Permutation & perm, size_t limit) const override;
 
+    ColumnPtr index(const IColumn & indexes, size_t limit) const override;
+
+    template <typename Type>
+    ColumnPtr indexImpl(const PaddedPODArray<Type> & indexes, size_t limit) const;
+
     ColumnPtr replicate(const Offsets & offsets) const override;
 
     MutableColumns scatter(ColumnIndex num_columns, const Selector & selector) const override

dbms/src/Columns/ColumnFunction.cpp

@@ -88,6 +88,15 @@ ColumnPtr ColumnFunction::permute(const Permutation & perm, size_t limit) const
     return ColumnFunction::create(limit, function, capture);
 }
 
+ColumnPtr ColumnFunction::index(const IColumn & indexes, size_t limit) const
+{
+    ColumnsWithTypeAndName capture = captured_columns;
+    for (auto & column : capture)
+        column.column = column.column->index(indexes, limit);
+
+    return ColumnFunction::create(limit, function, capture);
+}
+
 std::vector<MutableColumnPtr> ColumnFunction::scatter(IColumn::ColumnIndex num_columns,
                                                       const IColumn::Selector & selector) const
 {

dbms/src/Columns/ColumnFunction.h

@@ -33,6 +33,7 @@ public:
     ColumnPtr replicate(const Offsets & offsets) const override;
     ColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;
     ColumnPtr permute(const Permutation & perm, size_t limit) const override;
+    ColumnPtr index(const IColumn & indexes, size_t limit) const override;
     void insertDefault() override;
     void popBack(size_t n) override;
     std::vector<MutableColumnPtr> scatter(IColumn::ColumnIndex num_columns,

dbms/src/Columns/ColumnNullable.cpp

@@ -166,6 +166,13 @@ ColumnPtr ColumnNullable::permute(const Permutation & perm, size_t limit) const
     return ColumnNullable::create(permuted_data, permuted_null_map);
 }
 
+ColumnPtr ColumnNullable::index(const IColumn & indexes, size_t limit) const
+{
+    ColumnPtr indexed_data = getNestedColumn().index(indexes, limit);
+    ColumnPtr indexed_null_map = getNullMapColumn().index(indexes, limit);
+    return ColumnNullable::create(indexed_data, indexed_null_map);
+}
+
 int ColumnNullable::compareAt(size_t n, size_t m, const IColumn & rhs_, int null_direction_hint) const
 {
     /// NULL values share the properties of NaN values.

dbms/src/Columns/ColumnNullable.h

@@ -65,6 +65,7 @@ public:
     void popBack(size_t n) override;
     ColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;
     ColumnPtr permute(const Permutation & perm, size_t limit) const override;
+    ColumnPtr index(const IColumn & indexes, size_t limit) const override;
     int compareAt(size_t n, size_t m, const IColumn & rhs_, int null_direction_hint) const override;
     void getPermutation(bool reverse, size_t limit, int null_direction_hint, Permutation & res) const override;
     void reserve(size_t n) override;

dbms/src/Columns/ColumnString.cpp

@@ -159,6 +159,48 @@ ColumnPtr ColumnString::permute(const Permutation & perm, size_t limit) const
 }
 
 
+ColumnPtr ColumnString::index(const IColumn & indexes, size_t limit) const
+{
+    return selectIndexImpl(*this, indexes, limit);
+}
+
+template <typename Type>
+ColumnPtr ColumnString::indexImpl(const PaddedPODArray<Type> & indexes, size_t limit) const
+{
+    if (limit == 0)
+        return ColumnString::create();
+
+    auto res = ColumnString::create();
+
+    Chars_t & res_chars = res->chars;
+    Offsets & res_offsets = res->offsets;
+
+
+    size_t new_chars_size = 0;
+    for (size_t i = 0; i < limit; ++i)
+        new_chars_size += sizeAt(indexes[i]);
+    res_chars.resize(new_chars_size);
+
+    res_offsets.resize(limit);
+
+    Offset current_new_offset = 0;
+
+    for (size_t i = 0; i < limit; ++i)
+    {
+        size_t j = indexes[i];
+        size_t string_offset = j == 0 ? 0 : offsets[j - 1];
+        size_t string_size = offsets[j] - string_offset;
+
+        memcpySmallAllowReadWriteOverflow15(&res_chars[current_new_offset], &chars[string_offset], string_size);
+
+        current_new_offset += string_size;
+        res_offsets[i] = current_new_offset;
+    }
+
+    return std::move(res);
+}
+
+
 template <bool positive>
 struct ColumnString::less
 {

dbms/src/Columns/ColumnString.h

@@ -220,6 +220,11 @@ public:
 
     ColumnPtr permute(const Permutation & perm, size_t limit) const override;
 
+    ColumnPtr index(const IColumn & indexes, size_t limit) const override;
+
+    template <typename Type>
+    ColumnPtr indexImpl(const PaddedPODArray<Type> & indexes, size_t limit) const;
+
     void insertDefault() override
     {
         chars.push_back(0);

dbms/src/Columns/ColumnTuple.cpp

@@ -181,6 +181,17 @@ ColumnPtr ColumnTuple::permute(const Permutation & perm, size_t limit) const
     return ColumnTuple::create(new_columns);
 }
 
+ColumnPtr ColumnTuple::index(const IColumn & indexes, size_t limit) const
+{
+    const size_t tuple_size = columns.size();
+    Columns new_columns(tuple_size);
+
+    for (size_t i = 0; i < tuple_size; ++i)
+        new_columns[i] = columns[i]->index(indexes, limit);
+
+    return ColumnTuple::create(new_columns);
+}
+
 ColumnPtr ColumnTuple::replicate(const Offsets & offsets) const
 {
     const size_t tuple_size = columns.size();

dbms/src/Columns/ColumnTuple.h

@@ -60,6 +60,7 @@ public:
     void insertRangeFrom(const IColumn & src, size_t start, size_t length) override;
     ColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override;
     ColumnPtr permute(const Permutation & perm, size_t limit) const override;
+    ColumnPtr index(const IColumn & indexes, size_t limit) const override;
     ColumnPtr replicate(const Offsets & offsets) const override;
     MutableColumns scatter(ColumnIndex num_columns, const Selector & selector) const override;
     void gather(ColumnGathererStream & gatherer_stream) override;

dbms/src/Columns/ColumnUnique.h

@@ -0,0 +1,512 @@
+#pragma once
+#include <Columns/IColumnUnique.h>
+#include <Columns/ReverseIndex.h>
+
+#include <Columns/ColumnVector.h>
+#include <Columns/ColumnNullable.h>
+#include <Columns/ColumnString.h>
+#include <Columns/ColumnFixedString.h>
+
+#include <DataTypes/DataTypeNullable.h>
+#include <DataTypes/NumberTraits.h>
+
+#include <Common/typeid_cast.h>
+#include <ext/range.h>
+
+namespace DB
+{
+
+
+template <typename ColumnType>
+class ColumnUnique final : public COWPtrHelper<IColumnUnique, ColumnUnique<ColumnType>>
+{
+    friend class COWPtrHelper<IColumnUnique, ColumnUnique<ColumnType>>;
+
+private:
+    explicit ColumnUnique(MutableColumnPtr && holder, bool is_nullable);
+    explicit ColumnUnique(const IDataType & type);
+    ColumnUnique(const ColumnUnique & other);
+
+public:
+    MutableColumnPtr cloneEmpty() const override;
+
+    const ColumnPtr & getNestedColumn() const override;
+    const ColumnPtr & getNestedNotNullableColumn() const override { return column_holder; }
+
+    size_t uniqueInsert(const Field & x) override;
+    size_t uniqueInsertFrom(const IColumn & src, size_t n) override;
+    MutableColumnPtr uniqueInsertRangeFrom(const IColumn & src, size_t start, size_t length) override;
+    IColumnUnique::IndexesWithOverflow uniqueInsertRangeWithOverflow(const IColumn & src, size_t start, size_t length,
+                                                                     size_t max_dictionary_size) override;
+    size_t uniqueInsertData(const char * pos, size_t length) override;
+    size_t uniqueInsertDataWithTerminatingZero(const char * pos, size_t length) override;
+    size_t uniqueDeserializeAndInsertFromArena(const char * pos, const char *& new_pos) override;
+
+    size_t getDefaultValueIndex() const override { return is_nullable ? 1 : 0; }
+    size_t getNullValueIndex() const override;
+    bool canContainNulls() const override { return is_nullable; }
+
+    Field operator[](size_t n) const override { return (*getNestedColumn())[n]; }
+    void get(size_t n, Field & res) const override { getNestedColumn()->get(n, res); }
+    StringRef getDataAt(size_t n) const override { return getNestedColumn()->getDataAt(n); }
+    StringRef getDataAtWithTerminatingZero(size_t n) const override
+    {
+        return getNestedColumn()->getDataAtWithTerminatingZero(n);
+    }
+    UInt64 get64(size_t n) const override { return getNestedColumn()->get64(n); }
+    UInt64 getUInt(size_t n) const override { return getNestedColumn()->getUInt(n); }
+    Int64 getInt(size_t n) const override { return getNestedColumn()->getInt(n); }
+    bool isNullAt(size_t n) const override { return is_nullable && n == getNullValueIndex(); }
+    StringRef serializeValueIntoArena(size_t n, Arena & arena, char const *& begin) const override
+    {
+        return column_holder->serializeValueIntoArena(n, arena, begin);
+    }
+    void updateHashWithValue(size_t n, SipHash & hash) const override
+    {
+        return getNestedColumn()->updateHashWithValue(n, hash);
+    }
+
+    int compareAt(size_t n, size_t m, const IColumn & rhs, int nan_direction_hint) const override
+    {
+        auto & column_unique = static_cast<const IColumnUnique&>(rhs);
+        return getNestedColumn()->compareAt(n, m, *column_unique.getNestedColumn(), nan_direction_hint);
+    }
+
+    void getExtremes(Field & min, Field & max) const override { column_holder->getExtremes(min, max); }
+    bool valuesHaveFixedSize() const override { return column_holder->valuesHaveFixedSize(); }
+    bool isFixedAndContiguous() const override { return column_holder->isFixedAndContiguous(); }
+    size_t sizeOfValueIfFixed() const override { return column_holder->sizeOfValueIfFixed(); }
+    bool isNumeric() const override { return column_holder->isNumeric(); }
+
+    size_t byteSize() const override { return column_holder->byteSize(); }
+    size_t allocatedBytes() const override
+    {
+        return column_holder->allocatedBytes()
+               + index.allocatedBytes()
+               + (cached_null_mask ? cached_null_mask->allocatedBytes() : 0);
+    }
+    void forEachSubcolumn(IColumn::ColumnCallback callback) override
+    {
+        callback(column_holder);
+        index.setColumn(getRawColumnPtr());
+    }
+
+private:
+
+    ColumnPtr column_holder;
+    bool is_nullable;
+    ReverseIndex<UInt64, ColumnType> index;
+
+    /// For DataTypeNullable, stores null map.
+    mutable ColumnPtr cached_null_mask;
+    mutable ColumnPtr cached_column_nullable;
+
+    static size_t numSpecialValues(bool is_nullable) { return is_nullable ? 2 : 1; }
+    size_t numSpecialValues() const { return numSpecialValues(is_nullable); }
+
+    ColumnType * getRawColumnPtr() { return static_cast<ColumnType *>(column_holder->assumeMutable().get()); }
+    const ColumnType * getRawColumnPtr() const { return static_cast<const ColumnType *>(column_holder.get()); }
+
+    template <typename IndexType>
+    MutableColumnPtr uniqueInsertRangeImpl(
+        const IColumn & src,
+        size_t start,
+        size_t length,
+        size_t num_added_rows,
+        typename ColumnVector<IndexType>::MutablePtr && positions_column,
+        ReverseIndex<UInt64, ColumnType> * secondary_index,
+        size_t max_dictionary_size);
+};
+
+template <typename ColumnType>
+MutableColumnPtr ColumnUnique<ColumnType>::cloneEmpty() const
+{
+    return ColumnUnique<ColumnType>::create(column_holder->cloneResized(numSpecialValues()), is_nullable);
+}
+
+template <typename ColumnType>
+ColumnUnique<ColumnType>::ColumnUnique(const ColumnUnique & other)
+    : column_holder(other.column_holder)
+    , is_nullable(other.is_nullable)
+    , index(numSpecialValues(is_nullable), 0)
+{
+    index.setColumn(getRawColumnPtr());
+}
+
+template <typename ColumnType>
+ColumnUnique<ColumnType>::ColumnUnique(const IDataType & type)
+    : is_nullable(type.isNullable())
+    , index(numSpecialValues(is_nullable), 0)
+{
+    const auto & holder_type = is_nullable ? *static_cast<const DataTypeNullable &>(type).getNestedType() : type;
+    column_holder = holder_type.createColumn()->cloneResized(numSpecialValues());
+    index.setColumn(getRawColumnPtr());
+}
+
+template <typename ColumnType>
+ColumnUnique<ColumnType>::ColumnUnique(MutableColumnPtr && holder, bool is_nullable)
+    : column_holder(std::move(holder))
+    , is_nullable(is_nullable)
+    , index(numSpecialValues(is_nullable), 0)
+{
+    if (column_holder->size() < numSpecialValues())
+        throw Exception("Too small holder column for ColumnUnique.", ErrorCodes::ILLEGAL_COLUMN);
+    if (column_holder->isColumnNullable())
+        throw Exception("Holder column for ColumnUnique can't be nullable.", ErrorCodes::ILLEGAL_COLUMN);
+
+    index.setColumn(getRawColumnPtr());
+}
+
+template <typename ColumnType>
+const ColumnPtr & ColumnUnique<ColumnType>::getNestedColumn() const
+{
+    if (is_nullable)
+    {
+        size_t size = getRawColumnPtr()->size();
+        if (!cached_null_mask)
+        {
+            ColumnUInt8::MutablePtr null_mask = ColumnUInt8::create(size, UInt8(0));
+            null_mask->getData()[getNullValueIndex()] = 1;
+            cached_null_mask = std::move(null_mask);
+            cached_column_nullable = ColumnNullable::create(column_holder, cached_null_mask);
+        }
+
+        if (cached_null_mask->size() != size)
+        {
+            MutableColumnPtr null_mask = (*std::move(cached_null_mask)).mutate();
+            static_cast<ColumnUInt8 &>(*null_mask).getData().resize_fill(size);
+            cached_null_mask = std::move(null_mask);
+            cached_column_nullable = ColumnNullable::create(column_holder, cached_null_mask);
+        }
+
+        return cached_column_nullable;
+    }
+    return column_holder;
+}
+
+template <typename ColumnType>
+size_t ColumnUnique<ColumnType>::getNullValueIndex() const
+{
+    if (!is_nullable)
+        throw Exception("ColumnUnique can't contain null values.", ErrorCodes::LOGICAL_ERROR);
+
+    return 0;
+}
+
+template <typename ColumnType>
+size_t ColumnUnique<ColumnType>::uniqueInsert(const Field & x)
+{
+    if (x.getType() == Field::Types::Null)
+        return getNullValueIndex();
+
+    auto column = getRawColumnPtr();
+    auto prev_size = static_cast<UInt64>(column->size());
+
+    if ((*column)[getDefaultValueIndex()] == x)
+        return getDefaultValueIndex();
+
+    column->insert(x);
+    auto pos = index.insert(prev_size);
+    if (pos != prev_size)
+        column->popBack(1);
+
+    return pos;
+}
+
+template <typename ColumnType>
+size_t ColumnUnique<ColumnType>::uniqueInsertFrom(const IColumn & src, size_t n)
+{
+    if (is_nullable && src.isNullAt(n))
+        return getNullValueIndex();
+
+    if (auto * nullable = typeid_cast<const ColumnNullable *>(&src))
+        return uniqueInsertFrom(nullable->getNestedColumn(), n);
+
+    auto ref = src.getDataAt(n);
+    return uniqueInsertData(ref.data, ref.size);
+}
+
+template <typename ColumnType>
+size_t ColumnUnique<ColumnType>::uniqueInsertData(const char * pos, size_t length)
+{
+    auto column = getRawColumnPtr();
+
+    if (column->getDataAt(getDefaultValueIndex()) == StringRef(pos, length))
+        return getDefaultValueIndex();
+
+    UInt64 size = column->size();
+    UInt64 insertion_point = index.getInsertionPoint(StringRef(pos, length));
+
+    if (insertion_point == size)
+    {
+        column->insertData(pos, length);
+        index.insertFromLastRow();
+    }
+
+    return insertion_point;
+}
+
+template <typename ColumnType>
+size_t ColumnUnique<ColumnType>::uniqueInsertDataWithTerminatingZero(const char * pos, size_t length)
+{
+    if (std::is_same<ColumnType, ColumnString>::value)
+        return uniqueInsertData(pos, length - 1);
+
+    if (column_holder->valuesHaveFixedSize())
+        return uniqueInsertData(pos, length);
+
+    /// Don't know if data actually has terminating zero. So, insert it firstly.
+
+    auto column = getRawColumnPtr();
+    size_t prev_size = column->size();
+    column->insertDataWithTerminatingZero(pos, length);
+
+    if (column->compareAt(getDefaultValueIndex(), prev_size, *column, 1) == 0)
+    {
+        column->popBack(1);
+        return getDefaultValueIndex();
+    }
+
+    auto position = index.insert(prev_size);
+    if (position != prev_size)
+        column->popBack(1);
+
+    return static_cast<size_t>(position);
+}
+
+template <typename ColumnType>
+size_t ColumnUnique<ColumnType>::uniqueDeserializeAndInsertFromArena(const char * pos, const char *& new_pos)
+{
+    auto column = getRawColumnPtr();
+    size_t prev_size = column->size();
+    new_pos = column->deserializeAndInsertFromArena(pos);
+
+    if (column->compareAt(getDefaultValueIndex(), prev_size, *column, 1) == 0)
+    {
+        column->popBack(1);
+        return getDefaultValueIndex();
+    }
+
+    auto index_pos = index.insert(prev_size);
+    if (index_pos != prev_size)
+        column->popBack(1);
+
+    return static_cast<size_t>(index_pos);
+}
+
+template <typename IndexType>
+static void checkIndexes(const ColumnVector<IndexType> & indexes, size_t max_dictionary_size)
+{
+    auto & data = indexes.getData();
+    for (size_t i = 0; i < data.size(); ++i)
+    {
+        if (data[i] >= max_dictionary_size)
+        {
+            throw Exception("Found index " + toString(data[i]) + " at position " + toString(i)
+                            + " which is grated or equal than dictionary size " + toString(max_dictionary_size),
+                            ErrorCodes::LOGICAL_ERROR);
+        }
+    }
+}
+
+template <typename ColumnType>
+template <typename IndexType>
+MutableColumnPtr ColumnUnique<ColumnType>::uniqueInsertRangeImpl(
+    const IColumn & src,
+    size_t start,
+    size_t length,
+    size_t num_added_rows,
+    typename ColumnVector<IndexType>::MutablePtr && positions_column,
+    ReverseIndex<UInt64, ColumnType> * secondary_index,
+    size_t max_dictionary_size)
+{
+    const ColumnType * src_column;
+    const NullMap * null_map = nullptr;
+    auto & positions = positions_column->getData();
+
+    auto update_position = [&](UInt64 & next_position) -> MutableColumnPtr
+    {
+        constexpr auto next_size = NumberTraits::nextSize(sizeof(IndexType));
+        using SuperiorIndexType = typename NumberTraits::Construct<false, false, next_size>::Type;
+
+        ++next_position;
+
+        if (next_position > std::numeric_limits<IndexType>::max())
+        {
+            if (sizeof(SuperiorIndexType) == sizeof(IndexType))
+                throw Exception("Can't find superior index type for type " + demangle(typeid(IndexType).name()),
+                                ErrorCodes::LOGICAL_ERROR);
+
+            auto expanded_column = ColumnVector<SuperiorIndexType>::create(length);
+            auto & expanded_data = expanded_column->getData();
+            for (size_t i = 0; i < num_added_rows; ++i)
+                expanded_data[i] = positions[i];
+
+            return uniqueInsertRangeImpl<SuperiorIndexType>(
+                    src,
+                    start,
+                    length,
+                    num_added_rows,
+                    std::move(expanded_column),
+                    secondary_index,
+                    max_dictionary_size);
+        }
+
+        return nullptr;
+    };
+
+    if (auto nullable_column = typeid_cast<const ColumnNullable *>(&src))
+    {
+        src_column = typeid_cast<const ColumnType *>(&nullable_column->getNestedColumn());
+        null_map = &nullable_column->getNullMapData();
+    }
+    else
+        src_column = typeid_cast<const ColumnType *>(&src);
+
+    if (src_column == nullptr)
+        throw Exception("Invalid column type for ColumnUnique::insertRangeFrom. Expected " + column_holder->getName() +
+                        ", got " + src.getName(), ErrorCodes::ILLEGAL_COLUMN);
+
+    auto column = getRawColumnPtr();
+
+    UInt64 next_position = column->size();
+    if (secondary_index)
+        next_position += secondary_index->size();
+
+    auto check_inserted_position = [&next_position](UInt64 inserted_position)
+    {
+        if (inserted_position != next_position)
+            throw Exception("Inserted position " + toString(inserted_position)
+                            + " is not equal with expected " + toString(next_position), ErrorCodes::LOGICAL_ERROR);
+    };
+
+    auto insert_key = [&](const StringRef & ref, ReverseIndex<UInt64, ColumnType> * cur_index)
+    {
+        positions[num_added_rows] = next_position;
+        cur_index->getColumn()->insertData(ref.data, ref.size);
+        auto inserted_pos = cur_index->insertFromLastRow();
+        check_inserted_position(inserted_pos);
+        return update_position(next_position);
+    };
+
+    for (; num_added_rows < length; ++num_added_rows)
+    {
+        auto row = start + num_added_rows;
+
+        if (null_map && (*null_map)[row])
+            positions[num_added_rows] = getNullValueIndex();
+        else if (column->compareAt(getDefaultValueIndex(), row, *src_column, 1) == 0)
+            positions[num_added_rows] = getDefaultValueIndex();
+        else
+        {
+            auto ref = src_column->getDataAt(row);
+            auto cur_index = &index;
+            bool inserted = false;
+
+            while (!inserted)
+            {
+                auto insertion_point = cur_index->getInsertionPoint(ref);
+
+                if (insertion_point == cur_index->lastInsertionPoint())
+                {
+                    if (secondary_index && cur_index != secondary_index && next_position >= max_dictionary_size)
+                    {
+                        cur_index = secondary_index;
+                        continue;
+                    }
+
+                    if (auto res = insert_key(ref, cur_index))
+                        return res;
+                }
+                else
+                   positions[num_added_rows] = insertion_point;
+
+                inserted = true;
+            }
+        }
+    }
+
+    // checkIndexes(*positions_column, column->size() + (overflowed_keys ? overflowed_keys->size() : 0));
+    return std::move(positions_column);
+}
+
+template <typename ColumnType>
+MutableColumnPtr ColumnUnique<ColumnType>::uniqueInsertRangeFrom(const IColumn & src, size_t start, size_t length)
+{
+    auto callForType = [this, &src, start, length](auto x) -> MutableColumnPtr
+    {
+        size_t size = getRawColumnPtr()->size();
+
+        using IndexType = decltype(x);
+        if (size <= std::numeric_limits<IndexType>::max())
+        {
+            auto positions = ColumnVector<IndexType>::create(length);
+            return this->uniqueInsertRangeImpl<IndexType>(src, start, length, 0, std::move(positions), nullptr, 0);
+        }
+
+        return nullptr;
+    };
+
+    MutableColumnPtr positions_column;
+    if (!positions_column)
+        positions_column = callForType(UInt8());
+    if (!positions_column)
+        positions_column = callForType(UInt16());
+    if (!positions_column)
+        positions_column = callForType(UInt32());
+    if (!positions_column)
+        positions_column = callForType(UInt64());
+    if (!positions_column)
+        throw Exception("Can't find index type for ColumnUnique", ErrorCodes::LOGICAL_ERROR);
+
+    return positions_column;
+}
+
+template <typename ColumnType>
+IColumnUnique::IndexesWithOverflow ColumnUnique<ColumnType>::uniqueInsertRangeWithOverflow(
+    const IColumn & src,
+    size_t start,
+    size_t length,
+    size_t max_dictionary_size)
+{
+    auto overflowed_keys = column_holder->cloneEmpty();
+    auto overflowed_keys_ptr = typeid_cast<ColumnType *>(overflowed_keys.get());
+    if (!overflowed_keys_ptr)
+        throw Exception("Invalid keys type for ColumnUnique.", ErrorCodes::LOGICAL_ERROR);
+
+    auto callForType = [this, &src, start, length, overflowed_keys_ptr, max_dictionary_size](auto x) -> MutableColumnPtr
+    {
+        size_t size = getRawColumnPtr()->size();
+
+        using IndexType = decltype(x);
+        if (size <= std::numeric_limits<IndexType>::max())
+        {
+            auto positions = ColumnVector<IndexType>::create(length);
+            ReverseIndex<UInt64, ColumnType> secondary_index(0, max_dictionary_size);
+            secondary_index.setColumn(overflowed_keys_ptr);
+            return this->uniqueInsertRangeImpl<IndexType>(src, start, length, 0, std::move(positions),
+                                                          &secondary_index, max_dictionary_size);
+        }
+
+        return nullptr;
+    };
+
+    MutableColumnPtr positions_column;
+    if (!positions_column)
+        positions_column = callForType(UInt8());
+    if (!positions_column)
+        positions_column = callForType(UInt16());
+    if (!positions_column)
+        positions_column = callForType(UInt32());
+    if (!positions_column)
+        positions_column = callForType(UInt64());
+    if (!positions_column)
+        throw Exception("Can't find index type for ColumnUnique", ErrorCodes::LOGICAL_ERROR);
+
+    IColumnUnique::IndexesWithOverflow indexes_with_overflow;
+    indexes_with_overflow.indexes = std::move(positions_column);
+    indexes_with_overflow.overflowed_keys = std::move(overflowed_keys);
+    return indexes_with_overflow;
+}
+
+};

dbms/src/Columns/ColumnVector.cpp

@@ -17,6 +17,8 @@
 
 #if __SSE2__
     #include <emmintrin.h>
+#include <Columns/ColumnsCommon.h>
+
 #endif
 
 
@@ -230,6 +232,12 @@ ColumnPtr ColumnVector<T>::permute(const IColumn::Permutation & perm, size_t lim
     return std::move(res);
 }
 
+template <typename T>
+ColumnPtr ColumnVector<T>::index(const IColumn & indexes, size_t limit) const
+{
+    return selectIndexImpl(*this, indexes, limit);
+}
+
 template <typename T>
 ColumnPtr ColumnVector<T>::replicate(const IColumn::Offsets & offsets) const
 {

dbms/src/Columns/ColumnVector.h

@@ -252,6 +252,11 @@ public:
 
     ColumnPtr permute(const IColumn::Permutation & perm, size_t limit) const override;
 
+    ColumnPtr index(const IColumn & indexes, size_t limit) const override;
+
+    template <typename Type>
+    ColumnPtr indexImpl(const PaddedPODArray<Type> & indexes, size_t limit) const;
+
     ColumnPtr replicate(const IColumn::Offsets & offsets) const override;
 
     void getExtremes(Field & min, Field & max) const override;
@@ -295,5 +300,23 @@ protected:
     Container data;
 };
 
+template <typename T>
+template <typename Type>
+ColumnPtr ColumnVector<T>::indexImpl(const PaddedPODArray<Type> & indexes, size_t limit) const
+{
+    size_t size = indexes.size();
+
+    if (limit == 0)
+        limit = size;
+    else
+        limit = std::min(size, limit);
+
+    auto res = this->create(limit);
+    typename Self::Container & res_data = res->getData();
+    for (size_t i = 0; i < limit; ++i)
+        res_data[i] = data[indexes[i]];
+
+    return std::move(res);
+}
 
 }

dbms/src/Columns/ColumnWithDictionary.cpp

@@ -0,0 +1,619 @@
+#include <Columns/ColumnWithDictionary.h>
+#include <Columns/ColumnsNumber.h>
+#include <DataStreams/ColumnGathererStream.h>
+#include <DataTypes/NumberTraits.h>
+#include <Common/HashTable/HashMap.h>
+
+namespace DB
+{
+
+namespace
+{
+    template <typename T>
+    PaddedPODArray<T> * getIndexesData(IColumn & indexes)
+    {
+        auto * column = typeid_cast<ColumnVector<T> *>(&indexes);
+        if (column)
+            return &column->getData();
+
+        return nullptr;
+    }
+
+    template <typename T>
+    MutableColumnPtr mapUniqueIndexImplRef(PaddedPODArray<T> & index)
+    {
+        PaddedPODArray<T> copy(index.cbegin(), index.cend());
+
+        HashMap<T, T> hash_map;
+        for (auto val : index)
+            hash_map.insert({val, hash_map.size()});
+
+        auto res_col = ColumnVector<T>::create();
+        auto & data = res_col->getData();
+
+        data.resize(hash_map.size());
+        for (auto val : hash_map)
+            data[val.second] = val.first;
+
+        for (auto & ind : index)
+            ind = hash_map[ind];
+
+        for (size_t i = 0; i < index.size(); ++i)
+            if (data[index[i]] != copy[i])
+                throw Exception("Expected " + toString(data[index[i]]) + ", but got " + toString(copy[i]), ErrorCodes::LOGICAL_ERROR);
+
+        return std::move(res_col);
+    }
+
+    template <typename T>
+    MutableColumnPtr mapUniqueIndexImpl(PaddedPODArray<T> & index)
+    {
+        if (index.empty())
+            return ColumnVector<T>::create();
+
+        auto size = index.size();
+
+        T max_val = index[0];
+        for (size_t i = 1; i < size; ++i)
+            max_val = std::max(max_val, index[i]);
+
+        /// May happen when dictionary is shared.
+        if (max_val > size)
+            return mapUniqueIndexImplRef(index);
+
+        auto map_size = UInt64(max_val) + 1;
+        PaddedPODArray<T> map(map_size, 0);
+        T zero_pos_value = index[0];
+        index[0] = 0;
+        T cur_pos = 0;
+        for (size_t i = 1; i < size; ++i)
+        {
+            T val = index[i];
+            if (val != zero_pos_value && map[val] == 0)
+            {
+                ++cur_pos;
+                map[val] = cur_pos;
+            }
+
+            index[i] = map[val];
+        }
+
+        auto res_col = ColumnVector<T>::create(UInt64(cur_pos) + 1);
+        auto & data = res_col->getData();
+        data[0] = zero_pos_value;
+        for (size_t i = 0; i < map_size; ++i)
+        {
+            auto val = map[i];
+            if (val)
+                data[val] = static_cast<T>(i);
+        }
+
+        return std::move(res_col);
+    }
+
+    /// Returns unique values of column. Write new index to column.
+    MutableColumnPtr mapUniqueIndex(IColumn & column)
+    {
+        if (auto * data_uint8 = getIndexesData<UInt8>(column))
+            return mapUniqueIndexImpl(*data_uint8);
+        else if (auto * data_uint16 = getIndexesData<UInt16>(column))
+            return mapUniqueIndexImpl(*data_uint16);
+        else if (auto * data_uint32 = getIndexesData<UInt32>(column))
+            return mapUniqueIndexImpl(*data_uint32);
+        else if (auto * data_uint64 = getIndexesData<UInt64>(column))
+            return mapUniqueIndexImpl(*data_uint64);
+        else
+            throw Exception("Indexes column for getUniqueIndex must be ColumnUInt, got" + column.getName(),
+                            ErrorCodes::LOGICAL_ERROR);
+    }
+}
+
+
+ColumnWithDictionary::ColumnWithDictionary(MutableColumnPtr && column_unique_, MutableColumnPtr && indexes_)
+    : dictionary(std::move(column_unique_)), idx(std::move(indexes_))
+{
+    idx.check(getDictionary().size());
+}
+
+void ColumnWithDictionary::insert(const Field & x)
+{
+    compactIfSharedDictionary();
+    idx.insertPosition(dictionary.getColumnUnique().uniqueInsert(x));
+    idx.check(getDictionary().size());
+}
+
+void ColumnWithDictionary::insertDefault()
+{
+    idx.insertPosition(getDictionary().getDefaultValueIndex());
+}
+
+void ColumnWithDictionary::insertFrom(const IColumn & src, size_t n)
+{
+    auto * src_with_dict = typeid_cast<const ColumnWithDictionary *>(&src);
+
+    if (!src_with_dict)
+        throw Exception("Expected ColumnWithDictionary, got" + src.getName(), ErrorCodes::ILLEGAL_COLUMN);
+
+    size_t position = src_with_dict->getIndexes().getUInt(n);
+
+    if (&src_with_dict->getDictionary() == &getDictionary())
+    {
+        /// Dictionary is shared with src column. Insert only index.
+        idx.insertPosition(position);
+    }
+    else
+    {
+        compactIfSharedDictionary();
+        const auto & nested = *src_with_dict->getDictionary().getNestedColumn();
+        idx.insertPosition(dictionary.getColumnUnique().uniqueInsertFrom(nested, position));
+    }
+
+    idx.check(getDictionary().size());
+}
+
+void ColumnWithDictionary::insertFromFullColumn(const IColumn & src, size_t n)
+{
+    compactIfSharedDictionary();
+    idx.insertPosition(dictionary.getColumnUnique().uniqueInsertFrom(src, n));
+    idx.check(getDictionary().size());
+}
+
+void ColumnWithDictionary::insertRangeFrom(const IColumn & src, size_t start, size_t length)
+{
+    auto * src_with_dict = typeid_cast<const ColumnWithDictionary *>(&src);
+
+    if (!src_with_dict)
+        throw Exception("Expected ColumnWithDictionary, got" + src.getName(), ErrorCodes::ILLEGAL_COLUMN);
+
+    if (&src_with_dict->getDictionary() == &getDictionary())
+    {
+        /// Dictionary is shared with src column. Insert only indexes.
+        idx.insertPositionsRange(src_with_dict->getIndexes(), start, length);
+    }
+    else
+    {
+        compactIfSharedDictionary();
+
+        /// TODO: Support native insertion from other unique column. It will help to avoid null map creation.
+
+        auto sub_idx = (*src_with_dict->getIndexes().cut(start, length)).mutate();
+        auto idx_map = mapUniqueIndex(*sub_idx);
+
+        auto src_nested = src_with_dict->getDictionary().getNestedColumn();
+        auto used_keys = src_nested->index(*idx_map, 0);
+
+        auto inserted_indexes = dictionary.getColumnUnique().uniqueInsertRangeFrom(*used_keys, 0, used_keys->size());
+        idx.insertPositionsRange(*inserted_indexes->index(*sub_idx, 0), 0, length);
+    }
+    idx.check(getDictionary().size());
+}
+
+void ColumnWithDictionary::insertRangeFromFullColumn(const IColumn & src, size_t start, size_t length)
+{
+    compactIfSharedDictionary();
+    auto inserted_indexes = dictionary.getColumnUnique().uniqueInsertRangeFrom(src, start, length);
+    idx.insertPositionsRange(*inserted_indexes, 0, length);
+    idx.check(getDictionary().size());
+}
+
+void ColumnWithDictionary::insertRangeFromDictionaryEncodedColumn(const IColumn & keys, const IColumn & positions)
+{
+    Index(positions.getPtr()).check(keys.size());
+    compactIfSharedDictionary();
+    auto inserted_indexes = dictionary.getColumnUnique().uniqueInsertRangeFrom(keys, 0, keys.size());
+    idx.insertPositionsRange(*inserted_indexes->index(positions, 0), 0, positions.size());
+    idx.check(getDictionary().size());
+}
+
+void ColumnWithDictionary::insertData(const char * pos, size_t length)
+{
+    compactIfSharedDictionary();
+    idx.insertPosition(dictionary.getColumnUnique().uniqueInsertData(pos, length));
+    idx.check(getDictionary().size());
+}
+
+void ColumnWithDictionary::insertDataWithTerminatingZero(const char * pos, size_t length)
+{
+    compactIfSharedDictionary();
+    idx.insertPosition(dictionary.getColumnUnique().uniqueInsertDataWithTerminatingZero(pos, length));
+    idx.check(getDictionary().size());
+}
+
+StringRef ColumnWithDictionary::serializeValueIntoArena(size_t n, Arena & arena, char const *& begin) const
+{
+    return getDictionary().serializeValueIntoArena(getIndexes().getUInt(n), arena, begin);
+}
+
+const char * ColumnWithDictionary::deserializeAndInsertFromArena(const char * pos)
+{
+    compactIfSharedDictionary();
+
+    const char * new_pos;
+    idx.insertPosition(dictionary.getColumnUnique().uniqueDeserializeAndInsertFromArena(pos, new_pos));
+
+    idx.check(getDictionary().size());
+    return new_pos;
+}
+
+void ColumnWithDictionary::gather(ColumnGathererStream & gatherer)
+{
+    gatherer.gather(*this);
+}
+
+MutableColumnPtr ColumnWithDictionary::cloneResized(size_t size) const
+{
+    auto unique_ptr = dictionary.getColumnUniquePtr();
+    return ColumnWithDictionary::create((*std::move(unique_ptr)).mutate(), getIndexes().cloneResized(size));
+}
+
+int ColumnWithDictionary::compareAt(size_t n, size_t m, const IColumn & rhs, int nan_direction_hint) const
+{
+    const auto & column_with_dictionary = static_cast<const ColumnWithDictionary &>(rhs);
+    size_t n_index = getIndexes().getUInt(n);
+    size_t m_index = column_with_dictionary.getIndexes().getUInt(m);
+    return getDictionary().compareAt(n_index, m_index, column_with_dictionary.getDictionary(), nan_direction_hint);
+}
+
+void ColumnWithDictionary::getPermutation(bool reverse, size_t limit, int nan_direction_hint, Permutation & res) const
+{
+    if (limit == 0)
+        limit = size();
+
+    size_t unique_limit = std::min(limit, getDictionary().size());
+    Permutation unique_perm;
+    getDictionary().getNestedColumn()->getPermutation(reverse, unique_limit, nan_direction_hint, unique_perm);
+
+    /// TODO: optimize with sse.
+
+    /// Get indexes per row in column_unique.
+    std::vector<std::vector<size_t>> indexes_per_row(getDictionary().size());
+    size_t indexes_size = getIndexes().size();
+    for (size_t row = 0; row < indexes_size; ++row)
+        indexes_per_row[getIndexes().getUInt(row)].push_back(row);
+
+    /// Replicate permutation.
+    size_t perm_size = std::min(indexes_size, limit);
+    res.resize(perm_size);
+    size_t perm_index = 0;
+    for (size_t row = 0; row < indexes_size && perm_index < perm_size; ++row)
+    {
+        const auto & row_indexes = indexes_per_row[unique_perm[row]];
+        for (auto row_index : row_indexes)
+        {
+            res[perm_index] = row_index;
+            ++perm_index;
+
+            if (perm_index == perm_size)
+                break;
+        }
+    }
+}
+
+std::vector<MutableColumnPtr> ColumnWithDictionary::scatter(ColumnIndex num_columns, const Selector & selector) const
+{
+    auto columns = getIndexes().scatter(num_columns, selector);
+    for (auto & column : columns)
+    {
+        auto unique_ptr = dictionary.getColumnUniquePtr();
+        column = ColumnWithDictionary::create((*std::move(unique_ptr)).mutate(), std::move(column));
+    }
+
+    return columns;
+}
+
+void ColumnWithDictionary::setSharedDictionary(const ColumnPtr & column_unique)
+{
+    if (!empty())
+        throw Exception("Can't set ColumnUnique for ColumnWithDictionary because is't not empty.",
+                        ErrorCodes::LOGICAL_ERROR);
+
+    dictionary.setShared(column_unique);
+}
+
+ColumnWithDictionary::MutablePtr ColumnWithDictionary::compact()
+{
+    auto positions = idx.getPositions();
+    /// Create column with new indexes and old dictionary.
+    auto column = ColumnWithDictionary::create(getDictionary().assumeMutable(), (*std::move(positions)).mutate());
+    /// Will create new dictionary.
+    column->compactInplace();
+
+    return column;
+}
+
+ColumnWithDictionary::MutablePtr ColumnWithDictionary::cutAndCompact(size_t start, size_t length) const
+{
+    auto sub_positions = (*idx.getPositions()->cut(start, length)).mutate();
+    /// Create column with new indexes and old dictionary.
+    auto column = ColumnWithDictionary::create(getDictionary().assumeMutable(), std::move(sub_positions));
+    /// Will create new dictionary.
+    column->compactInplace();
+
+    return column;
+}
+
+void ColumnWithDictionary::compactInplace()
+{
+    auto positions = idx.detachPositions();
+    dictionary.compact(positions);
+    idx.attachPositions(std::move(positions));
+}
+
+void ColumnWithDictionary::compactIfSharedDictionary()
+{
+    if (dictionary.isShared())
+        compactInplace();
+}
+
+
+ColumnWithDictionary::DictionaryEncodedColumn
+ColumnWithDictionary::getMinimalDictionaryEncodedColumn(size_t offset, size_t limit) const
+{
+    MutableColumnPtr sub_indexes = (*std::move(idx.getPositions()->cut(offset, limit))).mutate();
+    auto indexes_map = mapUniqueIndex(*sub_indexes);
+    auto sub_keys = getDictionary().getNestedColumn()->index(*indexes_map, 0);
+
+    return {std::move(sub_keys), std::move(sub_indexes)};
+}
+
+
+ColumnWithDictionary::Index::Index() : positions(ColumnUInt8::create()), size_of_type(sizeof(UInt8)) {}
+
+ColumnWithDictionary::Index::Index(MutableColumnPtr && positions) : positions(std::move(positions))
+{
+    updateSizeOfType();
+}
+
+ColumnWithDictionary::Index::Index(ColumnPtr positions) : positions(std::move(positions))
+{
+    updateSizeOfType();
+}
+
+template <typename Callback>
+void ColumnWithDictionary::Index::callForType(Callback && callback, size_t size_of_type)
+{
+    switch (size_of_type)
+    {
+        case sizeof(UInt8): { callback(UInt8()); break; }
+        case sizeof(UInt16): { callback(UInt16()); break; }
+        case sizeof(UInt32): { callback(UInt32()); break; }
+        case sizeof(UInt64): { callback(UInt64()); break; }
+        default: {
+            throw Exception("Unexpected size of index type for ColumnWithDictionary: " + toString(size_of_type),
+                            ErrorCodes::LOGICAL_ERROR);
+        }
+    }
+}
+
+size_t ColumnWithDictionary::Index::getSizeOfIndexType(const IColumn & column, size_t hint)
+{
+    auto checkFor = [&](auto type) { return typeid_cast<const ColumnVector<decltype(type)> *>(&column) != nullptr; };
+    auto tryGetSizeFor = [&](auto type) -> size_t { return checkFor(type) ? sizeof(decltype(type)) : 0; };
+
+    if (hint)
+    {
+        size_t size = 0;
+        callForType([&](auto type) { size = tryGetSizeFor(type); }, hint);
+
+        if (size)
+            return size;
+    }
+
+    if (auto size = tryGetSizeFor(UInt8()))
+        return size;
+    if (auto size = tryGetSizeFor(UInt16()))
+        return size;
+    if (auto size = tryGetSizeFor(UInt32()))
+        return size;
+    if (auto size = tryGetSizeFor(UInt64()))
+        return size;
+
+    throw Exception("Unexpected indexes type for ColumnWithDictionary. Expected UInt, got " + column.getName(),
+                    ErrorCodes::ILLEGAL_COLUMN);
+}
+
+void ColumnWithDictionary::Index::attachPositions(ColumnPtr positions_)
+{
+    positions = std::move(positions_);
+    updateSizeOfType();
+}
+
+template <typename IndexType>
+typename ColumnVector<IndexType>::Container & ColumnWithDictionary::Index::getPositionsData()
+{
+    auto * positions_ptr = typeid_cast<ColumnVector<IndexType> *>(positions->assumeMutable().get());
+    if (!positions_ptr)
+        throw Exception("Invalid indexes type for ColumnWithDictionary."
+                        " Expected UInt" + toString(8 * sizeof(IndexType)) + ", got " + positions->getName(),
+                        ErrorCodes::LOGICAL_ERROR);
+
+    return positions_ptr->getData();
+}
+
+template <typename IndexType>
+void ColumnWithDictionary::Index::convertPositions()
+{
+    auto convert = [&](auto x)
+    {
+        using CurIndexType = decltype(x);
+        auto & data = getPositionsData<CurIndexType>();
+
+        if (sizeof(CurIndexType) > sizeof(IndexType))
+            throw Exception("Converting indexes to smaller type: from " + toString(sizeof(CurIndexType)) +
+                            " to " + toString(sizeof(IndexType)), ErrorCodes::LOGICAL_ERROR);
+
+        if (sizeof(CurIndexType) != sizeof(IndexType))
+        {
+            size_t size = data.size();
+            auto new_positions = ColumnVector<IndexType>::create(size);
+            auto & new_data = new_positions->getData();
+
+            /// TODO: Optimize with SSE?
+            for (size_t i = 0; i < size; ++i)
+                new_data[i] = data[i];
+
+            positions = std::move(new_positions);
+            size_of_type = sizeof(IndexType);
+        }
+    };
+
+    callForType(std::move(convert), size_of_type);
+
+    checkSizeOfType();
+}
+
+void ColumnWithDictionary::Index::expandType()
+{
+    auto expand = [&](auto type)
+    {
+        using CurIndexType = decltype(type);
+        constexpr auto next_size = NumberTraits::nextSize(sizeof(CurIndexType));
+        if (next_size == sizeof(CurIndexType))
+            throw Exception("Can't expand indexes type for ColumnWithDictionary from type: "
+                            + demangle(typeid(CurIndexType).name()), ErrorCodes::LOGICAL_ERROR);
+
+        using NewIndexType = typename NumberTraits::Construct<false, false, next_size>::Type;
+        convertPositions<NewIndexType>();
+    };
+
+    callForType(std::move(expand), size_of_type);
+}
+
+UInt64 ColumnWithDictionary::Index::getMaxPositionForCurrentType() const
+{
+    UInt64 value = 0;
+    callForType([&](auto type) { value = std::numeric_limits<decltype(type)>::max(); }, size_of_type);
+    return value;
+}
+
+void ColumnWithDictionary::Index::insertPosition(UInt64 position)
+{
+    while (position > getMaxPositionForCurrentType())
+        expandType();
+
+    positions->assumeMutableRef().insert(UInt64(position));
+    checkSizeOfType();
+}
+
+void ColumnWithDictionary::Index::insertPositionsRange(const IColumn & column, size_t offset, size_t limit)
+{
+    auto insertForType = [&](auto type)
+    {
+        using ColumnType = decltype(type);
+        const auto * column_ptr = typeid_cast<const ColumnVector<ColumnType> *>(&column);
+
+        if (!column_ptr)
+            return false;
+
+        if (size_of_type < sizeof(ColumnType))
+            convertPositions<ColumnType>();
+
+        if (size_of_type == sizeof(ColumnType))
+            positions->assumeMutableRef().insertRangeFrom(column, offset, limit);
+        else
+        {
+            auto copy = [&](auto cur_type)
+            {
+                using CurIndexType = decltype(cur_type);
+                auto & positions_data = getPositionsData<CurIndexType>();
+                const auto & column_data = column_ptr->getData();
+
+                size_t size = positions_data.size();
+                positions_data.resize(size + limit);
+
+                for (size_t i = 0; i < limit; ++i)
+                    positions_data[size + i] = column_data[offset + i];
+            };
+
+            callForType(std::move(copy), size_of_type);
+        }
+
+        return true;
+    };
+
+    if (!insertForType(UInt8()) &&
+        !insertForType(UInt16()) &&
+        !insertForType(UInt32()) &&
+        !insertForType(UInt64()))
+        throw Exception("Invalid column for ColumnWithDictionary index. Expected UInt, got " + column.getName(),
+                        ErrorCodes::ILLEGAL_COLUMN);
+
+    checkSizeOfType();
+}
+
+void ColumnWithDictionary::Index::check(size_t /*max_dictionary_size*/)
+{
+    /// TODO: remove
+    /*
+    auto check = [&](auto cur_type)
+    {
+        using CurIndexType = decltype(cur_type);
+        auto & positions_data = getPositionsData<CurIndexType>();
+
+        for (size_t i = 0; i < positions_data.size(); ++i)
+        {
+            if (positions_data[i] >= max_dictionary_size)
+            {
+                throw Exception("Found index " + toString(positions_data[i]) + " at position " + toString(i)
+                                + " which is grated or equal than dictionary size " + toString(max_dictionary_size),
+                                ErrorCodes::LOGICAL_ERROR);
+            }
+        }
+    };
+
+    callForType(std::move(check), size_of_type);
+     */
+}
+
+void ColumnWithDictionary::Index::checkSizeOfType()
+{
+    if (size_of_type != getSizeOfIndexType(*positions, size_of_type))
+        throw Exception("Invalid size of type. Expected "  + toString(8 * size_of_type) +
+                        ", but positions are " + positions->getName(), ErrorCodes::LOGICAL_ERROR);
+}
+
+
+ColumnWithDictionary::Dictionary::Dictionary(MutableColumnPtr && column_unique_)
+    : column_unique(std::move(column_unique_))
+{
+    checkColumn(*column_unique);
+}
+ColumnWithDictionary::Dictionary::Dictionary(ColumnPtr column_unique_)
+    : column_unique(std::move(column_unique_))
+{
+    checkColumn(*column_unique);
+}
+
+void ColumnWithDictionary::Dictionary::checkColumn(const IColumn & column)
+{
+
+    if (!dynamic_cast<const IColumnUnique *>(&column))
+        throw Exception("ColumnUnique expected as an argument of ColumnWithDictionary.", ErrorCodes::ILLEGAL_COLUMN);
+}
+
+void ColumnWithDictionary::Dictionary::setShared(const ColumnPtr & dictionary)
+{
+    checkColumn(*dictionary);
+
+    column_unique = dictionary;
+    shared = true;
+}
+
+void ColumnWithDictionary::Dictionary::compact(ColumnPtr & positions)
+{
+    auto new_column_unique = column_unique->cloneEmpty();
+
+    auto & unique = getColumnUnique();
+    auto & new_unique = static_cast<IColumnUnique &>(*new_column_unique);
+
+    auto indexes = mapUniqueIndex(positions->assumeMutableRef());
+    auto sub_keys = unique.getNestedColumn()->index(*indexes, 0);
+    auto new_indexes = new_unique.uniqueInsertRangeFrom(*sub_keys, 0, sub_keys->size());
+
+    positions = (*new_indexes->index(*positions, 0)).mutate();
+    column_unique = std::move(new_column_unique);
+
+    shared = false;
+}
+
+}

dbms/src/Columns/ColumnWithDictionary.h

@@ -0,0 +1,248 @@
+#pragma once
+#include <Columns/IColumn.h>
+#include <Columns/IColumnUnique.h>
+#include <Common/typeid_cast.h>
+#include <AggregateFunctions/AggregateFunctionCount.h>
+#include "ColumnsNumber.h"
+
+namespace DB
+{
+
+namespace ErrorCodes
+{
+    extern const int ILLEGAL_COLUMN;
+}
+
+class ColumnWithDictionary final : public COWPtrHelper<IColumn, ColumnWithDictionary>
+{
+    friend class COWPtrHelper<IColumn, ColumnWithDictionary>;
+
+    ColumnWithDictionary(MutableColumnPtr && column_unique, MutableColumnPtr && indexes);
+    ColumnWithDictionary(const ColumnWithDictionary & other) = default;
+
+public:
+    /** Create immutable column using immutable arguments. This arguments may be shared with other columns.
+      * Use IColumn::mutate in order to make mutable column and mutate shared nested columns.
+      */
+    using Base = COWPtrHelper<IColumn, ColumnWithDictionary>;
+    static Ptr create(const ColumnPtr & column_unique_, const ColumnPtr & indexes_)
+    {
+        return ColumnWithDictionary::create(column_unique_->assumeMutable(), indexes_->assumeMutable());
+    }
+
+    template <typename ... Args, typename = typename std::enable_if<IsMutableColumns<Args ...>::value>::type>
+    static MutablePtr create(Args &&... args) { return Base::create(std::forward<Args>(args)...); }
+
+
+    std::string getName() const override { return "ColumnWithDictionary"; }
+    const char * getFamilyName() const override { return "ColumnWithDictionary"; }
+
+    ColumnPtr convertToFullColumn() const { return getDictionary().getNestedColumn()->index(getIndexes(), 0); }
+    ColumnPtr convertToFullColumnIfWithDictionary() const override { return convertToFullColumn(); }
+
+    MutableColumnPtr cloneResized(size_t size) const override;
+    size_t size() const override { return getIndexes().size(); }
+
+    Field operator[](size_t n) const override { return getDictionary()[getIndexes().getUInt(n)]; }
+    void get(size_t n, Field & res) const override { getDictionary().get(getIndexes().getUInt(n), res); }
+
+    StringRef getDataAt(size_t n) const override { return getDictionary().getDataAt(getIndexes().getUInt(n)); }
+    StringRef getDataAtWithTerminatingZero(size_t n) const override
+    {
+        return getDictionary().getDataAtWithTerminatingZero(getIndexes().getUInt(n));
+    }
+
+    UInt64 get64(size_t n) const override { return getDictionary().get64(getIndexes().getUInt(n)); }
+    UInt64 getUInt(size_t n) const override { return getDictionary().getUInt(getIndexes().getUInt(n)); }
+    Int64 getInt(size_t n) const override { return getDictionary().getInt(getIndexes().getUInt(n)); }
+    bool isNullAt(size_t n) const override { return getDictionary().isNullAt(getIndexes().getUInt(n)); }
+    ColumnPtr cut(size_t start, size_t length) const override
+    {
+        return ColumnWithDictionary::create(dictionary.getColumnUniquePtr(), getIndexes().cut(start, length));
+    }
+
+    void insert(const Field & x) override;
+    void insertDefault() override;
+
+    void insertFrom(const IColumn & src, size_t n) override;
+    void insertFromFullColumn(const IColumn & src, size_t n);
+
+    void insertRangeFrom(const IColumn & src, size_t start, size_t length) override;
+    void insertRangeFromFullColumn(const IColumn & src, size_t start, size_t length);
+    void insertRangeFromDictionaryEncodedColumn(const IColumn & keys, const IColumn & positions);
+
+    void insertData(const char * pos, size_t length) override;
+    void insertDataWithTerminatingZero(const char * pos, size_t length) override;
+
+
+    void popBack(size_t n) override { idx.popBack(n); }
+
+    StringRef serializeValueIntoArena(size_t n, Arena & arena, char const *& begin) const override;
+
+    const char * deserializeAndInsertFromArena(const char * pos) override;
+
+    void updateHashWithValue(size_t n, SipHash & hash) const override
+    {
+        return getDictionary().updateHashWithValue(getIndexes().getUInt(n), hash);
+    }
+
+    ColumnPtr filter(const Filter & filt, ssize_t result_size_hint) const override
+    {
+        return ColumnWithDictionary::create(dictionary.getColumnUniquePtr(), getIndexes().filter(filt, result_size_hint));
+    }
+
+    ColumnPtr permute(const Permutation & perm, size_t limit) const override
+    {
+        return ColumnWithDictionary::create(dictionary.getColumnUniquePtr(), getIndexes().permute(perm, limit));
+    }
+
+    ColumnPtr index(const IColumn & indexes_, size_t limit) const override
+    {
+        return ColumnWithDictionary::create(dictionary.getColumnUniquePtr(), getIndexes().index(indexes_, limit));
+    }
+
+    int compareAt(size_t n, size_t m, const IColumn & rhs, int nan_direction_hint) const override;
+
+    void getPermutation(bool reverse, size_t limit, int nan_direction_hint, Permutation & res) const override;
+
+    ColumnPtr replicate(const Offsets & offsets) const override
+    {
+        return ColumnWithDictionary::create(dictionary.getColumnUniquePtr(), getIndexes().replicate(offsets));
+    }
+
+    std::vector<MutableColumnPtr> scatter(ColumnIndex num_columns, const Selector & selector) const override;
+
+    void gather(ColumnGathererStream & gatherer_stream) override ;
+    void getExtremes(Field & min, Field & max) const override {
+        return getDictionary().index(getIndexes(), 0)->getExtremes(min, max); /// TODO: optimize
+    }
+
+    void reserve(size_t n) override { idx.reserve(n); }
+
+    size_t byteSize() const override { return idx.getPositions()->byteSize() + getDictionary().byteSize(); }
+    size_t allocatedBytes() const override { return idx.getPositions()->allocatedBytes() + getDictionary().allocatedBytes(); }
+
+    void forEachSubcolumn(ColumnCallback callback) override
+    {
+        callback(idx.getPositionsPtr());
+
+        /// Column doesn't own dictionary if it's shared.
+        if (!dictionary.isShared())
+            callback(dictionary.getColumnUniquePtr());
+    }
+
+    bool valuesHaveFixedSize() const override { return getDictionary().valuesHaveFixedSize(); }
+    bool isFixedAndContiguous() const override { return getDictionary().isFixedAndContiguous(); }
+    size_t sizeOfValueIfFixed() const override { return getDictionary().sizeOfValueIfFixed(); }
+    bool isNumeric() const override { return getDictionary().isNumeric(); }
+    bool withDictionary() const override { return true; }
+
+    const IColumnUnique & getDictionary() const { return dictionary.getColumnUnique(); }
+    /// IColumnUnique & getUnique() { return static_cast<IColumnUnique &>(*column_unique->assumeMutable()); }
+    /// ColumnPtr getUniquePtr() const { return column_unique; }
+
+    /// IColumn & getIndexes() { return idx.getPositions()->assumeMutableRef(); }
+    const IColumn & getIndexes() const { return *idx.getPositions(); }
+    const ColumnPtr & getIndexesPtr() const { return idx.getPositions(); }
+
+    ///void setIndexes(MutableColumnPtr && indexes_) { indexes = std::move(indexes_); }
+
+    /// Set shared ColumnUnique for empty column with dictionary.
+    void setSharedDictionary(const ColumnPtr & column_unique);
+
+    /// Create column new dictionary with only keys that are mentioned in index.
+    MutablePtr compact();
+
+    /// Cut + compact.
+    MutablePtr cutAndCompact(size_t start, size_t length) const;
+
+    struct DictionaryEncodedColumn
+    {
+        ColumnPtr dictionary;
+        ColumnPtr indexes;
+    };
+
+    DictionaryEncodedColumn getMinimalDictionaryEncodedColumn(size_t offset, size_t limit) const;
+
+    class Index
+    {
+    public:
+        Index();
+        Index(const Index & other) = default;
+        explicit Index(MutableColumnPtr && positions);
+        explicit Index(ColumnPtr positions);
+
+        const ColumnPtr & getPositions() const { return positions; }
+        ColumnPtr & getPositionsPtr() { return positions; }
+        void insertPosition(UInt64 position);
+        void insertPositionsRange(const IColumn & column, size_t offset, size_t limit);
+
+        void popBack(size_t n) { positions->assumeMutableRef().popBack(n); }
+        void reserve(size_t n) { positions->assumeMutableRef().reserve(n); }
+
+        UInt64 getMaxPositionForCurrentType() const;
+
+        static size_t getSizeOfIndexType(const IColumn & column, size_t hint);
+
+        void check(size_t max_dictionary_size);
+        void checkSizeOfType();
+
+        ColumnPtr detachPositions() { return std::move(positions); }
+        void attachPositions(ColumnPtr positions_);
+
+    private:
+        ColumnPtr positions;
+        size_t size_of_type = 0;
+
+        void updateSizeOfType() { size_of_type = getSizeOfIndexType(*positions, size_of_type); }
+        void expandType();
+
+        template <typename IndexType>
+        typename ColumnVector<IndexType>::Container & getPositionsData();
+
+        template <typename IndexType>
+        void convertPositions();
+
+        template <typename Callback>
+        static void callForType(Callback && callback, size_t size_of_type);
+    };
+
+private:
+    class Dictionary
+    {
+    public:
+        Dictionary(const Dictionary & other) = default;
+        explicit Dictionary(MutableColumnPtr && column_unique);
+        explicit Dictionary(ColumnPtr column_unique);
+
+        const ColumnPtr & getColumnUniquePtr() const { return column_unique; }
+        ColumnPtr & getColumnUniquePtr() { return column_unique; }
+
+        const IColumnUnique & getColumnUnique() const { return static_cast<const IColumnUnique &>(*column_unique); }
+        IColumnUnique & getColumnUnique() { return static_cast<IColumnUnique &>(column_unique->assumeMutableRef()); }
+
+        /// Dictionary may be shared for several mutable columns.
+        /// Immutable columns may have the same column unique, which isn't necessarily shared dictionary.
+        void setShared(const ColumnPtr & dictionary);
+        bool isShared() const { return shared; }
+
+        /// Create new dictionary with only keys that are mentioned in positions.
+        void compact(ColumnPtr & positions);
+
+    private:
+        ColumnPtr column_unique;
+        bool shared = false;
+
+        void checkColumn(const IColumn & column);
+    };
+
+    Dictionary dictionary;
+    Index idx;
+
+    void compactInplace();
+    void compactIfSharedDictionary();
+};
+
+
+
+}

dbms/src/Columns/ColumnsCommon.cpp

@@ -3,6 +3,10 @@
 #endif
 
 #include <Columns/IColumn.h>
+#include <Common/typeid_cast.h>
+#include <Columns/ColumnVector.h>
+#include <Common/HashTable/HashSet.h>
+#include <Common/HashTable/HashMap.h>
 
 
 namespace DB
@@ -308,4 +312,22 @@ INSTANTIATE(Float64)
 
 #undef INSTANTIATE
 
+namespace detail
+{
+    template <typename T>
+    const PaddedPODArray<T> * getIndexesData(const IColumn & indexes)
+    {
+        auto * column = typeid_cast<const ColumnVector<T> *>(&indexes);
+        if (column)
+            return &column->getData();
+
+        return nullptr;
+    }
+
+    template const PaddedPODArray<UInt8> * getIndexesData<UInt8>(const IColumn & indexes);
+    template const PaddedPODArray<UInt16> * getIndexesData<UInt16>(const IColumn & indexes);
+    template const PaddedPODArray<UInt32> * getIndexesData<UInt32>(const IColumn & indexes);
+    template const PaddedPODArray<UInt64> * getIndexesData<UInt64>(const IColumn & indexes);
+}
+
 }

dbms/src/Columns/ColumnsCommon.h

@@ -8,6 +8,11 @@
 namespace DB
 {
 
+namespace ErrorCodes
+{
+    extern const int LOGICAL_ERROR;
+}
+
 /// Counts how many bytes of `filt` are greater than zero.
 size_t countBytesInFilter(const IColumn::Filter & filt);
 
@@ -33,4 +38,38 @@ void filterArraysImplOnlyData(
     PaddedPODArray<T> & res_elems,
     const IColumn::Filter & filt, ssize_t result_size_hint);
 
+namespace detail
+{
+    template <typename T>
+    const PaddedPODArray<T> * getIndexesData(const IColumn & indexes);
+}
+
+/// Check limit <= indexes->size() and call column.indexImpl(const PaddedPodArray<Type> & indexes, size_t limit).
+template <typename Column>
+ColumnPtr selectIndexImpl(const Column & column, const IColumn & indexes, size_t limit)
+{
+    if (limit == 0)
+        limit = indexes.size();
+
+    if (indexes.size() < limit)
+        throw Exception("Size of indexes is less than required.", ErrorCodes::SIZES_OF_COLUMNS_DOESNT_MATCH);
+
+    if (auto * data_uint8 = detail::getIndexesData<UInt8>(indexes))
+        return column.template indexImpl<UInt8>(*data_uint8, limit);
+    else if (auto * data_uint16 = detail::getIndexesData<UInt16>(indexes))
+        return column.template indexImpl<UInt16>(*data_uint16, limit);
+    else if (auto * data_uint32 = detail::getIndexesData<UInt32>(indexes))
+        return column.template indexImpl<UInt32>(*data_uint32, limit);
+    else if (auto * data_uint64 = detail::getIndexesData<UInt64>(indexes))
+        return column.template indexImpl<UInt64>(*data_uint64, limit);
+    else
+        throw Exception("Indexes column for IColumn::select must be ColumnUInt, got" + indexes.getName(),
+                        ErrorCodes::LOGICAL_ERROR);
+}
+
+#define INSTANTIATE_INDEX_IMPL(Column) \
+    template ColumnPtr Column::indexImpl<UInt8>(const PaddedPODArray<UInt8> & indexes, size_t limit) const; \
+    template ColumnPtr Column::indexImpl<UInt16>(const PaddedPODArray<UInt16> & indexes, size_t limit) const; \
+    template ColumnPtr Column::indexImpl<UInt32>(const PaddedPODArray<UInt32> & indexes, size_t limit) const; \
+    template ColumnPtr Column::indexImpl<UInt64>(const PaddedPODArray<UInt64> & indexes, size_t limit) const;
 }

dbms/src/Columns/FilterDescription.cpp

@@ -27,14 +27,14 @@ ConstantFilterDescription::ConstantFilterDescription(const IColumn & column)
     if (column.isColumnConst())
     {
         const ColumnConst & column_const = static_cast<const ColumnConst &>(column);
-        const IColumn & column_nested = column_const.getDataColumn();
+        ColumnPtr column_nested = column_const.getDataColumnPtr()->convertToFullColumnIfWithDictionary();
 
-        if (!typeid_cast<const ColumnUInt8 *>(&column_nested))
+        if (!typeid_cast<const ColumnUInt8 *>(column_nested.get()))
         {
-            const ColumnNullable * column_nested_nullable = typeid_cast<const ColumnNullable *>(&column_nested);
+            const ColumnNullable * column_nested_nullable = typeid_cast<const ColumnNullable *>(column_nested.get());
             if (!column_nested_nullable || !typeid_cast<const ColumnUInt8 *>(&column_nested_nullable->getNestedColumn()))
             {
-                throw Exception("Illegal type " + column_nested.getName() + " of column for constant filter. Must be UInt8 or Nullable(UInt8).",
+                throw Exception("Illegal type " + column_nested->getName() + " of column for constant filter. Must be UInt8 or Nullable(UInt8).",
                                 ErrorCodes::ILLEGAL_TYPE_OF_COLUMN_FOR_FILTER);
             }
         }
@@ -48,8 +48,13 @@ ConstantFilterDescription::ConstantFilterDescription(const IColumn & column)
 }
 
 
-FilterDescription::FilterDescription(const IColumn & column)
+FilterDescription::FilterDescription(const IColumn & column_)
 {
+    if (column_.withDictionary())
+        data_holder = column_.convertToFullColumnIfWithDictionary();
+
+    const auto & column = data_holder ? *data_holder : column_;
+
     if (const ColumnUInt8 * concrete_column = typeid_cast<const ColumnUInt8 *>(&column))
     {
         data = &concrete_column->getData();

dbms/src/Columns/IColumn.h

@@ -47,6 +47,10 @@ public:
       */
     virtual Ptr convertToFullColumnIfConst() const { return {}; }
 
+    /// If column isn't ColumnWithDictionary, return itself.
+    /// If column is ColumnWithDictionary, transforms is to full column.
+    virtual Ptr convertToFullColumnIfWithDictionary() const { return getPtr(); }
+
     /// Creates empty column with the same type.
     virtual MutablePtr cloneEmpty() const { return cloneResized(0); }
 
@@ -188,6 +192,10 @@ public:
     using Permutation = PaddedPODArray<size_t>;
     virtual Ptr permute(const Permutation & perm, size_t limit) const = 0;
 
+    /// Creates new column with values column[indexes[:limit]]. If limit is 0, all indexes are used.
+    /// Indexes must be one of the ColumnUInt. For default implementation, see selectIndexImpl from ColumnsCommon.h
+    virtual Ptr index(const IColumn & indexes, size_t limit) const = 0;
+
     /** Compares (*this)[n] and rhs[m].
       * Returns negative number, 0, or positive number (*this)[n] is less, equal, greater than rhs[m] respectively.
       * Is used in sortings.
@@ -325,6 +333,8 @@ public:
     /// Can be inside ColumnNullable.
     virtual bool canBeInsideNullable() const { return false; }
 
+    virtual bool withDictionary() const { return false; }
+
 
     virtual ~IColumn() {}
 

dbms/src/Columns/IColumnDummy.h

@@ -87,6 +87,14 @@ public:
         return cloneDummy(limit ? std::min(s, limit) : s);
     }
 
+    ColumnPtr index(const IColumn & indexes, size_t limit) const override
+    {
+        if (indexes.size() < limit)
+            throw Exception("Size of indexes is less than required.", ErrorCodes::SIZES_OF_COLUMNS_DOESNT_MATCH);
+
+        return cloneDummy(limit ? limit : s);
+    }
+
     void getPermutation(bool /*reverse*/, size_t /*limit*/, int /*nan_direction_hint*/, Permutation & res) const override
     {
         res.resize(s);

dbms/src/Columns/IColumnUnique.h

@@ -0,0 +1,129 @@
+#pragma once
+#include <Columns/IColumn.h>
+
+namespace DB
+{
+
+class IColumnUnique : public IColumn
+{
+public:
+    using ColumnUniquePtr = IColumn::template immutable_ptr<IColumnUnique>;
+    using MutableColumnUniquePtr = IColumn::template mutable_ptr<IColumnUnique>;
+
+    /// Column always contains Null if it's Nullable and empty string if it's String or Nullable(String).
+    /// So, size may be greater than the number of inserted unique values.
+    virtual const ColumnPtr & getNestedColumn() const = 0;
+    /// The same as getNestedColumn, but removes null map if nested column is nullable.
+    virtual const ColumnPtr & getNestedNotNullableColumn() const = 0;
+
+    size_t size() const override { return getNestedColumn()->size(); }
+
+    /// Appends new value at the end of column (column's size is increased by 1).
+    /// Is used to transform raw strings to Blocks (for example, inside input format parsers)
+    virtual size_t uniqueInsert(const Field & x) = 0;
+
+    virtual size_t uniqueInsertFrom(const IColumn & src, size_t n) = 0;
+    /// Appends range of elements from other column.
+    /// Could be used to concatenate columns.
+    virtual MutableColumnPtr uniqueInsertRangeFrom(const IColumn & src, size_t start, size_t length) = 0;
+
+    struct IndexesWithOverflow
+    {
+        MutableColumnPtr indexes;
+        MutableColumnPtr overflowed_keys;
+    };
+    /// Like uniqueInsertRangeFrom, but doesn't insert keys if inner dictionary has more than max_dictionary_size keys.
+    /// Keys that won't be inserted into dictionary will be into overflowed_keys, indexes will be calculated for
+    /// concatenation of nested column (which can be got from getNestedColumn() function) and overflowed_keys.
+    virtual IndexesWithOverflow uniqueInsertRangeWithOverflow(const IColumn & src, size_t start,
+                                                              size_t length, size_t max_dictionary_size) = 0;
+
+    /// Appends data located in specified memory chunk if it is possible (throws an exception if it cannot be implemented).
+    /// Is used to optimize some computations (in aggregation, for example).
+    /// Parameter length could be ignored if column values have fixed size.
+    virtual size_t uniqueInsertData(const char * pos, size_t length) = 0;
+    virtual size_t uniqueInsertDataWithTerminatingZero(const char * pos, size_t length) = 0;
+
+    virtual size_t getDefaultValueIndex() const = 0;
+    virtual size_t getNullValueIndex() const = 0;
+    virtual bool canContainNulls() const = 0;
+
+    virtual size_t uniqueDeserializeAndInsertFromArena(const char * pos, const char *& new_pos) = 0;
+
+    const char * getFamilyName() const override { return "ColumnUnique"; }
+
+    void insert(const Field &) override
+    {
+        throw Exception("Method insert is not supported for ColumnUnique.", ErrorCodes::NOT_IMPLEMENTED);
+    }
+
+    void insertRangeFrom(const IColumn &, size_t, size_t) override
+    {
+        throw Exception("Method insertRangeFrom is not supported for ColumnUnique.", ErrorCodes::NOT_IMPLEMENTED);
+    }
+
+    void insertData(const char *, size_t) override
+    {
+        throw Exception("Method insertData is not supported for ColumnUnique.", ErrorCodes::NOT_IMPLEMENTED);
+    }
+
+    void insertDefault() override
+    {
+        throw Exception("Method insertDefault is not supported for ColumnUnique.", ErrorCodes::NOT_IMPLEMENTED);
+    }
+
+    void popBack(size_t) override
+    {
+        throw Exception("Method popBack is not supported for ColumnUnique.", ErrorCodes::NOT_IMPLEMENTED);
+    }
+
+    void gather(ColumnGathererStream &) override
+    {
+        throw Exception("Method gather is not supported for ColumnUnique.", ErrorCodes::NOT_IMPLEMENTED);
+    }
+
+    const char * deserializeAndInsertFromArena(const char *) override
+    {
+        throw Exception("Method deserializeAndInsertFromArena is not supported for ColumnUnique.", ErrorCodes::NOT_IMPLEMENTED);
+    }
+
+    ColumnPtr index(const IColumn &, size_t) const override
+    {
+        throw Exception("Method index is not supported for ColumnUnique.", ErrorCodes::NOT_IMPLEMENTED);
+    }
+
+    ColumnPtr cut(size_t, size_t) const override
+    {
+        throw Exception("Method cut is not supported for ColumnUnique.", ErrorCodes::NOT_IMPLEMENTED);
+    }
+
+    ColumnPtr filter(const IColumn::Filter &, ssize_t) const override
+    {
+        throw Exception("Method filter is not supported for ColumnUnique.", ErrorCodes::NOT_IMPLEMENTED);
+    }
+
+    ColumnPtr permute(const IColumn::Permutation &, size_t) const override
+    {
+        throw Exception("Method permute is not supported for ColumnUnique.", ErrorCodes::NOT_IMPLEMENTED);
+    }
+
+    ColumnPtr replicate(const IColumn::Offsets &) const override
+    {
+        throw Exception("Method replicate is not supported for ColumnUnique.", ErrorCodes::NOT_IMPLEMENTED);
+    }
+
+    void getPermutation(bool, size_t, int, IColumn::Permutation &) const override
+    {
+        throw Exception("Method getPermutation is not supported for ColumnUnique.", ErrorCodes::NOT_IMPLEMENTED);
+    }
+
+    std::vector<MutableColumnPtr> scatter(IColumn::ColumnIndex, const IColumn::Selector &) const override
+    {
+        throw Exception("Method scatter is not supported for ColumnUnique.", ErrorCodes::NOT_IMPLEMENTED);
+    }
+};
+
+using ColumnUniquePtr = IColumnUnique::ColumnUniquePtr;
+using MutableColumnUniquePtr = IColumnUnique::MutableColumnUniquePtr;
+
+}

dbms/src/Columns/ReverseIndex.h

@@ -0,0 +1,412 @@
+#pragma once
+
+#include <Common/HashTable/Hash.h>
+#include <Common/HashTable/HashTable.h>
+#include <Common/HashTable/HashTableAllocator.h>
+
+#include <Columns/ColumnString.h>
+#include <ext/range.h>
+
+namespace DB
+{
+
+namespace
+{
+    template <typename ColumnType, bool with_saved_hash, bool has_base_index>
+    struct ReverseIndexHashTableState;
+
+    template <typename ColumnType>
+    struct ReverseIndexHashTableState<ColumnType, /* with_saved_hash */ false, /* has_base_index */ false>
+    {
+        constexpr static bool with_saved_hash = false;
+        constexpr static bool has_base_index = false;
+
+        ColumnType * index_column;
+    };
+
+    template <typename ColumnType>
+    struct ReverseIndexHashTableState<ColumnType, /* with_saved_hash */ false, /* has_base_index */ true>
+    {
+        constexpr static bool with_saved_hash = false;
+        constexpr static bool has_base_index = true;
+
+        ColumnType * index_column;
+        size_t base_index;
+    };
+
+    template <typename ColumnType>
+    struct ReverseIndexHashTableState<ColumnType, /* with_saved_hash = */ true, /* has_base_index */ false>
+    {
+        constexpr static bool with_saved_hash = true;
+        constexpr static bool has_base_index = false;
+
+        ColumnType * index_column;
+        typename ColumnVector<UInt64>::Container * saved_hash_column;
+    };
+
+    template <typename ColumnType>
+    struct ReverseIndexHashTableState<ColumnType, /* with_saved_hash = */ true, /* has_base_index */ true>
+    {
+        constexpr static bool with_saved_hash = true;
+        constexpr static bool has_base_index = true;
+
+        ColumnType * index_column;
+        typename ColumnVector<UInt64>::Container * saved_hash_column;
+        size_t base_index;
+    };
+
+
+    template <typename Hash>
+    struct ReverseIndexHash : public Hash
+    {
+        template <typename T>
+        size_t operator()(T) const
+        {
+            throw Exception("operator()(key) is not implemented for ReverseIndexHash.", ErrorCodes::LOGICAL_ERROR);
+        }
+
+        template <typename State, typename T>
+        size_t operator()(const State & state, T key) const
+        {
+            auto index = key;
+            if constexpr (State::has_base_index)
+                index -= state.base_index;
+
+            return Hash::operator()(state.index_column->getElement(index));
+        }
+    };
+
+    using ReverseIndexStringHash = ReverseIndexHash<StringRefHash>;
+
+    template <typename IndexType>
+    using ReverseIndexNumberHash = ReverseIndexHash<DefaultHash<IndexType>>;
+
+
+    template <typename IndexType, typename Hash, typename HashTable, typename ColumnType, bool string_hash, bool has_base_index>
+    struct  ReverseIndexHashTableCell
+        : public HashTableCell<IndexType, Hash, ReverseIndexHashTableState<ColumnType, string_hash, has_base_index>>
+    {
+        using Base = HashTableCell<IndexType, Hash, ReverseIndexHashTableState<ColumnType, string_hash, has_base_index>>;
+        using State = typename Base::State;
+        using Base::Base;
+        using Base::key;
+        using Base::keyEquals;
+        using Base::isZero;
+
+        template <typename T>
+        static bool isZero(const T &, const State & /*state*/)
+        {
+            static_assert(!std::is_same_v<typename std::decay<T>::type, typename std::decay<IndexType>::type>);
+            return false;
+        }
+        /// Special case when we want to compare with something not in index_column.
+        /// When we compare something inside column default keyEquals checks only that row numbers are equal.
+        bool keyEquals(const StringRef & object, size_t hash_ [[maybe_unused]], const State & state) const
+        {
+            auto index = key;
+            if constexpr (has_base_index)
+                index -= state.base_index;
+
+            if constexpr (string_hash)
+                return hash_ == (*state.saved_hash_column)[index] && object == state.index_column->getDataAt(index);
+            else
+                return object == state.index_column->getDataAt(index);
+        }
+
+        size_t getHash(const Hash & hash) const
+        {
+            auto index = key;
+
+            /// Hack. HashTable is Hash itself.
+            const auto & state = static_cast<const State &>(static_cast<const HashTable &>(hash));
+
+            if constexpr (has_base_index)
+                index -= state.base_index;
+
+            if constexpr (string_hash)
+                return (*state.saved_hash_column)[index];
+            else
+                return hash(state, key);
+        }
+    };
+
+
+    template <typename Key, typename Cell, typename Hash>
+    class HashTableWithPublicState : public HashTable<Key, Cell, Hash, HashTableGrower<>, HashTableAllocator>
+    {
+        using State = typename Cell::State;
+        using Base = HashTable<Key, Cell, Hash, HashTableGrower<>, HashTableAllocator>;
+
+    public:
+        using Base::Base;
+        State & getState() { return *this; }
+    };
+
+    template <typename IndexType, typename ColumnType, bool has_base_index>
+    class ReverseIndexStringHashTable : public HashTableWithPublicState<
+            IndexType,
+            ReverseIndexHashTableCell<
+                    IndexType,
+                    ReverseIndexStringHash,
+                    ReverseIndexStringHashTable<IndexType, ColumnType, has_base_index>,
+                    ColumnType,
+                    true,
+                    has_base_index>,
+            ReverseIndexStringHash>
+    {
+        using Base = HashTableWithPublicState<
+                IndexType,
+                ReverseIndexHashTableCell<
+                        IndexType,
+                        ReverseIndexStringHash,
+                        ReverseIndexStringHashTable<IndexType, ColumnType, has_base_index>,
+                        ColumnType,
+                        true,
+                        has_base_index>,
+                ReverseIndexStringHash>;
+    public:
+        using Base::Base;
+        friend struct ReverseIndexHashTableCell<
+                IndexType,
+                ReverseIndexStringHash,
+                ReverseIndexStringHashTable<IndexType, ColumnType, has_base_index>,
+                ColumnType,
+                true,
+                has_base_index>;
+    };
+
+    template <typename IndexType, typename ColumnType, bool has_base_index>
+    class ReverseIndexNumberHashTable : public HashTableWithPublicState<
+            IndexType,
+            ReverseIndexHashTableCell<
+                    IndexType,
+                    ReverseIndexNumberHash<typename ColumnType::value_type>,
+                    ReverseIndexNumberHashTable<IndexType, ColumnType, has_base_index>,
+                    ColumnType,
+                    false,
+                    has_base_index>,
+            ReverseIndexNumberHash<typename ColumnType::value_type>>
+    {
+        using Base = HashTableWithPublicState<
+                IndexType,
+                ReverseIndexHashTableCell<
+                        IndexType,
+                        ReverseIndexNumberHash<typename ColumnType::value_type>,
+                        ReverseIndexNumberHashTable<IndexType, ColumnType, has_base_index>,
+                        ColumnType,
+                        false,
+                        has_base_index>,
+                ReverseIndexNumberHash<typename ColumnType::value_type>>;
+    public:
+        using Base::Base;
+        friend struct ReverseIndexHashTableCell<
+                IndexType,
+                ReverseIndexNumberHash<typename ColumnType::value_type>,
+                ReverseIndexNumberHashTable<IndexType, ColumnType, has_base_index>,
+                ColumnType,
+                false,
+                has_base_index>;
+    };
+
+
+    template <typename IndexType, typename ColumnType, bool has_base_index, bool is_numeric_column>
+    struct SelectReverseIndexHashTable;
+
+    template <typename IndexType, typename ColumnType, bool has_base_index>
+    struct SelectReverseIndexHashTable<IndexType, ColumnType, has_base_index, true>
+    {
+        using Type = ReverseIndexNumberHashTable<IndexType, ColumnType, has_base_index>;
+    };
+
+    template <typename IndexType, typename ColumnType, bool has_base_index>
+    struct SelectReverseIndexHashTable<IndexType, ColumnType, has_base_index, false>
+    {
+        using Type = ReverseIndexStringHashTable<IndexType, ColumnType, has_base_index>;
+    };
+
+
+    template <typename T>
+    constexpr bool isNumericColumn(const T *) { return false; }
+
+    template <typename T>
+    constexpr bool isNumericColumn(const ColumnVector<T> *) { return true; }
+
+    static_assert(isNumericColumn(static_cast<ColumnVector<UInt8> *>(nullptr)));
+    static_assert(!isNumericColumn(static_cast<ColumnString *>(nullptr)));
+
+
+    template <typename IndexType, typename ColumnType, bool has_base_index>
+    using ReverseIndexHashTable = typename SelectReverseIndexHashTable<IndexType, ColumnType, has_base_index,
+            isNumericColumn(static_cast<ColumnType *>(nullptr))>::Type;
+}
+
+
+template <typename IndexType, typename ColumnType>
+class ReverseIndex
+{
+public:
+    explicit ReverseIndex(UInt64 num_prefix_rows_to_skip, UInt64 base_index)
+            : num_prefix_rows_to_skip(num_prefix_rows_to_skip), base_index(base_index) {}
+
+    void setColumn(ColumnType * column_);
+
+    static constexpr bool is_numeric_column = isNumericColumn(static_cast<ColumnType *>(nullptr));
+    static constexpr bool use_saved_hash = !is_numeric_column;
+
+    UInt64 insert(UInt64 from_position);  /// Insert into index column[from_position];
+    UInt64 insertFromLastRow();
+    UInt64 getInsertionPoint(const StringRef & data);
+    UInt64 lastInsertionPoint() const { return size() + base_index; }
+
+    ColumnType * getColumn() const { return column; }
+    size_t size() const;
+
+    size_t allocatedBytes() const { return index ? index->getBufferSizeInBytes() : 0; }
+
+private:
+    ColumnType * column = nullptr;
+    UInt64 num_prefix_rows_to_skip; /// The number prefix tows in column which won't be sored at index.
+    UInt64 base_index; /// This values will be added to row number which is inserted into index.
+
+    using IndexMapType = ReverseIndexHashTable<IndexType, ColumnType, true>;
+
+    /// Lazy initialized.
+    std::unique_ptr<IndexMapType> index;
+    ColumnUInt64::MutablePtr saved_hash;
+
+    void buildIndex();
+
+    UInt64 getHash(const StringRef & ref) const
+    {
+        if constexpr (is_numeric_column)
+        {
+            using ValueType = typename ColumnType::value_type;
+            ValueType value = *reinterpret_cast<const ValueType *>(ref.data);
+            return DefaultHash<ValueType>()(value);
+        }
+        else
+            return StringRefHash()(ref);
+    }
+};
+
+
+
+template <typename IndexType, typename ColumnType>
+void ReverseIndex<IndexType, ColumnType>:: setColumn(ColumnType * column_)
+{
+    if (column != column_)
+        index = nullptr;
+
+    column = column_;
+}
+
+template <typename IndexType, typename ColumnType>
+size_t ReverseIndex<IndexType, ColumnType>::size() const
+{
+    if (!column)
+        throw Exception("ReverseIndex has not size because index column wasn't set.", ErrorCodes::LOGICAL_ERROR);
+
+    return column->size();
+}
+
+template <typename IndexType, typename ColumnType>
+void ReverseIndex<IndexType, ColumnType>::buildIndex()
+{
+    if (index)
+        return;
+
+    if (!column)
+        throw Exception("ReverseIndex can't build index because index column wasn't set.", ErrorCodes::LOGICAL_ERROR);
+
+    auto size = column->size();
+    index = std::make_unique<IndexMapType>(size);
+
+    if constexpr (use_saved_hash)
+        saved_hash = ColumnUInt64::create(size);
+
+    auto & state = index->getState();
+    state.index_column = column;
+    state.base_index = base_index;
+    if constexpr (use_saved_hash)
+        state.saved_hash_column = &saved_hash->getData();
+
+    using IteratorType = typename IndexMapType::iterator;
+    IteratorType iterator;
+    bool inserted;
+
+    for (auto row : ext::range(num_prefix_rows_to_skip, size))
+    {
+        auto hash = getHash(column->getDataAt(row));
+
+        if constexpr (use_saved_hash)
+            saved_hash->getElement(row) = hash;
+
+        index->emplace(row + base_index, iterator, inserted, hash);
+
+        if (!inserted)
+            throw Exception("Duplicating keys found in ReverseIndex.", ErrorCodes::LOGICAL_ERROR);
+    }
+}
+
+template <typename IndexType, typename ColumnType>
+UInt64 ReverseIndex<IndexType, ColumnType>::insert(UInt64 from_position)
+{
+    if (!index)
+        buildIndex();
+
+    using IteratorType = typename IndexMapType::iterator;
+    IteratorType iterator;
+    bool inserted;
+
+    auto hash = getHash(column->getDataAt(from_position));
+
+    if constexpr (use_saved_hash)
+    {
+        auto & data = saved_hash->getData();
+        if (data.size() <= from_position)
+            data.resize(from_position + 1);
+        data[from_position] = hash;
+    }
+
+    index->emplace(from_position + base_index, iterator, inserted, hash);
+
+    return *iterator;
+}
+
+template <typename IndexType, typename ColumnType>
+UInt64 ReverseIndex<IndexType, ColumnType>::insertFromLastRow()
+{
+    if (!column)
+        throw Exception("ReverseIndex can't insert row from column because index column wasn't set.",
+                        ErrorCodes::LOGICAL_ERROR);
+
+    UInt64 num_rows = size();
+
+    if (num_rows == 0)
+        throw Exception("ReverseIndex can't insert row from column because it is empty.", ErrorCodes::LOGICAL_ERROR);
+
+    UInt64 position = num_rows - 1;
+    UInt64 inserted_pos = insert(position);
+    if (position + base_index != inserted_pos)
+        throw Exception("Can't insert into reverse index from last row (" + toString(position + base_index)
+                        + ") because the same row is in position " + toString(inserted_pos), ErrorCodes::LOGICAL_ERROR);
+
+    return inserted_pos;
+}
+
+template <typename IndexType, typename ColumnType>
+UInt64 ReverseIndex<IndexType, ColumnType>::getInsertionPoint(const StringRef & data)
+{
+    if (!index)
+        buildIndex();
+
+    using IteratorType = typename IndexMapType::iterator;
+    IteratorType iterator;
+
+    auto hash = getHash(data);
+    iterator = index->find(data, hash);
+
+    return iterator == index->end() ? size() + base_index : *iterator;
+}
+
+}

dbms/src/Common/ErrorCodes.cpp

@@ -380,9 +380,10 @@ namespace ErrorCodes
     extern const int INVALID_JOIN_ON_EXPRESSION = 403;
     extern const int BAD_ODBC_CONNECTION_STRING = 404;
     extern const int PARTITION_SIZE_EXCEEDS_MAX_DROP_SIZE_LIMIT = 405;
-    extern const int BAD_REQUEST_PARAMETER = 406;
-    extern const int EXTERNAL_EXECUTABLE_NOT_FOUND = 407;
-    extern const int EXTERNAL_SERVER_IS_NOT_RESPONDING = 408;
+    extern const int TOP_AND_LIMIT_TOGETHER = 406;
+    extern const int BAD_REQUEST_PARAMETER = 407;
+    extern const int EXTERNAL_EXECUTABLE_NOT_FOUND = 408;
+    extern const int EXTERNAL_SERVER_IS_NOT_RESPONDING = 409;
 
     extern const int KEEPER_EXCEPTION = 999;
     extern const int POCO_EXCEPTION = 1000;

dbms/src/Common/HashTable/HashMap.h

@@ -53,6 +53,7 @@ struct HashMapCell
 
     bool keyEquals(const Key & key_) const { return value.first == key_; }
     bool keyEquals(const Key & key_, size_t /*hash_*/) const { return value.first == key_; }
+    bool keyEquals(const Key & key_, size_t /*hash_*/, const State & /*state*/) const { return value.first == key_; }
 
     void setHash(size_t /*hash_value*/) {}
     size_t getHash(const Hash & hash) const { return hash(value.first); }
@@ -112,6 +113,7 @@ struct HashMapCellWithSavedHash : public HashMapCell<Key, TMapped, Hash, TState>
 
     bool keyEquals(const Key & key_) const { return this->value.first == key_; }
     bool keyEquals(const Key & key_, size_t hash_) const { return saved_hash == hash_ && this->value.first == key_; }
+    bool keyEquals(const Key & key_, size_t hash_, const typename Base::State &) const { return keyEquals(key_, hash_); }
 
     void setHash(size_t hash_value) { saved_hash = hash_value; }
     size_t getHash(const Hash & /*hash_function*/) const { return saved_hash; }

dbms/src/Common/HashTable/HashSet.h

@@ -75,6 +75,7 @@ struct HashSetCellWithSavedHash : public HashTableCell<Key, Hash, TState>
 
     bool keyEquals(const Key & key_) const { return this->key == key_; }
     bool keyEquals(const Key & key_, size_t hash_) const { return saved_hash == hash_ && this->key == key_; }
+    bool keyEquals(const Key & key_, size_t hash_, const typename Base::State &) const { return keyEquals(key_, hash_); }
 
     void setHash(size_t hash_value) { saved_hash = hash_value; }
     size_t getHash(const Hash & /*hash_function*/) const { return saved_hash; }

dbms/src/Common/HashTable/HashTable.h

@@ -108,6 +108,7 @@ struct HashTableCell
     /// Are the keys at the cells equal?
     bool keyEquals(const Key & key_) const { return key == key_; }
     bool keyEquals(const Key & key_, size_t /*hash_*/) const { return key == key_; }
+    bool keyEquals(const Key & key_, size_t /*hash_*/, const State & /*state*/) const { return key == key_; }
 
     /// If the cell can remember the value of the hash function, then remember it.
     void setHash(size_t /*hash_value*/) {}
@@ -280,9 +281,10 @@ protected:
 #endif
 
     /// Find a cell with the same key or an empty cell, starting from the specified position and further along the collision resolution chain.
-    size_t ALWAYS_INLINE findCell(const Key & x, size_t hash_value, size_t place_value) const
+    template <typename ObjectToCompareWith>
+    size_t ALWAYS_INLINE findCell(const ObjectToCompareWith & x, size_t hash_value, size_t place_value) const
     {
-        while (!buf[place_value].isZero(*this) && !buf[place_value].keyEquals(x, hash_value))
+        while (!buf[place_value].isZero(*this) && !buf[place_value].keyEquals(x, hash_value, *this))
         {
             place_value = grower.next(place_value);
 #ifdef DBMS_HASH_MAP_COUNT_COLLISIONS
@@ -734,7 +736,8 @@ public:
     }
 
 
-    iterator ALWAYS_INLINE find(Key x)
+    template <typename ObjectToCompareWith>
+    iterator ALWAYS_INLINE find(ObjectToCompareWith x)
     {
         if (Cell::isZero(x, *this))
             return this->hasZero() ? iteratorToZero() : end();
@@ -745,7 +748,8 @@ public:
     }
 
 
-    const_iterator ALWAYS_INLINE find(Key x) const
+    template <typename ObjectToCompareWith>
+    const_iterator ALWAYS_INLINE find(ObjectToCompareWith x) const
     {
         if (Cell::isZero(x, *this))
             return this->hasZero() ? iteratorToZero() : end();
@@ -756,7 +760,8 @@ public:
     }
 
 
-    iterator ALWAYS_INLINE find(Key x, size_t hash_value)
+    template <typename ObjectToCompareWith>
+    iterator ALWAYS_INLINE find(ObjectToCompareWith x, size_t hash_value)
     {
         if (Cell::isZero(x, *this))
             return this->hasZero() ? iteratorToZero() : end();
@@ -766,7 +771,8 @@ public:
     }
 
 
-    const_iterator ALWAYS_INLINE find(Key x, size_t hash_value) const
+    template <typename ObjectToCompareWith>
+    const_iterator ALWAYS_INLINE find(ObjectToCompareWith x, size_t hash_value) const
     {
         if (Cell::isZero(x, *this))
             return this->hasZero() ? iteratorToZero() : end();

dbms/src/Core/Block.cpp

@@ -368,6 +368,18 @@ Block Block::cloneWithColumns(const Columns & columns) const
 }
 
 
+Block Block::cloneWithoutColumns() const
+{
+    Block res;
+
+    size_t num_columns = data.size();
+    for (size_t i = 0; i < num_columns; ++i)
+        res.insert({ nullptr, data[i].type, data[i].name });
+
+    return res;
+}
+
+
 Block Block::sortColumns() const
 {
     Block sorted_block;

dbms/src/Core/Block.h

@@ -109,6 +109,7 @@ public:
     Columns getColumns() const;
     void setColumns(const Columns & columns);
     Block cloneWithColumns(const Columns & columns) const;
+    Block cloneWithoutColumns() const;
 
     /** Get empty columns with the same types as in block. */
     MutableColumns cloneEmptyColumns() const;

dbms/src/DataStreams/ConvertColumnWithDictionaryToFullBlockInputStream.h

@@ -0,0 +1,57 @@
+#pragma once
+
+#include <DataStreams/IProfilingBlockInputStream.h>
+#include <Columns/ColumnWithDictionary.h>
+#include <DataTypes/DataTypeWithDictionary.h>
+
+namespace DB
+{
+
+
+/** Combines several sources into one.
+  * Unlike UnionBlockInputStream, it does this sequentially.
+  * Blocks of different sources are not interleaved with each other.
+  */
+class ConvertColumnWithDictionaryToFullBlockInputStream : public IProfilingBlockInputStream
+{
+public:
+    explicit ConvertColumnWithDictionaryToFullBlockInputStream(const BlockInputStreamPtr & input)
+    {
+        children.push_back(input);
+    }
+
+    String getName() const override { return "ConvertColumnWithDictionaryToFull"; }
+
+    Block getHeader() const override { return convert(children.at(0)->getHeader()); }
+
+protected:
+    Block readImpl() override { return convert(children.back()->read()); }
+
+private:
+    Block convert(Block && block) const
+    {
+        for (auto & column : block)
+        {
+            auto * type_with_dict = typeid_cast<const DataTypeWithDictionary *>(column.type.get());
+            auto * col_with_dict = typeid_cast<const ColumnWithDictionary *>(column.column.get());
+
+            if (type_with_dict && !col_with_dict)
+                throw Exception("Invalid column for " + type_with_dict->getName() + ": " + column.column->getName(),
+                                ErrorCodes::LOGICAL_ERROR);
+
+            if (!type_with_dict && col_with_dict)
+                throw Exception("Invalid type for " + col_with_dict->getName() + ": " + column.type->getName(),
+                                ErrorCodes::LOGICAL_ERROR);
+
+            if (type_with_dict && col_with_dict)
+            {
+                column.column = col_with_dict->convertToFullColumn();
+                column.type = type_with_dict->getDictionaryType();
+            }
+        }
+
+        return std::move(block);
+    }
+};
+
+}

dbms/src/DataStreams/NativeBlockInputStream.cpp

@@ -59,8 +59,14 @@ NativeBlockInputStream::NativeBlockInputStream(ReadBuffer & istr_, UInt64 server
 
 void NativeBlockInputStream::readData(const IDataType & type, IColumn & column, ReadBuffer & istr, size_t rows, double avg_value_size_hint)
 {
-    IDataType::InputStreamGetter input_stream_getter = [&] (const IDataType::SubstreamPath &) { return &istr; };
-    type.deserializeBinaryBulkWithMultipleStreams(column, input_stream_getter, rows, avg_value_size_hint, false, {});
+    IDataType::DeserializeBinaryBulkSettings settings;
+    settings.getter = [&](IDataType::SubstreamPath) -> ReadBuffer * { return &istr; };
+    settings.avg_value_size_hint = avg_value_size_hint;
+    settings.position_independent_encoding = false;
+
+    IDataType::DeserializeBinaryBulkStatePtr state;
+    type.deserializeBinaryBulkStatePrefix(settings, state);
+    type.deserializeBinaryBulkWithMultipleStreams(column, rows, settings, state);
 
     if (column.size() != rows)
         throw Exception("Cannot read all data in NativeBlockInputStream. Rows read: " + toString(column.size()) + ". Rows expected: " + toString(rows) + ".",

dbms/src/DataStreams/NativeBlockOutputStream.cpp

@@ -52,8 +52,15 @@ void NativeBlockOutputStream::writeData(const IDataType & type, const ColumnPtr
     else
         full_column = column;
 
-    IDataType::OutputStreamGetter output_stream_getter = [&] (const IDataType::SubstreamPath &) { return &ostr; };
-    type.serializeBinaryBulkWithMultipleStreams(*full_column, output_stream_getter, offset, limit, false, {});
+    IDataType::SerializeBinaryBulkSettings settings;
+    settings.getter = [&ostr](IDataType::SubstreamPath) -> WriteBuffer * { return &ostr; };
+    settings.position_independent_encoding = false;
+    settings.low_cardinality_max_dictionary_size = 0;
+
+    IDataType::SerializeBinaryBulkStatePtr state;
+    type.serializeBinaryBulkStatePrefix(settings, state);
+    type.serializeBinaryBulkWithMultipleStreams(*full_column, offset, limit, settings, state);
+    type.serializeBinaryBulkStateSuffix(settings, state);
 }
 
 

dbms/src/DataTypes/DataTypeArray.cpp

@@ -145,37 +145,67 @@ namespace
 }
 
 
-void DataTypeArray::enumerateStreams(StreamCallback callback, SubstreamPath path) const
+void DataTypeArray::enumerateStreams(const StreamCallback & callback, SubstreamPath & path) const
 {
     path.push_back(Substream::ArraySizes);
     callback(path);
     path.back() = Substream::ArrayElements;
     nested->enumerateStreams(callback, path);
+    path.pop_back();
+}
+
+
+void DataTypeArray::serializeBinaryBulkStatePrefix(
+    SerializeBinaryBulkSettings & settings,
+    SerializeBinaryBulkStatePtr & state) const
+{
+    settings.path.push_back(Substream::ArrayElements);
+    nested->serializeBinaryBulkStatePrefix(settings, state);
+    settings.path.pop_back();
+}
+
+
+void DataTypeArray::serializeBinaryBulkStateSuffix(
+    SerializeBinaryBulkSettings & settings,
+    SerializeBinaryBulkStatePtr & state) const
+{
+    settings.path.push_back(Substream::ArrayElements);
+    nested->serializeBinaryBulkStateSuffix(settings, state);
+    settings.path.pop_back();
+}
+
+
+void DataTypeArray::deserializeBinaryBulkStatePrefix(
+    DeserializeBinaryBulkSettings & settings,
+    DeserializeBinaryBulkStatePtr & state) const
+{
+    settings.path.push_back(Substream::ArrayElements);
+    nested->deserializeBinaryBulkStatePrefix(settings, state);
+    settings.path.pop_back();
 }
 
 
 void DataTypeArray::serializeBinaryBulkWithMultipleStreams(
     const IColumn & column,
-    OutputStreamGetter getter,
     size_t offset,
     size_t limit,
-    bool position_independent_encoding,
-    SubstreamPath path) const
+    SerializeBinaryBulkSettings & settings,
+    SerializeBinaryBulkStatePtr & state) const
 {
     const ColumnArray & column_array = typeid_cast<const ColumnArray &>(column);
 
     /// First serialize array sizes.
-    path.push_back(Substream::ArraySizes);
-    if (auto stream = getter(path))
+    settings.path.push_back(Substream::ArraySizes);
+    if (auto stream = settings.getter(settings.path))
     {
-        if (position_independent_encoding)
+        if (settings.position_independent_encoding)
             serializeArraySizesPositionIndependent(column, *stream, offset, limit);
         else
             DataTypeNumber<ColumnArray::Offset>().serializeBinaryBulk(*column_array.getOffsetsPtr(), *stream, offset, limit);
     }
 
     /// Then serialize contents of arrays.
-    path.back() = Substream::ArrayElements;
+    settings.path.back() = Substream::ArrayElements;
     const ColumnArray::Offsets & offset_values = column_array.getOffsets();
 
     if (offset > offset_values.size())
@@ -197,30 +227,29 @@ void DataTypeArray::serializeBinaryBulkWithMultipleStreams(
         : 0;
 
     if (limit == 0 || nested_limit)
-        nested->serializeBinaryBulkWithMultipleStreams(column_array.getData(), getter, nested_offset, nested_limit, position_independent_encoding, path);
+        nested->serializeBinaryBulkWithMultipleStreams(column_array.getData(), nested_offset, nested_limit, settings, state);
+    settings.path.pop_back();
 }
 
 
 void DataTypeArray::deserializeBinaryBulkWithMultipleStreams(
     IColumn & column,
-    InputStreamGetter getter,
     size_t limit,
-    double /*avg_value_size_hint*/,
-    bool position_independent_encoding,
-    SubstreamPath path) const
+    DeserializeBinaryBulkSettings & settings,
+    DeserializeBinaryBulkStatePtr & state) const
 {
     ColumnArray & column_array = typeid_cast<ColumnArray &>(column);
 
-    path.push_back(Substream::ArraySizes);
-    if (auto stream = getter(path))
+    settings.path.push_back(Substream::ArraySizes);
+    if (auto stream = settings.getter(settings.path))
     {
-        if (position_independent_encoding)
+        if (settings.position_independent_encoding)
             deserializeArraySizesPositionIndependent(column, *stream, limit);
         else
             DataTypeNumber<ColumnArray::Offset>().deserializeBinaryBulk(column_array.getOffsetsColumn(), *stream, limit, 0);
     }
 
-    path.back() = Substream::ArrayElements;
+    settings.path.back() = Substream::ArrayElements;
 
     ColumnArray::Offsets & offset_values = column_array.getOffsets();
     IColumn & nested_column = column_array.getData();
@@ -230,7 +259,8 @@ void DataTypeArray::deserializeBinaryBulkWithMultipleStreams(
     if (last_offset < nested_column.size())
         throw Exception("Nested column is longer than last offset", ErrorCodes::LOGICAL_ERROR);
     size_t nested_limit = last_offset - nested_column.size();
-    nested->deserializeBinaryBulkWithMultipleStreams(nested_column, getter, nested_limit, 0, position_independent_encoding, path);
+    nested->deserializeBinaryBulkWithMultipleStreams(nested_column, nested_limit, settings, state);
+    settings.path.pop_back();
 
     /// Check consistency between offsets and elements subcolumns.
     /// But if elements column is empty - it's ok for columns of Nested types that was added by ALTER.

dbms/src/DataTypes/DataTypeArray.h

@@ -61,23 +61,32 @@ public:
       * This is necessary, because when implementing nested structures, several arrays can have common sizes.
       */
 
-    void enumerateStreams(StreamCallback callback, SubstreamPath path) const override;
+    void enumerateStreams(const StreamCallback & callback, SubstreamPath & path) const override;
+
+    void serializeBinaryBulkStatePrefix(
+            SerializeBinaryBulkSettings & settings,
+            SerializeBinaryBulkStatePtr & state) const override;
+
+    void serializeBinaryBulkStateSuffix(
+            SerializeBinaryBulkSettings & settings,
+            SerializeBinaryBulkStatePtr & state) const override;
+
+    void deserializeBinaryBulkStatePrefix(
+            DeserializeBinaryBulkSettings & settings,
+            DeserializeBinaryBulkStatePtr & state) const override;
 
     void serializeBinaryBulkWithMultipleStreams(
             const IColumn & column,
-        OutputStreamGetter getter,
             size_t offset,
             size_t limit,
-        bool position_independent_encoding,
-        SubstreamPath path) const override;
+            SerializeBinaryBulkSettings & settings,
+            SerializeBinaryBulkStatePtr & state) const override;
 
     void deserializeBinaryBulkWithMultipleStreams(
             IColumn & column,
-        InputStreamGetter getter,
             size_t limit,
-        double avg_value_size_hint,
-        bool position_independent_encoding,
-        SubstreamPath path) const override;
+            DeserializeBinaryBulkSettings & settings,
+            DeserializeBinaryBulkStatePtr & state) const override;
 
     MutableColumnPtr createColumn() const override;
 

dbms/src/DataTypes/DataTypeFactory.cpp

@@ -6,6 +6,7 @@
 #include <Parsers/ASTLiteral.h>
 #include <Common/typeid_cast.h>
 #include <Poco/String.h>
+#include <Common/StringUtils/StringUtils.h>
 
 
 namespace DB
@@ -55,6 +56,23 @@ DataTypePtr DataTypeFactory::get(const String & family_name_param, const ASTPtr
 {
     String family_name = getAliasToOrName(family_name_param);
 
+    if (endsWith(family_name, "WithDictionary"))
+    {
+        ASTPtr low_cardinality_params = std::make_shared<ASTExpressionList>();
+        String param_name = family_name.substr(0, family_name.size() - strlen("WithDictionary"));
+        if (parameters)
+        {
+            auto func = std::make_shared<ASTFunction>();
+            func->name = param_name;
+            func->arguments = parameters;
+            low_cardinality_params->children.push_back(func);
+        }
+        else
+            low_cardinality_params->children.push_back(std::make_shared<ASTIdentifier>(param_name));
+
+        return get("LowCardinality", low_cardinality_params);
+    }
+
     {
         DataTypesDictionary::const_iterator it = data_types.find(family_name);
         if (data_types.end() != it)
@@ -124,6 +142,7 @@ void registerDataTypeUUID(DataTypeFactory & factory);
 void registerDataTypeAggregateFunction(DataTypeFactory & factory);
 void registerDataTypeNested(DataTypeFactory & factory);
 void registerDataTypeInterval(DataTypeFactory & factory);
+void registerDataTypeWithDictionary(DataTypeFactory & factory);
 
 
 DataTypeFactory::DataTypeFactory()
@@ -142,6 +161,7 @@ DataTypeFactory::DataTypeFactory()
     registerDataTypeAggregateFunction(*this);
     registerDataTypeNested(*this);
     registerDataTypeInterval(*this);
+    registerDataTypeWithDictionary(*this);
 }
 
 }

dbms/src/DataTypes/DataTypeNullable.cpp

@@ -37,53 +37,83 @@ bool DataTypeNullable::onlyNull() const
 }
 
 
-void DataTypeNullable::enumerateStreams(StreamCallback callback, SubstreamPath path) const
+void DataTypeNullable::enumerateStreams(const StreamCallback & callback, SubstreamPath & path) const
 {
     path.push_back(Substream::NullMap);
     callback(path);
     path.back() = Substream::NullableElements;
     nested_data_type->enumerateStreams(callback, path);
+    path.pop_back();
+}
+
+
+void DataTypeNullable::serializeBinaryBulkStatePrefix(
+        SerializeBinaryBulkSettings & settings,
+        SerializeBinaryBulkStatePtr & state) const
+{
+    settings.path.push_back(Substream::NullableElements);
+    nested_data_type->serializeBinaryBulkStatePrefix(settings, state);
+    settings.path.pop_back();
+}
+
+
+void DataTypeNullable::serializeBinaryBulkStateSuffix(
+    SerializeBinaryBulkSettings & settings,
+    SerializeBinaryBulkStatePtr & state) const
+{
+    settings.path.push_back(Substream::NullableElements);
+    nested_data_type->serializeBinaryBulkStateSuffix(settings, state);
+    settings.path.pop_back();
+}
+
+
+void DataTypeNullable::deserializeBinaryBulkStatePrefix(
+    DeserializeBinaryBulkSettings & settings,
+    DeserializeBinaryBulkStatePtr & state) const
+{
+    settings.path.push_back(Substream::NullableElements);
+    nested_data_type->deserializeBinaryBulkStatePrefix(settings, state);
+    settings.path.pop_back();
 }
 
 
 void DataTypeNullable::serializeBinaryBulkWithMultipleStreams(
     const IColumn & column,
-    OutputStreamGetter getter,
     size_t offset,
     size_t limit,
-    bool position_independent_encoding,
-    SubstreamPath path) const
+    SerializeBinaryBulkSettings & settings,
+    SerializeBinaryBulkStatePtr & state) const
 {
     const ColumnNullable & col = static_cast<const ColumnNullable &>(column);
     col.checkConsistency();
 
     /// First serialize null map.
-    path.push_back(Substream::NullMap);
-    if (auto stream = getter(path))
+    settings.path.push_back(Substream::NullMap);
+    if (auto stream = settings.getter(settings.path))
         DataTypeUInt8().serializeBinaryBulk(col.getNullMapColumn(), *stream, offset, limit);
 
     /// Then serialize contents of arrays.
-    path.back() = Substream::NullableElements;
-    nested_data_type->serializeBinaryBulkWithMultipleStreams(col.getNestedColumn(), getter, offset, limit, position_independent_encoding, path);
+    settings.path.back() = Substream::NullableElements;
+    nested_data_type->serializeBinaryBulkWithMultipleStreams(col.getNestedColumn(), offset, limit, settings, state);
+    settings.path.pop_back();
 }
 
 
 void DataTypeNullable::deserializeBinaryBulkWithMultipleStreams(
     IColumn & column,
-    InputStreamGetter getter,
     size_t limit,
-    double avg_value_size_hint,
-    bool position_independent_encoding,
-    SubstreamPath path) const
+    DeserializeBinaryBulkSettings & settings,
+    DeserializeBinaryBulkStatePtr & state) const
 {
     ColumnNullable & col = static_cast<ColumnNullable &>(column);
 
-    path.push_back(Substream::NullMap);
-    if (auto stream = getter(path))
+    settings.path.push_back(Substream::NullMap);
+    if (auto stream = settings.getter(settings.path))
         DataTypeUInt8().deserializeBinaryBulk(col.getNullMapColumn(), *stream, limit, 0);
 
-    path.back() = Substream::NullableElements;
-    nested_data_type->deserializeBinaryBulkWithMultipleStreams(col.getNestedColumn(), getter, limit, avg_value_size_hint, position_independent_encoding, path);
+    settings.path.back() = Substream::NullableElements;
+    nested_data_type->deserializeBinaryBulkWithMultipleStreams(col.getNestedColumn(), limit, settings, state);
+    settings.path.pop_back();
 }
 
 

dbms/src/DataTypes/DataTypeNullable.h

@@ -13,27 +13,36 @@ class DataTypeNullable final : public IDataType
 public:
     static constexpr bool is_parametric = true;
 
-    DataTypeNullable(const DataTypePtr & nested_data_type_);
+    explicit DataTypeNullable(const DataTypePtr & nested_data_type_);
     std::string getName() const override { return "Nullable(" + nested_data_type->getName() + ")"; }
     const char * getFamilyName() const override { return "Nullable"; }
 
-    void enumerateStreams(StreamCallback callback, SubstreamPath path) const override;
+    void enumerateStreams(const StreamCallback & callback, SubstreamPath & path) const override;
+
+    void serializeBinaryBulkStatePrefix(
+            SerializeBinaryBulkSettings & settings,
+            SerializeBinaryBulkStatePtr & state) const override;
+
+    void serializeBinaryBulkStateSuffix(
+            SerializeBinaryBulkSettings & settings,
+            SerializeBinaryBulkStatePtr & state) const override;
+
+    void deserializeBinaryBulkStatePrefix(
+            DeserializeBinaryBulkSettings & settings,
+            DeserializeBinaryBulkStatePtr & state) const override;
 
     void serializeBinaryBulkWithMultipleStreams(
             const IColumn & column,
-        OutputStreamGetter getter,
             size_t offset,
             size_t limit,
-        bool position_independent_encoding,
-        SubstreamPath path) const override;
+            SerializeBinaryBulkSettings & settings,
+            SerializeBinaryBulkStatePtr & state) const override;
 
     void deserializeBinaryBulkWithMultipleStreams(
             IColumn & column,
-        InputStreamGetter getter,
             size_t limit,
-        double avg_value_size_hint,
-        bool position_independent_encoding,
-        SubstreamPath path) const override;
+            DeserializeBinaryBulkSettings & settings,
+            DeserializeBinaryBulkStatePtr & state) const override;
 
     void serializeBinary(const Field & field, WriteBuffer & ostr) const override { nested_data_type->serializeBinary(field, ostr); }
     void deserializeBinary(Field & field, ReadBuffer & istr) const override { nested_data_type->deserializeBinary(field, istr); }

dbms/src/DataTypes/DataTypeTuple.cpp

@@ -283,7 +283,7 @@ void DataTypeTuple::deserializeTextCSV(IColumn & column, ReadBuffer & istr, cons
     });
 }
 
-void DataTypeTuple::enumerateStreams(StreamCallback callback, SubstreamPath path) const
+void DataTypeTuple::enumerateStreams(const StreamCallback & callback, SubstreamPath & path) const
 {
     path.push_back(Substream::TupleElement);
     for (const auto i : ext::range(0, ext::size(elems)))
@@ -291,40 +291,139 @@ void DataTypeTuple::enumerateStreams(StreamCallback callback, SubstreamPath path
         path.back().tuple_element_name = names[i];
         elems[i]->enumerateStreams(callback, path);
     }
+    path.pop_back();
+}
+
+struct SerializeBinaryBulkStateTuple : public IDataType::SerializeBinaryBulkState
+{
+    std::vector<IDataType::SerializeBinaryBulkStatePtr> states;
+};
+
+struct DeserializeBinaryBulkStateTuple : public IDataType::DeserializeBinaryBulkState
+{
+    std::vector<IDataType::DeserializeBinaryBulkStatePtr> states;
+};
+
+static SerializeBinaryBulkStateTuple * checkAndGetTupleSerializeState(IDataType::SerializeBinaryBulkStatePtr & state)
+{
+    if (!state)
+        throw Exception("Got empty state for DataTypeTuple.", ErrorCodes::LOGICAL_ERROR);
+
+    auto * tuple_state = typeid_cast<SerializeBinaryBulkStateTuple *>(state.get());
+    if (!tuple_state)
+    {
+        auto & state_ref = *state;
+        throw Exception("Invalid SerializeBinaryBulkState for DataTypeTuple. Expected: "
+                        + demangle(typeid(SerializeBinaryBulkStateTuple).name()) + ", got "
+                        + demangle(typeid(state_ref).name()), ErrorCodes::LOGICAL_ERROR);
+    }
+
+    return tuple_state;
+}
+
+static DeserializeBinaryBulkStateTuple * checkAndGetTupleDeserializeState(IDataType::DeserializeBinaryBulkStatePtr & state)
+{
+    if (!state)
+        throw Exception("Got empty state for DataTypeTuple.", ErrorCodes::LOGICAL_ERROR);
+
+    auto * tuple_state = typeid_cast<DeserializeBinaryBulkStateTuple *>(state.get());
+    if (!tuple_state)
+    {
+        auto & state_ref = *state;
+        throw Exception("Invalid DeserializeBinaryBulkState for DataTypeTuple. Expected: "
+                        + demangle(typeid(DeserializeBinaryBulkStateTuple).name()) + ", got "
+                        + demangle(typeid(state_ref).name()), ErrorCodes::LOGICAL_ERROR);
+    }
+
+    return tuple_state;
+}
+
+void DataTypeTuple::serializeBinaryBulkStatePrefix(
+    SerializeBinaryBulkSettings & settings,
+    SerializeBinaryBulkStatePtr & state) const
+{
+    auto tuple_state = std::make_shared<SerializeBinaryBulkStateTuple>();
+    tuple_state->states.resize(elems.size());
+
+    settings.path.push_back(Substream::TupleElement);
+    for (size_t i = 0; i < elems.size(); ++i)
+    {
+        settings.path.back().tuple_element_name = names[i];
+        elems[i]->serializeBinaryBulkStatePrefix(settings, tuple_state->states[i]);
+    }
+    settings.path.pop_back();
+
+    state = std::move(tuple_state);
+}
+
+void DataTypeTuple::serializeBinaryBulkStateSuffix(
+    SerializeBinaryBulkSettings & settings,
+    SerializeBinaryBulkStatePtr & state) const
+{
+    auto * tuple_state = checkAndGetTupleSerializeState(state);
+
+    settings.path.push_back(Substream::TupleElement);
+    for (size_t i = 0; i < elems.size(); ++i)
+    {
+        settings.path.back().tuple_element_name = names[i];
+        elems[i]->serializeBinaryBulkStateSuffix(settings, tuple_state->states[i]);
+    }
+    settings.path.pop_back();
+}
+
+void DataTypeTuple::deserializeBinaryBulkStatePrefix(
+        DeserializeBinaryBulkSettings & settings,
+        DeserializeBinaryBulkStatePtr & state) const
+{
+    auto tuple_state = std::make_shared<DeserializeBinaryBulkStateTuple>();
+    tuple_state->states.resize(elems.size());
+
+    settings.path.push_back(Substream::TupleElement);
+    for (size_t i = 0; i < elems.size(); ++i)
+    {
+        settings.path.back().tuple_element_name = names[i];
+        elems[i]->deserializeBinaryBulkStatePrefix(settings, tuple_state->states[i]);
+    }
+    settings.path.pop_back();
+
+    state = std::move(tuple_state);
 }
 
 void DataTypeTuple::serializeBinaryBulkWithMultipleStreams(
     const IColumn & column,
-    OutputStreamGetter getter,
     size_t offset,
     size_t limit,
-    bool position_independent_encoding,
-    SubstreamPath path) const
+    SerializeBinaryBulkSettings & settings,
+    SerializeBinaryBulkStatePtr & state) const
 {
-    path.push_back(Substream::TupleElement);
+    auto * tuple_state = checkAndGetTupleSerializeState(state);
+
+    settings.path.push_back(Substream::TupleElement);
     for (const auto i : ext::range(0, ext::size(elems)))
     {
-        path.back().tuple_element_name = names[i];
-        elems[i]->serializeBinaryBulkWithMultipleStreams(
-            extractElementColumn(column, i), getter, offset, limit, position_independent_encoding, path);
+        settings.path.back().tuple_element_name = names[i];
+        auto & element_col = extractElementColumn(column, i);
+        elems[i]->serializeBinaryBulkWithMultipleStreams(element_col, offset, limit, settings, tuple_state->states[i]);
     }
+    settings.path.pop_back();
 }
 
 void DataTypeTuple::deserializeBinaryBulkWithMultipleStreams(
     IColumn & column,
-    InputStreamGetter getter,
     size_t limit,
-    double avg_value_size_hint,
-    bool position_independent_encoding,
-    SubstreamPath path) const
+    DeserializeBinaryBulkSettings & settings,
+    DeserializeBinaryBulkStatePtr & state) const
 {
-    path.push_back(Substream::TupleElement);
+    auto * tuple_state = checkAndGetTupleDeserializeState(state);
+
+    settings.path.push_back(Substream::TupleElement);
     for (const auto i : ext::range(0, ext::size(elems)))
     {
-        path.back().tuple_element_name = names[i];
-        elems[i]->deserializeBinaryBulkWithMultipleStreams(
-            extractElementColumn(column, i), getter, limit, avg_value_size_hint, position_independent_encoding, path);
+        settings.path.back().tuple_element_name = names[i];
+        auto & element_col = extractElementColumn(column, i);
+        elems[i]->deserializeBinaryBulkWithMultipleStreams(element_col, limit, settings, tuple_state->states[i]);
     }
+    settings.path.pop_back();
 }
 
 MutableColumnPtr DataTypeTuple::createColumn() const

dbms/src/DataTypes/DataTypeTuple.h

@@ -53,23 +53,32 @@ public:
 
     /** Each sub-column in a tuple is serialized in separate stream.
       */
-    void enumerateStreams(StreamCallback callback, SubstreamPath path) const override;
+    void enumerateStreams(const StreamCallback & callback, SubstreamPath & path) const override;
+
+    void serializeBinaryBulkStatePrefix(
+            SerializeBinaryBulkSettings & settings,
+            SerializeBinaryBulkStatePtr & state) const override;
+
+    void serializeBinaryBulkStateSuffix(
+            SerializeBinaryBulkSettings & settings,
+            SerializeBinaryBulkStatePtr & state) const override;
+
+    void deserializeBinaryBulkStatePrefix(
+            DeserializeBinaryBulkSettings & settings,
+            DeserializeBinaryBulkStatePtr & state) const override;
 
     void serializeBinaryBulkWithMultipleStreams(
             const IColumn & column,
-        OutputStreamGetter getter,
             size_t offset,
             size_t limit,
-        bool position_independent_encoding,
-        SubstreamPath path) const override;
+            SerializeBinaryBulkSettings & settings,
+            SerializeBinaryBulkStatePtr & state) const override;
 
     void deserializeBinaryBulkWithMultipleStreams(
             IColumn & column,
-        InputStreamGetter getter,
             size_t limit,
-        double avg_value_size_hint,
-        bool position_independent_encoding,
-        SubstreamPath path) const override;
+            DeserializeBinaryBulkSettings & settings,
+            DeserializeBinaryBulkStatePtr & state) const override;
 
     MutableColumnPtr createColumn() const override;
 

dbms/src/DataTypes/DataTypeWithDictionary.cpp

@@ -0,0 +1,840 @@
+#include <Columns/ColumnWithDictionary.h>
+#include <Columns/ColumnUnique.h>
+#include <Columns/ColumnFixedString.h>
+#include <Columns/ColumnsCommon.h>
+#include <Common/HashTable/HashMap.h>
+#include <Common/typeid_cast.h>
+#include <Core/TypeListNumber.h>
+#include <DataTypes/DataTypeFactory.h>
+#include <DataTypes/DataTypeWithDictionary.h>
+#include <DataTypes/DataTypeNullable.h>
+#include <DataTypes/DataTypeDate.h>
+#include <DataTypes/DataTypeDateTime.h>
+#include <Parsers/IAST.h>
+
+namespace DB
+{
+
+namespace ErrorCodes
+{
+    extern const int NUMBER_OF_ARGUMENTS_DOESNT_MATCH;
+    extern const int LOGICAL_ERROR;
+    extern const int ILLEGAL_TYPE_OF_ARGUMENT;
+}
+
+namespace
+{
+    const ColumnWithDictionary & getColumnWithDictionary(const IColumn & column)
+    {
+        return typeid_cast<const ColumnWithDictionary &>(column);
+    }
+
+    ColumnWithDictionary & getColumnWithDictionary(IColumn & column)
+    {
+        return typeid_cast<ColumnWithDictionary &>(column);
+    }
+}
+
+DataTypeWithDictionary::DataTypeWithDictionary(DataTypePtr dictionary_type_)
+        : dictionary_type(std::move(dictionary_type_))
+{
+    auto inner_type = dictionary_type;
+    if (dictionary_type->isNullable())
+        inner_type = static_cast<const DataTypeNullable &>(*dictionary_type).getNestedType();
+
+    if (!inner_type->isStringOrFixedString()
+        && !inner_type->isDateOrDateTime()
+        && !inner_type->isNumber())
+        throw Exception("DataTypeWithDictionary is supported only for numbers, strings, Date or DateTime, but got "
+                        + dictionary_type->getName(), ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
+}
+
+void DataTypeWithDictionary::enumerateStreams(const StreamCallback & callback, SubstreamPath & path) const
+{
+    path.push_back(Substream::DictionaryKeys);
+    dictionary_type->enumerateStreams(callback, path);
+    path.back() = Substream::DictionaryIndexes;
+    callback(path);
+    path.pop_back();
+}
+
+struct  KeysSerializationVersion
+{
+    enum Value
+    {
+        /// Version is written at the start of <name.dict.bin>.
+        /// Dictionary is written as number N and N keys after them.
+        /// Dictionary can be shared for continuous range of granules, so some marks may point to the same position.
+        /// Shared dictionary is stored in state and is read once.
+        SharedDictionariesWithAdditionalKeys = 1,
+    };
+
+    Value value;
+
+    static void checkVersion(UInt64 version)
+    {
+        if (version != SharedDictionariesWithAdditionalKeys)
+            throw Exception("Invalid version for DataTypeWithDictionary key column.", ErrorCodes::LOGICAL_ERROR);
+    }
+
+    KeysSerializationVersion(UInt64 version) : value(static_cast<Value>(version)) { checkVersion(version); }
+};
+
+/// Version is stored at the start of each granule. It's used to store indexes type and flags.
+struct IndexesSerializationType
+{
+    using SerializationType = UInt64;
+    /// Need to read dictionary if it wasn't.
+    static constexpr SerializationType NeedGlobalDictionaryBit = 1u << 8u;
+    /// Need to read additional keys. Additional keys are stored before indexes as value N and N keys after them.
+    static constexpr SerializationType HasAdditionalKeysBit = 1u << 9u;
+    /// Need to update dictionary. It means that previous granule has different dictionary.
+    static constexpr SerializationType NeedUpdateDictionary = 1u << 10u;
+
+    enum Type
+    {
+        TUInt8 = 0,
+        TUInt16,
+        TUInt32,
+        TUInt64,
+    };
+
+    Type type;
+    bool has_additional_keys;
+    bool need_global_dictionary;
+    bool need_update_dictionary;
+
+    static constexpr SerializationType resetFlags(SerializationType type)
+    {
+        return type & (~(HasAdditionalKeysBit | NeedGlobalDictionaryBit | NeedUpdateDictionary));
+    }
+
+    static void checkType(SerializationType type)
+    {
+        UInt64 value = resetFlags(type);
+        if (value <= TUInt64)
+            return;
+
+        throw Exception("Invalid type for DataTypeWithDictionary index column.", ErrorCodes::LOGICAL_ERROR);
+    }
+
+    void serialize(WriteBuffer & buffer) const
+    {
+        SerializationType val = type;
+        if (has_additional_keys)
+            val |= HasAdditionalKeysBit;
+        if (need_global_dictionary)
+            val |= NeedGlobalDictionaryBit;
+        if (need_update_dictionary)
+            val |= NeedUpdateDictionary;
+        writeIntBinary(val, buffer);
+    }
+
+    void deserialize(ReadBuffer & buffer)
+    {
+        SerializationType val;
+        readIntBinary(val, buffer);
+        checkType(val);
+        has_additional_keys = (val & HasAdditionalKeysBit) != 0;
+        need_global_dictionary = (val & NeedGlobalDictionaryBit) != 0;
+        need_update_dictionary = (val & NeedUpdateDictionary) != 0;
+        type = static_cast<Type>(resetFlags(val));
+    }
+
+    IndexesSerializationType(const IColumn & column,
+                             bool has_additional_keys,
+                             bool need_global_dictionary,
+                             bool enumerate_dictionaries)
+        : has_additional_keys(has_additional_keys)
+        , need_global_dictionary(need_global_dictionary)
+        , need_update_dictionary(enumerate_dictionaries)
+    {
+        if (typeid_cast<const ColumnUInt8 *>(&column))
+            type = TUInt8;
+        else if (typeid_cast<const ColumnUInt16 *>(&column))
+            type = TUInt16;
+        else if (typeid_cast<const ColumnUInt32 *>(&column))
+            type = TUInt32;
+        else if (typeid_cast<const ColumnUInt64 *>(&column))
+            type = TUInt64;
+        else
+            throw Exception("Invalid Indexes column for IndexesSerializationType. Expected ColumnUInt*, got "
+                            + column.getName(), ErrorCodes::LOGICAL_ERROR);
+    }
+
+    DataTypePtr getDataType() const
+    {
+        if (type == TUInt8)
+            return std::make_shared<DataTypeUInt8>();
+        if (type == TUInt16)
+            return std::make_shared<DataTypeUInt16>();
+        if (type == TUInt32)
+            return std::make_shared<DataTypeUInt32>();
+        if (type == TUInt64)
+            return std::make_shared<DataTypeUInt64>();
+
+        throw Exception("Can't create DataType from IndexesSerializationType.", ErrorCodes::LOGICAL_ERROR);
+    }
+
+    IndexesSerializationType() = default;
+};
+
+struct SerializeStateWithDictionary : public IDataType::SerializeBinaryBulkState
+{
+    KeysSerializationVersion key_version;
+    MutableColumnUniquePtr shared_dictionary;
+
+    explicit SerializeStateWithDictionary(UInt64 key_version) : key_version(key_version) {}
+};
+
+struct DeserializeStateWithDictionary : public IDataType::DeserializeBinaryBulkState
+{
+    KeysSerializationVersion key_version;
+    ColumnUniquePtr global_dictionary;
+
+    IndexesSerializationType index_type;
+    ColumnPtr additional_keys;
+    UInt64 num_pending_rows = 0;
+
+    explicit DeserializeStateWithDictionary(UInt64 key_version) : key_version(key_version) {}
+};
+
+static SerializeStateWithDictionary * checkAndGetWithDictionarySerializeState(
+    IDataType::SerializeBinaryBulkStatePtr & state)
+{
+    if (!state)
+        throw Exception("Got empty state for DataTypeWithDictionary.", ErrorCodes::LOGICAL_ERROR);
+
+    auto * with_dictionary_state = typeid_cast<SerializeStateWithDictionary *>(state.get());
+    if (!with_dictionary_state)
+    {
+        auto & state_ref = *state;
+        throw Exception("Invalid SerializeBinaryBulkState for DataTypeWithDictionary. Expected: "
+                        + demangle(typeid(SerializeStateWithDictionary).name()) + ", got "
+                        + demangle(typeid(state_ref).name()), ErrorCodes::LOGICAL_ERROR);
+    }
+
+    return with_dictionary_state;
+}
+
+static DeserializeStateWithDictionary * checkAndGetWithDictionaryDeserializeState(
+    IDataType::DeserializeBinaryBulkStatePtr & state)
+{
+    if (!state)
+        throw Exception("Got empty state for DataTypeWithDictionary.", ErrorCodes::LOGICAL_ERROR);
+
+    auto * with_dictionary_state = typeid_cast<DeserializeStateWithDictionary *>(state.get());
+    if (!with_dictionary_state)
+    {
+        auto & state_ref = *state;
+        throw Exception("Invalid DeserializeBinaryBulkState for DataTypeWithDictionary. Expected: "
+                        + demangle(typeid(DeserializeStateWithDictionary).name()) + ", got "
+                        + demangle(typeid(state_ref).name()), ErrorCodes::LOGICAL_ERROR);
+    }
+
+    return with_dictionary_state;
+}
+
+void DataTypeWithDictionary::serializeBinaryBulkStatePrefix(
+    SerializeBinaryBulkSettings & settings,
+    SerializeBinaryBulkStatePtr & state) const
+{
+    settings.path.push_back(Substream::DictionaryKeys);
+    auto * stream = settings.getter(settings.path);
+    settings.path.pop_back();
+
+    if (!stream)
+        throw Exception("Got empty stream in DataTypeWithDictionary::serializeBinaryBulkStatePrefix",
+                        ErrorCodes::LOGICAL_ERROR);
+
+    /// Write version and create SerializeBinaryBulkState.
+    UInt64 key_version = KeysSerializationVersion::SharedDictionariesWithAdditionalKeys;
+
+    writeIntBinary(key_version, *stream);
+
+    state = std::make_shared<SerializeStateWithDictionary>(key_version);
+}
+
+void DataTypeWithDictionary::serializeBinaryBulkStateSuffix(
+    SerializeBinaryBulkSettings & settings,
+    SerializeBinaryBulkStatePtr & state) const
+{
+    auto * state_with_dictionary = checkAndGetWithDictionarySerializeState(state);
+    KeysSerializationVersion::checkVersion(state_with_dictionary->key_version.value);
+
+    if (state_with_dictionary->shared_dictionary && settings.low_cardinality_max_dictionary_size)
+    {
+        auto nested_column = state_with_dictionary->shared_dictionary->getNestedNotNullableColumn();
+
+        settings.path.push_back(Substream::DictionaryKeys);
+        auto * stream = settings.getter(settings.path);
+        settings.path.pop_back();
+
+        if (!stream)
+            throw Exception("Got empty stream in DataTypeWithDictionary::serializeBinaryBulkStateSuffix",
+                            ErrorCodes::LOGICAL_ERROR);
+
+        UInt64 num_keys = nested_column->size();
+        writeIntBinary(num_keys, *stream);
+        removeNullable(dictionary_type)->serializeBinaryBulk(*nested_column, *stream, 0, num_keys);
+        state_with_dictionary->shared_dictionary = nullptr;
+    }
+}
+
+void DataTypeWithDictionary::deserializeBinaryBulkStatePrefix(
+    DeserializeBinaryBulkSettings & settings,
+    DeserializeBinaryBulkStatePtr & state) const
+{
+    settings.path.push_back(Substream::DictionaryKeys);
+    auto * stream = settings.getter(settings.path);
+    settings.path.pop_back();
+
+    if (!stream)
+        throw Exception("Got empty stream in DataTypeWithDictionary::deserializeBinaryBulkStatePrefix",
+                        ErrorCodes::LOGICAL_ERROR);
+
+    UInt64 keys_version;
+    readIntBinary(keys_version, *stream);
+
+    state = std::make_shared<DeserializeStateWithDictionary>(keys_version);
+}
+
+namespace
+{
+    template <typename T>
+    PaddedPODArray<T> * getIndexesData(IColumn & indexes)
+    {
+        auto * column = typeid_cast<ColumnVector<T> *>(&indexes);
+        if (column)
+            return &column->getData();
+
+        return nullptr;
+    }
+
+    struct IndexMapsWithAdditionalKeys
+    {
+        MutableColumnPtr dictionary_map;
+        MutableColumnPtr additional_keys_map;
+    };
+
+    template <typename T>
+    IndexMapsWithAdditionalKeys mapIndexWithAdditionalKeysRef(PaddedPODArray<T> & index, size_t dict_size)
+    {
+        PaddedPODArray<T> copy(index.cbegin(), index.cend());
+
+        HashMap<T, T> dict_map;
+        HashMap<T, T> add_keys_map;
+
+        for (auto val : index)
+        {
+            if (val < dict_size)
+                dict_map.insert({val, dict_map.size()});
+            else
+                add_keys_map.insert({val, add_keys_map.size()});
+        }
+
+        auto dictionary_map = ColumnVector<T>::create(dict_map.size());
+        auto additional_keys_map = ColumnVector<T>::create(add_keys_map.size());
+        auto & dict_data = dictionary_map->getData();
+        auto & add_keys_data = additional_keys_map->getData();
+
+        for (auto val : dict_map)
+            dict_data[val.second] = val.first;
+
+        for (auto val : add_keys_map)
+            add_keys_data[val.second] = val.first - dict_size;
+
+        for (auto & val : index)
+            val = val < dict_size ? dict_map[val]
+                                  : add_keys_map[val] + dict_map.size();
+
+        for (size_t i = 0; i < index.size(); ++i)
+        {
+            T expected = index[i] < dict_data.size() ? dict_data[index[i]]
+                                                     : add_keys_data[index[i] - dict_data.size()] + dict_size;
+            if (expected != copy[i])
+                throw Exception("Expected " + toString(expected) + ", but got " + toString(copy[i]), ErrorCodes::LOGICAL_ERROR);
+
+        }
+
+        return {std::move(dictionary_map), std::move(additional_keys_map)};
+    }
+
+    template <typename T>
+    IndexMapsWithAdditionalKeys mapIndexWithAdditionalKeys(PaddedPODArray<T> & index, size_t dict_size)
+    {
+        T max_less_dict_size = 0;
+        T max_value = 0;
+
+        auto size = index.size();
+        if (size == 0)
+            return {ColumnVector<T>::create(), ColumnVector<T>::create()};
+
+        for (size_t i = 0; i < size; ++i)
+        {
+            auto val = index[i];
+            if (val < dict_size)
+                max_less_dict_size = std::max(max_less_dict_size, val);
+
+            max_value = std::max(max_value, val);
+        }
+
+        auto map_size = UInt64(max_less_dict_size) + 1;
+        auto overflow_map_size = max_value >= dict_size ? (UInt64(max_value - dict_size) + 1) : 0;
+        PaddedPODArray<T> map(map_size, 0);
+        PaddedPODArray<T> overflow_map(overflow_map_size, 0);
+
+        T zero_pos_value = 0;
+        T zero_pos_overflowed_value = 0;
+        UInt64 cur_pos = 0;
+        UInt64 cur_overflowed_pos = 0;
+
+        for (size_t i = 0; i < size; ++i)
+        {
+            T val = index[i];
+            if (val < dict_size)
+            {
+                if (cur_pos == 0)
+                {
+                    zero_pos_value = val;
+                    ++cur_pos;
+                }
+                else if (map[val] == 0 && val != zero_pos_value)
+                {
+                    map[val] = cur_pos;
+                    ++cur_pos;
+                }
+            }
+            else
+            {
+                T shifted_val = val - dict_size;
+                if (cur_overflowed_pos == 0)
+                {
+                    zero_pos_overflowed_value = shifted_val;
+                    ++cur_overflowed_pos;
+                }
+                else if (overflow_map[shifted_val] == 0 && shifted_val != zero_pos_overflowed_value)
+                {
+                    overflow_map[shifted_val] = cur_overflowed_pos;
+                    ++cur_overflowed_pos;
+                }
+            }
+        }
+
+        auto dictionary_map = ColumnVector<T>::create(cur_pos);
+        auto additional_keys_map = ColumnVector<T>::create(cur_overflowed_pos);
+        auto & dict_data = dictionary_map->getData();
+        auto & add_keys_data = additional_keys_map->getData();
+
+        for (size_t i = 0; i < map_size; ++i)
+            if (map[i])
+                dict_data[map[i]] = static_cast<T>(i);
+
+        for (size_t i = 0; i < overflow_map_size; ++i)
+            if (overflow_map[i])
+                add_keys_data[overflow_map[i]] = static_cast<T>(i);
+
+        if (cur_pos)
+            dict_data[0] = zero_pos_value;
+        if (cur_overflowed_pos)
+            add_keys_data[0] = zero_pos_overflowed_value;
+
+        for (size_t i = 0; i < size; ++i)
+        {
+            T & val = index[i];
+            if (val < dict_size)
+                val = map[val];
+            else
+                val = overflow_map[val - dict_size] + cur_pos;
+        }
+
+        return {std::move(dictionary_map), std::move(additional_keys_map)};
+    }
+
+    /// Update column and return map with old indexes.
+    /// Let N is the number of distinct values which are less than max_size;
+    ///     old_column - column before function call;
+    ///     new_column - column after function call:
+    /// * if old_column[i] < max_size, than
+    ///       dictionary_map[new_column[i]] = old_column[i]
+    /// * else
+    ///       additional_keys_map[new_column[i]] = old_column[i] - dict_size + N
+    IndexMapsWithAdditionalKeys mapIndexWithAdditionalKeys(IColumn & column, size_t dict_size)
+    {
+        if (auto * data_uint8 = getIndexesData<UInt8>(column))
+            return mapIndexWithAdditionalKeys(*data_uint8, dict_size);
+        else if (auto * data_uint16 = getIndexesData<UInt16>(column))
+            return mapIndexWithAdditionalKeys(*data_uint16, dict_size);
+        else if (auto * data_uint32 = getIndexesData<UInt32>(column))
+            return mapIndexWithAdditionalKeys(*data_uint32, dict_size);
+        else if (auto * data_uint64 = getIndexesData<UInt64>(column))
+            return mapIndexWithAdditionalKeys(*data_uint64, dict_size);
+        else
+            throw Exception("Indexes column for mapIndexWithAdditionalKeys must be UInt, got" + column.getName(),
+                            ErrorCodes::LOGICAL_ERROR);
+    }
+}
+
+void DataTypeWithDictionary::serializeBinaryBulkWithMultipleStreams(
+    const IColumn & column,
+    size_t offset,
+    size_t limit,
+    SerializeBinaryBulkSettings & settings,
+    SerializeBinaryBulkStatePtr & state) const
+{
+    settings.path.push_back(Substream::DictionaryKeys);
+    auto * keys_stream = settings.getter(settings.path);
+    settings.path.back() = Substream::DictionaryIndexes;
+    auto * indexes_stream = settings.getter(settings.path);
+    settings.path.pop_back();
+
+    if (!keys_stream && !indexes_stream)
+        return;
+
+    if (!keys_stream)
+        throw Exception("Got empty stream for DataTypeWithDictionary keys.", ErrorCodes::LOGICAL_ERROR);
+
+    if (!indexes_stream)
+        throw Exception("Got empty stream for DataTypeWithDictionary indexes.", ErrorCodes::LOGICAL_ERROR);
+
+    const ColumnWithDictionary & column_with_dictionary = typeid_cast<const ColumnWithDictionary &>(column);
+
+    auto * state_with_dictionary = checkAndGetWithDictionarySerializeState(state);
+    auto & global_dictionary = state_with_dictionary->shared_dictionary;
+    KeysSerializationVersion::checkVersion(state_with_dictionary->key_version.value);
+
+    bool need_update_dictionary = global_dictionary == nullptr;
+    if (need_update_dictionary)
+        global_dictionary = createColumnUnique(*dictionary_type);
+
+    size_t max_limit = column.size() - offset;
+    limit = limit ? std::min(limit, max_limit) : max_limit;
+
+    auto sub_column = column_with_dictionary.cutAndCompact(offset, limit);
+    ColumnPtr positions = sub_column->getIndexesPtr();
+    ColumnPtr keys = sub_column->getDictionary().getNestedColumn();
+
+    if (settings.low_cardinality_max_dictionary_size)
+    {
+        /// Insert used_keys into global dictionary and update sub_index.
+        auto indexes_with_overflow = global_dictionary->uniqueInsertRangeWithOverflow(*keys, 0, keys->size(),
+                                                                                      settings.low_cardinality_max_dictionary_size);
+        size_t max_size = settings.low_cardinality_max_dictionary_size + indexes_with_overflow.overflowed_keys->size();
+        ColumnWithDictionary::Index(indexes_with_overflow.indexes->getPtr()).check(max_size);
+
+        if (global_dictionary->size() > settings.low_cardinality_max_dictionary_size)
+            throw Exception("Got dictionary with size " + toString(global_dictionary->size()) +
+                            " but max dictionary size is " + toString(settings.low_cardinality_max_dictionary_size),
+                            ErrorCodes::LOGICAL_ERROR);
+
+        positions = indexes_with_overflow.indexes->index(*positions, 0);
+        keys = std::move(indexes_with_overflow.overflowed_keys);
+
+        if (global_dictionary->size() < settings.low_cardinality_max_dictionary_size && !keys->empty())
+            throw Exception("Has additional keys, but dict size is " + toString(global_dictionary->size()) +
+                            " which is less then max dictionary size (" + toString(settings.low_cardinality_max_dictionary_size) + ")",
+                            ErrorCodes::LOGICAL_ERROR);
+    }
+
+    if (auto nullable_keys = typeid_cast<const ColumnNullable *>(keys.get()))
+        keys = nullable_keys->getNestedColumnPtr();
+
+    bool need_additional_keys = !keys->empty();
+    bool need_dictionary = settings.low_cardinality_max_dictionary_size != 0;
+    bool need_write_dictionary = !settings.low_cardinality_use_single_dictionary_for_part
+                                 && global_dictionary->size() >= settings.low_cardinality_max_dictionary_size;
+
+    IndexesSerializationType index_version(*positions, need_additional_keys, need_dictionary, need_update_dictionary);
+    index_version.serialize(*indexes_stream);
+
+    if (need_write_dictionary)
+    {
+        const auto & nested_column = global_dictionary->getNestedNotNullableColumn();
+        UInt64 num_keys = nested_column->size();
+        writeIntBinary(num_keys, *keys_stream);
+        removeNullable(dictionary_type)->serializeBinaryBulk(*nested_column, *keys_stream, 0, num_keys);
+        state_with_dictionary->shared_dictionary = nullptr;
+    }
+
+    if (need_additional_keys)
+    {
+        UInt64 num_keys = keys->size();
+        writeIntBinary(num_keys, *indexes_stream);
+        removeNullable(dictionary_type)->serializeBinaryBulk(*keys, *indexes_stream, 0, num_keys);
+    }
+
+    UInt64 num_rows = positions->size();
+    writeIntBinary(num_rows, *indexes_stream);
+    index_version.getDataType()->serializeBinaryBulk(*positions, *indexes_stream, 0, num_rows);
+}
+
+void DataTypeWithDictionary::deserializeBinaryBulkWithMultipleStreams(
+    IColumn & column,
+    size_t limit,
+    DeserializeBinaryBulkSettings & settings,
+    DeserializeBinaryBulkStatePtr & state) const
+{
+    ColumnWithDictionary & column_with_dictionary = typeid_cast<ColumnWithDictionary &>(column);
+
+    auto * state_with_dictionary = checkAndGetWithDictionaryDeserializeState(state);
+    KeysSerializationVersion::checkVersion(state_with_dictionary->key_version.value);
+
+    settings.path.push_back(Substream::DictionaryKeys);
+    auto * keys_stream = settings.getter(settings.path);
+    settings.path.back() = Substream::DictionaryIndexes;
+    auto * indexes_stream = settings.getter(settings.path);
+    settings.path.pop_back();
+
+    if (!keys_stream && !indexes_stream)
+        return;
+
+    if (!keys_stream)
+        throw Exception("Got empty stream for DataTypeWithDictionary keys.", ErrorCodes::LOGICAL_ERROR);
+
+    if (!indexes_stream)
+        throw Exception("Got empty stream for DataTypeWithDictionary indexes.", ErrorCodes::LOGICAL_ERROR);
+
+    auto readDictionary = [this, state_with_dictionary, keys_stream]()
+    {
+        UInt64 num_keys;
+        readIntBinary(num_keys, *keys_stream);
+
+        auto keys_type = removeNullable(dictionary_type);
+        auto global_dict_keys = keys_type->createColumn();
+        keys_type->deserializeBinaryBulk(*global_dict_keys, *keys_stream, num_keys, 0);
+
+        auto column_unique = createColumnUnique(*dictionary_type, std::move(global_dict_keys));
+        state_with_dictionary->global_dictionary = std::move(column_unique);
+    };
+
+    auto readAdditionalKeys = [this, state_with_dictionary, indexes_stream]()
+    {
+        UInt64 num_keys;
+        readIntBinary(num_keys, *indexes_stream);
+        auto keys_type = removeNullable(dictionary_type);
+        auto additional_keys = keys_type->createColumn();
+        keys_type->deserializeBinaryBulk(*additional_keys, *indexes_stream, num_keys, 0);
+        state_with_dictionary->additional_keys = std::move(additional_keys);
+    };
+
+    auto readIndexes = [this, state_with_dictionary, indexes_stream, &column_with_dictionary](UInt64 num_rows)
+    {
+        auto indexes_type = state_with_dictionary->index_type.getDataType();
+        MutableColumnPtr indexes_column = indexes_type->createColumn();
+        indexes_type->deserializeBinaryBulk(*indexes_column, *indexes_stream, num_rows, 0);
+
+        auto & global_dictionary = state_with_dictionary->global_dictionary;
+        const auto & additional_keys = state_with_dictionary->additional_keys;
+
+        bool has_additional_keys = state_with_dictionary->index_type.has_additional_keys;
+        bool column_is_empty = column_with_dictionary.empty();
+
+        if (!state_with_dictionary->index_type.need_global_dictionary)
+        {
+            column_with_dictionary.insertRangeFromDictionaryEncodedColumn(*additional_keys, *indexes_column);
+        }
+        else if (!has_additional_keys)
+        {
+            if (column_is_empty)
+                column_with_dictionary.setSharedDictionary(global_dictionary);
+
+            auto local_column = ColumnWithDictionary::create(global_dictionary, std::move(indexes_column));
+            column_with_dictionary.insertRangeFrom(*local_column, 0, num_rows);
+        }
+        else
+        {
+            auto maps = mapIndexWithAdditionalKeys(*indexes_column, global_dictionary->size());
+
+            ColumnWithDictionary::Index(maps.additional_keys_map->getPtr()).check(additional_keys->size());
+
+            ColumnWithDictionary::Index(indexes_column->getPtr()).check(
+                    maps.dictionary_map->size() + maps.additional_keys_map->size());
+
+            auto used_keys = (*std::move(global_dictionary->getNestedColumn()->index(*maps.dictionary_map, 0))).mutate();
+
+            if (!maps.additional_keys_map->empty())
+            {
+                auto used_add_keys = additional_keys->index(*maps.additional_keys_map, 0);
+
+                if (dictionary_type->isNullable())
+                {
+                    ColumnPtr null_map = ColumnUInt8::create(used_add_keys->size(), 0);
+                    used_add_keys = ColumnNullable::create(used_add_keys, null_map);
+                }
+
+                used_keys->insertRangeFrom(*used_add_keys, 0, used_add_keys->size());
+            }
+
+            column_with_dictionary.insertRangeFromDictionaryEncodedColumn(*used_keys, *indexes_column);
+        }
+    };
+
+    while (limit)
+    {
+        if (state_with_dictionary->num_pending_rows == 0)
+        {
+            if (indexes_stream->eof())
+                break;
+
+            auto & index_type = state_with_dictionary->index_type;
+            auto & global_dictionary = state_with_dictionary->global_dictionary;
+
+            index_type.deserialize(*indexes_stream);
+
+            if (index_type.need_global_dictionary && (!global_dictionary || index_type.need_update_dictionary))
+                readDictionary();
+
+            if (state_with_dictionary->index_type.has_additional_keys)
+                readAdditionalKeys();
+            else
+                state_with_dictionary->additional_keys = nullptr;
+
+            readIntBinary(state_with_dictionary->num_pending_rows, *indexes_stream);
+        }
+
+        size_t num_rows_to_read = std::min(limit, state_with_dictionary->num_pending_rows);
+        readIndexes(num_rows_to_read);
+        limit -= num_rows_to_read;
+        state_with_dictionary->num_pending_rows -= num_rows_to_read;
+    }
+}
+
+void DataTypeWithDictionary::serializeBinary(const Field & field, WriteBuffer & ostr) const
+{
+    dictionary_type->serializeBinary(field, ostr);
+}
+void DataTypeWithDictionary::deserializeBinary(Field & field, ReadBuffer & istr) const
+{
+    dictionary_type->deserializeBinary(field, istr);
+}
+
+template <typename ... Args>
+void DataTypeWithDictionary::serializeImpl(
+        const IColumn & column, size_t row_num, WriteBuffer & ostr,
+        DataTypeWithDictionary::SerealizeFunctionPtr<Args ...> func, Args & ... args) const
+{
+    auto & column_with_dictionary = getColumnWithDictionary(column);
+    size_t unique_row_number = column_with_dictionary.getIndexes().getUInt(row_num);
+    (dictionary_type.get()->*func)(*column_with_dictionary.getDictionary().getNestedColumn(), unique_row_number, ostr, std::forward<Args>(args)...);
+}
+
+template <typename ... Args>
+void DataTypeWithDictionary::deserializeImpl(
+        IColumn & column, ReadBuffer & istr,
+        DataTypeWithDictionary::DeserealizeFunctionPtr<Args ...> func, Args & ... args) const
+{
+    auto & column_with_dictionary = getColumnWithDictionary(column);
+    auto temp_column = column_with_dictionary.getDictionary().getNestedColumn()->cloneEmpty();
+
+    (dictionary_type.get()->*func)(*temp_column, istr, std::forward<Args>(args)...);
+
+    column_with_dictionary.insertFromFullColumn(*temp_column, 0);
+}
+
+namespace
+{
+    template <typename Creator>
+    struct CreateColumnVector
+    {
+        MutableColumnUniquePtr & column;
+        const IDataType & keys_type;
+        const Creator & creator;
+
+        CreateColumnVector(MutableColumnUniquePtr & column, const IDataType & keys_type, const Creator & creator)
+                : column(column), keys_type(keys_type), creator(creator)
+        {
+        }
+
+        template <typename T, size_t>
+        void operator()()
+        {
+            if (typeid_cast<const DataTypeNumber<T> *>(&keys_type))
+                column = creator((ColumnVector<T> *)(nullptr));
+        }
+    };
+}
+
+template <typename Creator>
+MutableColumnUniquePtr DataTypeWithDictionary::createColumnUniqueImpl(const IDataType & keys_type,
+                                                                      const Creator & creator)
+{
+    auto * type = &keys_type;
+    if (auto * nullable_type = typeid_cast<const DataTypeNullable *>(&keys_type))
+        type = nullable_type->getNestedType().get();
+
+    if (type->isString())
+        return creator((ColumnString *)(nullptr));
+    if (type->isFixedString())
+        return creator((ColumnFixedString *)(nullptr));
+    if (typeid_cast<const DataTypeDate *>(type))
+        return creator((ColumnVector<UInt16> *)(nullptr));
+    if (typeid_cast<const DataTypeDateTime *>(type))
+        return creator((ColumnVector<UInt32> *)(nullptr));
+    if (type->isNumber())
+    {
+        MutableColumnUniquePtr column;
+        TypeListNumbers::forEach(CreateColumnVector(column, *type, creator));
+
+        if (!column)
+            throw Exception("Unexpected numeric type: " + type->getName(), ErrorCodes::LOGICAL_ERROR);
+
+        return column;
+    }
+
+    throw Exception("Unexpected dictionary type for DataTypeWithDictionary: " + type->getName(),
+                    ErrorCodes::LOGICAL_ERROR);
+}
+
+
+MutableColumnUniquePtr DataTypeWithDictionary::createColumnUnique(const IDataType & keys_type)
+{
+    auto creator = [&](auto x)
+    {
+        using ColumnType = typename std::remove_pointer<decltype(x)>::type;
+        return ColumnUnique<ColumnType>::create(keys_type);
+    };
+    return createColumnUniqueImpl(keys_type, creator);
+}
+
+MutableColumnUniquePtr DataTypeWithDictionary::createColumnUnique(const IDataType & keys_type, MutableColumnPtr && keys)
+{
+    auto creator = [&](auto x)
+    {
+        using ColumnType = typename std::remove_pointer<decltype(x)>::type;
+        return ColumnUnique<ColumnType>::create(std::move(keys), keys_type.isNullable());
+    };
+    return createColumnUniqueImpl(keys_type, creator);
+}
+
+MutableColumnPtr DataTypeWithDictionary::createColumn() const
+{
+    MutableColumnPtr indexes = DataTypeUInt8().createColumn();
+    MutableColumnPtr dictionary = createColumnUnique(*dictionary_type);
+    return ColumnWithDictionary::create(std::move(dictionary), std::move(indexes));
+}
+
+bool DataTypeWithDictionary::equals(const IDataType & rhs) const
+{
+    if (typeid(rhs) != typeid(*this))
+        return false;
+
+    auto & rhs_with_dictionary = static_cast<const DataTypeWithDictionary &>(rhs);
+    return dictionary_type->equals(*rhs_with_dictionary.dictionary_type);
+}
+
+
+static DataTypePtr create(const ASTPtr & arguments)
+{
+    if (!arguments || arguments->children.size() != 1)
+        throw Exception("LowCardinality data type family must have single argument - type of elements",
+                        ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
+
+    return std::make_shared<DataTypeWithDictionary>(DataTypeFactory::instance().get(arguments->children[0]));
+}
+
+void registerDataTypeWithDictionary(DataTypeFactory & factory)
+{
+    factory.registerDataType("LowCardinality", create);
+}
+
+}

dbms/src/DataTypes/DataTypeWithDictionary.h

@@ -0,0 +1,170 @@
+#pragma once
+#include <DataTypes/IDataType.h>
+#include <Columns/IColumnUnique.h>
+
+namespace DB
+{
+
+class DataTypeWithDictionary : public IDataType
+{
+private:
+    DataTypePtr dictionary_type;
+
+public:
+    DataTypeWithDictionary(DataTypePtr dictionary_type_);
+
+    const DataTypePtr & getDictionaryType() const { return dictionary_type; }
+
+    String getName() const override
+    {
+        return "LowCardinality(" + dictionary_type->getName() + ")";
+    }
+    const char * getFamilyName() const override { return "LowCardinality"; }
+
+    void enumerateStreams(const StreamCallback & callback, SubstreamPath & path) const override;
+
+    void serializeBinaryBulkStatePrefix(
+            SerializeBinaryBulkSettings & settings,
+            SerializeBinaryBulkStatePtr & state) const override;
+
+    void serializeBinaryBulkStateSuffix(
+            SerializeBinaryBulkSettings & settings,
+            SerializeBinaryBulkStatePtr & state) const override;
+
+    void deserializeBinaryBulkStatePrefix(
+            DeserializeBinaryBulkSettings & settings,
+            DeserializeBinaryBulkStatePtr & state) const override;
+
+    void serializeBinaryBulkWithMultipleStreams(
+            const IColumn & column,
+            size_t offset,
+            size_t limit,
+            SerializeBinaryBulkSettings & settings,
+            SerializeBinaryBulkStatePtr & state) const override;
+
+    void deserializeBinaryBulkWithMultipleStreams(
+            IColumn & column,
+            size_t limit,
+            DeserializeBinaryBulkSettings & settings,
+            DeserializeBinaryBulkStatePtr & state) const override;
+
+    void serializeBinary(const Field & field, WriteBuffer & ostr) const override;
+    void deserializeBinary(Field & field, ReadBuffer & istr) const override;
+
+    void serializeBinary(const IColumn & column, size_t row_num, WriteBuffer & ostr) const override
+    {
+        serializeImpl(column, row_num, ostr, &IDataType::serializeBinary);
+    }
+    void deserializeBinary(IColumn & column, ReadBuffer & istr) const override
+    {
+        deserializeImpl(column, istr, &IDataType::deserializeBinary);
+    }
+
+    void serializeTextEscaped(const IColumn & column, size_t row_num, WriteBuffer & ostr, const FormatSettings & settings) const override
+    {
+        serializeImpl(column, row_num, ostr, &IDataType::serializeTextEscaped, settings);
+    }
+
+    void deserializeTextEscaped(IColumn & column, ReadBuffer & istr, const FormatSettings & settings) const override
+    {
+        deserializeImpl(column, istr, &IDataType::deserializeTextEscaped, settings);
+    }
+
+    void serializeTextQuoted(const IColumn & column, size_t row_num, WriteBuffer & ostr, const FormatSettings & settings) const override
+    {
+        serializeImpl(column, row_num, ostr, &IDataType::serializeTextQuoted, settings);
+    }
+
+    void deserializeTextQuoted(IColumn & column, ReadBuffer & istr, const FormatSettings & settings) const override
+    {
+        deserializeImpl(column, istr, &IDataType::deserializeTextQuoted, settings);
+    }
+
+    void serializeTextCSV(const IColumn & column, size_t row_num, WriteBuffer & ostr, const FormatSettings & settings) const override
+    {
+        serializeImpl(column, row_num, ostr, &IDataType::serializeTextCSV, settings);
+    }
+
+    void deserializeTextCSV(IColumn & column, ReadBuffer & istr, const FormatSettings & settings) const override
+    {
+        deserializeImpl(column, istr, &IDataType::deserializeTextCSV, settings);
+    }
+
+    void serializeText(const IColumn & column, size_t row_num, WriteBuffer & ostr, const FormatSettings & settings) const override
+    {
+        serializeImpl(column, row_num, ostr, &IDataType::serializeText, settings);
+    }
+
+    void serializeTextJSON(const IColumn & column, size_t row_num, WriteBuffer & ostr, const FormatSettings &  settings) const override
+    {
+        serializeImpl(column, row_num, ostr, &IDataType::serializeTextJSON, settings);
+    }
+    void deserializeTextJSON(IColumn & column, ReadBuffer & istr, const FormatSettings & settings) const override
+    {
+        deserializeImpl(column, istr, &IDataType::deserializeTextJSON, settings);
+    }
+
+    void serializeTextXML(const IColumn & column, size_t row_num, WriteBuffer & ostr, const FormatSettings & settings) const override
+    {
+        serializeImpl(column, row_num, ostr, &IDataType::serializeTextXML, settings);
+    }
+
+    MutableColumnPtr createColumn() const override;
+
+    Field getDefault() const override { return dictionary_type->getDefault(); }
+
+    bool equals(const IDataType & rhs) const override;
+
+    bool isParametric() const override { return true; }
+    bool haveSubtypes() const override { return true; }
+    bool cannotBeStoredInTables() const override { return dictionary_type->cannotBeStoredInTables(); }
+    bool shouldAlignRightInPrettyFormats() const override { return dictionary_type->shouldAlignRightInPrettyFormats(); }
+    bool textCanContainOnlyValidUTF8() const override { return dictionary_type->textCanContainOnlyValidUTF8(); }
+    bool isComparable() const override { return dictionary_type->isComparable(); }
+    bool canBeComparedWithCollation() const override { return dictionary_type->canBeComparedWithCollation(); }
+    bool canBeUsedAsVersion() const override { return dictionary_type->canBeUsedAsVersion(); }
+    bool isSummable() const override { return dictionary_type->isSummable(); };
+    bool canBeUsedInBitOperations() const override { return dictionary_type->canBeUsedInBitOperations(); };
+    bool canBeUsedInBooleanContext() const override { return dictionary_type->canBeUsedInBooleanContext(); };
+    bool isNumber() const override { return false; }
+    bool isInteger() const override { return false; }
+    bool isUnsignedInteger() const override { return false; }
+    bool isDateOrDateTime() const override { return false; }
+    bool isValueRepresentedByNumber() const override { return dictionary_type->isValueRepresentedByNumber(); }
+    bool isValueRepresentedByInteger() const override { return dictionary_type->isValueRepresentedByInteger(); }
+    bool isValueUnambiguouslyRepresentedInContiguousMemoryRegion() const override { return true; }
+    bool isString() const override { return false; }
+    bool isFixedString() const override { return false; }
+    bool haveMaximumSizeOfValue() const override { return dictionary_type->haveMaximumSizeOfValue(); }
+    size_t getMaximumSizeOfValueInMemory() const override { return dictionary_type->getMaximumSizeOfValueInMemory(); }
+    size_t getSizeOfValueInMemory() const override { return dictionary_type->getSizeOfValueInMemory(); }
+    bool isCategorial() const override { return false; }
+    bool isEnum() const override { return false; }
+    bool isNullable() const override { return false; }
+    bool onlyNull() const override { return false; }
+    bool withDictionary() const override { return true; }
+
+    static MutableColumnUniquePtr createColumnUnique(const IDataType & keys_type);
+    static MutableColumnUniquePtr createColumnUnique(const IDataType & keys_type, MutableColumnPtr && keys);
+
+private:
+
+    template <typename ... Args>
+    using SerealizeFunctionPtr = void (IDataType::*)(const IColumn &, size_t, WriteBuffer &, Args & ...) const;
+
+    template <typename ... Args>
+    void serializeImpl(const IColumn & column, size_t row_num, WriteBuffer & ostr,
+                       SerealizeFunctionPtr<Args ...> func, Args & ... args) const;
+
+    template <typename ... Args>
+    using DeserealizeFunctionPtr = void (IDataType::*)(IColumn &, ReadBuffer &, Args & ...) const;
+
+    template <typename ... Args>
+    void deserializeImpl(IColumn & column, ReadBuffer & istr,
+                         DeserealizeFunctionPtr<Args ...> func, Args & ... args) const;
+
+    template <typename Creator>
+    static MutableColumnUniquePtr createColumnUniqueImpl(const IDataType & keys_type, const Creator & creator);
+};
+
+}

dbms/src/DataTypes/IDataType.cpp

@@ -97,6 +97,8 @@ String IDataType::getFileNameForStream(const String & column_name, const IDataTy
             ///  and name is encoded as a whole.
             stream_name += "%2E" + escapeForFileName(elem.tuple_element_name);
         }
+        else if (elem.type == Substream::DictionaryKeys)
+            stream_name += ".dict";
     }
     return stream_name;
 }

dbms/src/DataTypes/IDataType.h

@@ -79,6 +79,9 @@ public:
             NullMap,
 
             TupleElement,
+
+            DictionaryKeys,
+            DictionaryIndexes,
         };
         Type type;
 
@@ -91,14 +94,64 @@ public:
     using SubstreamPath = std::vector<Substream>;
 
     using StreamCallback = std::function<void(const SubstreamPath &)>;
-    virtual void enumerateStreams(StreamCallback callback, SubstreamPath path) const
+    virtual void enumerateStreams(const StreamCallback & callback, SubstreamPath & path) const
     {
         callback(path);
     }
+    void enumerateStreams(const StreamCallback & callback, SubstreamPath && path) const { enumerateStreams(callback, path); }
+    void enumerateStreams(const StreamCallback & callback) const { enumerateStreams(callback, {}); }
 
     using OutputStreamGetter = std::function<WriteBuffer*(const SubstreamPath &)>;
     using InputStreamGetter = std::function<ReadBuffer*(const SubstreamPath &)>;
 
+    struct SerializeBinaryBulkState
+    {
+        virtual ~SerializeBinaryBulkState() = default;
+    };
+    struct DeserializeBinaryBulkState
+    {
+        virtual ~DeserializeBinaryBulkState() = default;
+    };
+
+    using SerializeBinaryBulkStatePtr = std::shared_ptr<SerializeBinaryBulkState>;
+    using DeserializeBinaryBulkStatePtr = std::shared_ptr<DeserializeBinaryBulkState>;
+
+    struct SerializeBinaryBulkSettings
+    {
+        OutputStreamGetter getter;
+        SubstreamPath path;
+
+        size_t low_cardinality_max_dictionary_size = 0;
+        bool low_cardinality_use_single_dictionary_for_part = true;
+
+        bool position_independent_encoding = true;
+    };
+
+    struct DeserializeBinaryBulkSettings
+    {
+        InputStreamGetter getter;
+        SubstreamPath path;
+
+        bool position_independent_encoding = true;
+        /// If not zero, may be used to avoid reallocations while reading column of String type.
+        double avg_value_size_hint = 0;
+    };
+
+    /// Call before serializeBinaryBulkWithMultipleStreams chain to write something before first mark.
+    virtual void serializeBinaryBulkStatePrefix(
+            SerializeBinaryBulkSettings & /*settings*/,
+            SerializeBinaryBulkStatePtr & /*state*/) const {}
+
+    /// Call after serializeBinaryBulkWithMultipleStreams chain to finish serialization.
+    virtual void serializeBinaryBulkStateSuffix(
+        SerializeBinaryBulkSettings & /*settings*/,
+        SerializeBinaryBulkStatePtr & /*state*/) const {}
+
+    /// Call before before deserializeBinaryBulkWithMultipleStreams chain to get DeserializeBinaryBulkStatePtr.
+    virtual void deserializeBinaryBulkStatePrefix(
+        DeserializeBinaryBulkSettings & /*settings*/,
+        DeserializeBinaryBulkStatePtr & /*state*/) const {}
+
     /** 'offset' and 'limit' are used to specify range.
       * limit = 0 - means no limit.
       * offset must be not greater than size of column.
@@ -107,29 +160,24 @@ public:
       */
     virtual void serializeBinaryBulkWithMultipleStreams(
         const IColumn & column,
-        OutputStreamGetter getter,
         size_t offset,
         size_t limit,
-        bool /*position_independent_encoding*/,
-        SubstreamPath path) const
+        SerializeBinaryBulkSettings & settings,
+        SerializeBinaryBulkStatePtr & /*state*/) const
     {
-        if (WriteBuffer * stream = getter(path))
+        if (WriteBuffer * stream = settings.getter(settings.path))
             serializeBinaryBulk(column, *stream, offset, limit);
     }
 
-    /** Read no more than limit values and append them into column.
-      * avg_value_size_hint - if not zero, may be used to avoid reallocations while reading column of String type.
-      */
+    /// Read no more than limit values and append them into column.
     virtual void deserializeBinaryBulkWithMultipleStreams(
         IColumn & column,
-        InputStreamGetter getter,
         size_t limit,
-        double avg_value_size_hint,
-        bool /*position_independent_encoding*/,
-        SubstreamPath path) const
+        DeserializeBinaryBulkSettings & settings,
+        DeserializeBinaryBulkStatePtr & /*state*/) const
     {
-        if (ReadBuffer * stream = getter(path))
-            deserializeBinaryBulk(column, *stream, limit, avg_value_size_hint);
+        if (ReadBuffer * stream = settings.getter(settings.path))
+            deserializeBinaryBulk(column, *stream, limit, settings.avg_value_size_hint);
     }
 
     /** Override these methods for data types that require just single stream (most of data types).
@@ -359,6 +407,8 @@ public:
       */
     virtual bool canBeInsideNullable() const { return false; }
 
+    virtual bool withDictionary() const { return false; }
+
 
     /// Updates avg_value_size_hint for newly read column. Uses to optimize deserialization. Zero expected for first column.
     static void updateAvgValueSizeHint(const IColumn & column, double & avg_value_size_hint);

dbms/src/DataTypes/tests/data_type_string.cpp

@@ -22,6 +22,7 @@ try
     size_t size = strlen(s) + 1;
     DataTypeString data_type;
 
+
     {
         auto column = ColumnString::create();
         ColumnString::Chars_t & data = column->getChars();
@@ -37,8 +38,14 @@ try
 
         WriteBufferFromFile out_buf("test");
 
+        IDataType::SerializeBinaryBulkSettings settings;
+        IDataType::SerializeBinaryBulkStatePtr state;
+        settings.getter = [&](const IDataType::SubstreamPath &){ return &out_buf; };
+
         stopwatch.restart();
-        data_type.serializeBinaryBulkWithMultipleStreams(*column, [&](const IDataType::SubstreamPath &){ return &out_buf; }, 0, 0, true, {});
+        data_type.serializeBinaryBulkStatePrefix(settings, state);
+        data_type.serializeBinaryBulkWithMultipleStreams(*column, 0, 0, settings, state);
+        data_type.serializeBinaryBulkStateSuffix(settings, state);
         stopwatch.stop();
 
         std::cout << "Writing, elapsed: " << stopwatch.elapsedSeconds() << std::endl;
@@ -49,8 +56,13 @@ try
 
         ReadBufferFromFile in_buf("test");
 
+        IDataType::DeserializeBinaryBulkSettings settings;
+        IDataType::DeserializeBinaryBulkStatePtr state;
+        settings.getter = [&](const IDataType::SubstreamPath &){ return &in_buf; };
+
         stopwatch.restart();
-        data_type.deserializeBinaryBulkWithMultipleStreams(*column, [&](const IDataType::SubstreamPath &){ return &in_buf; }, n, 0, true, {});
+        data_type.deserializeBinaryBulkStatePrefix(settings, state);
+        data_type.deserializeBinaryBulkWithMultipleStreams(*column, n, settings, state);
         stopwatch.stop();
 
         std::cout << "Reading, elapsed: " << stopwatch.elapsedSeconds() << std::endl;

dbms/src/DataTypes/tests/data_types_number_fixed.cpp

@@ -27,7 +27,12 @@ int main(int, char **)
     WriteBufferFromOStream out_buf(ostr);
 
     stopwatch.restart();
-    data_type.serializeBinaryBulkWithMultipleStreams(*column, [&](const IDataType::SubstreamPath &){ return &out_buf; }, 0, 0, true, {});
+    IDataType::SerializeBinaryBulkSettings settings;
+    settings.getter = [&](const IDataType::SubstreamPath &){ return &out_buf; };
+    IDataType::SerializeBinaryBulkStatePtr state;
+    data_type.serializeBinaryBulkStatePrefix(settings, state);
+    data_type.serializeBinaryBulkWithMultipleStreams(*column, 0, 0, settings, state);
+    data_type.serializeBinaryBulkStateSuffix(settings, state);
     stopwatch.stop();
 
     std::cout << "Elapsed: " << stopwatch.elapsedSeconds() << std::endl;

dbms/src/Functions/FunctionsArithmetic.h

@@ -749,7 +749,7 @@ static bool castTypeToEither(const IDataType * type, F && f)
 }
 
 
-template <template <typename, typename> class Op, typename Name>
+template <template <typename, typename> class Op, typename Name, bool CanBeExecutedOnDefaultArguments = true>
 class FunctionBinaryArithmetic : public IFunction
 {
     const Context & context;
@@ -993,6 +993,8 @@ public:
         return result;
     }
 #endif
+
+    bool canBeExecutedOnDefaultArguments() const override { return CanBeExecutedOnDefaultArguments; }
 };
 
 
@@ -1141,9 +1143,9 @@ using FunctionPlus = FunctionBinaryArithmetic<PlusImpl, NamePlus>;
 using FunctionMinus = FunctionBinaryArithmetic<MinusImpl, NameMinus>;
 using FunctionMultiply = FunctionBinaryArithmetic<MultiplyImpl, NameMultiply>;
 using FunctionDivideFloating = FunctionBinaryArithmetic<DivideFloatingImpl, NameDivideFloating>;
-using FunctionDivideIntegral = FunctionBinaryArithmetic<DivideIntegralImpl, NameDivideIntegral>;
+using FunctionDivideIntegral = FunctionBinaryArithmetic<DivideIntegralImpl, NameDivideIntegral, false>;
 using FunctionDivideIntegralOrZero = FunctionBinaryArithmetic<DivideIntegralOrZeroImpl, NameDivideIntegralOrZero>;
-using FunctionModulo = FunctionBinaryArithmetic<ModuloImpl, NameModulo>;
+using FunctionModulo = FunctionBinaryArithmetic<ModuloImpl, NameModulo, false>;
 using FunctionNegate = FunctionUnaryArithmetic<NegateImpl, NameNegate, true>;
 using FunctionAbs = FunctionUnaryArithmetic<AbsImpl, NameAbs, false>;
 using FunctionBitAnd = FunctionBinaryArithmetic<BitAndImpl, NameBitAnd>;
@@ -1157,8 +1159,8 @@ using FunctionBitRotateRight = FunctionBinaryArithmetic<BitRotateRightImpl, Name
 using FunctionBitTest = FunctionBinaryArithmetic<BitTestImpl, NameBitTest>;
 using FunctionLeast = FunctionBinaryArithmetic<LeastImpl, NameLeast>;
 using FunctionGreatest = FunctionBinaryArithmetic<GreatestImpl, NameGreatest>;
-using FunctionGCD = FunctionBinaryArithmetic<GCDImpl, NameGCD>;
-using FunctionLCM = FunctionBinaryArithmetic<LCMImpl, NameLCM>;
+using FunctionGCD = FunctionBinaryArithmetic<GCDImpl, NameGCD, false>;
+using FunctionLCM = FunctionBinaryArithmetic<LCMImpl, NameLCM, false>;
 /// Assumed to be injective for the purpose of query optimization, but in fact it is not injective because of possible overflow.
 using FunctionIntExp2 = FunctionUnaryArithmetic<IntExp2Impl, NameIntExp2, true>;
 using FunctionIntExp10 = FunctionUnaryArithmetic<IntExp10Impl, NameIntExp10, true>;

dbms/src/Functions/FunctionsArray.cpp

@@ -1339,7 +1339,7 @@ void FunctionArrayDistinct::executeImpl(Block & block, const ColumnNumbers & arg
         || executeNumber<Float32>(*inner_col, offsets, res_data, res_offsets, nullable_col)
         || executeNumber<Float64>(*inner_col, offsets, res_data, res_offsets, nullable_col)
         || executeString(*inner_col, offsets, res_data, res_offsets, nullable_col)))
-        executeHashed(offsets, original_data_columns, res_data, res_offsets);
+        executeHashed(offsets, original_data_columns, res_data, res_offsets, nullable_col);
 
     block.getByPosition(result).column = std::move(res_ptr);
 }
@@ -1449,13 +1449,21 @@ void FunctionArrayDistinct::executeHashed(
     const ColumnArray::Offsets & offsets,
     const ColumnRawPtrs & columns,
     IColumn & res_data_col,
-    ColumnArray::Offsets & res_offsets)
+    ColumnArray::Offsets & res_offsets,
+    const ColumnNullable * nullable_col)
 {
     size_t count = columns.size();
 
     using Set = ClearableHashSet<UInt128, UInt128TrivialHash, HashTableGrower<INITIAL_SIZE_DEGREE>,
         HashTableAllocatorWithStackMemory<(1ULL << INITIAL_SIZE_DEGREE) * sizeof(UInt128)>>;
 
+    const PaddedPODArray<UInt8> * src_null_map = nullptr;
+
+    if (nullable_col)
+    {
+        src_null_map = &static_cast<const ColumnUInt8 *>(&nullable_col->getNullMapColumn())->getData();
+    }
+
     Set set;
     size_t prev_off = 0;
     for (size_t i = 0; i < offsets.size(); ++i)
@@ -1465,7 +1473,7 @@ void FunctionArrayDistinct::executeHashed(
         for (size_t j = prev_off; j < off; ++j)
         {
             auto hash = hash128(j, count, columns);
-            if (set.find(hash) == set.end())
+            if (set.find(hash) == set.end() && (!nullable_col || (*src_null_map)[j] == 0))
             {
                 set.insert(hash);
                 res_data_col.insertFrom(*columns[0], j);

dbms/src/Functions/FunctionsArray.h

@@ -1255,7 +1255,8 @@ private:
         const ColumnArray::Offsets & offsets,
         const ColumnRawPtrs & columns,
         IColumn & res_data_col,
-        ColumnArray::Offsets & res_offsets);
+        ColumnArray::Offsets & res_offsets,
+        const ColumnNullable * nullable_col);
 };
 
 

dbms/src/Functions/FunctionsCoding.h

@@ -1111,7 +1111,7 @@ public:
         return std::make_shared<DataTypeUUID>();
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers &, size_t result, size_t input_rows_count) override
     {

dbms/src/Functions/FunctionsConversion.h

@@ -36,6 +36,8 @@
 #include <Functions/FunctionsMiscellaneous.h>
 #include <Functions/FunctionsDateTime.h>
 #include <Functions/FunctionHelpers.h>
+#include <DataTypes/DataTypeWithDictionary.h>
+#include <Columns/ColumnWithDictionary.h>
 
 
 namespace DB
@@ -728,6 +730,7 @@ public:
 
     bool useDefaultImplementationForConstants() const override { return true; }
     ColumnNumbers getArgumentsThatAreAlwaysConstant() const override { return {1}; }
+    bool canBeExecutedOnDefaultArguments() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t input_rows_count) override
     {
@@ -1245,6 +1248,7 @@ protected:
 
     bool useDefaultImplementationForNulls() const override { return false; }
     bool useDefaultImplementationForConstants() const override { return true; }
+    bool useDefaultImplementationForColumnsWithDictionary() const override { return false; }
     ColumnNumbers getArgumentsThatAreAlwaysConstant() const override { return {1}; }
 
 private:
@@ -1273,7 +1277,8 @@ public:
 
     PreparedFunctionPtr prepare(const Block & /*sample_block*/) const override
     {
-        return std::make_shared<PreparedFunctionCast>(prepare(getArgumentTypes()[0], getReturnType()), name);
+        return std::make_shared<PreparedFunctionCast>(
+                prepareUnpackDictionaries(getArgumentTypes()[0], getReturnType()), name);
     }
 
     String getName() const override { return name; }
@@ -1378,7 +1383,7 @@ private:
             throw Exception{"CAST AS Array can only be performed between same-dimensional array types or from String", ErrorCodes::TYPE_MISMATCH};
 
         /// Prepare nested type conversion
-        const auto nested_function = prepare(from_nested_type, to_nested_type);
+        const auto nested_function = prepareUnpackDictionaries(from_nested_type, to_nested_type);
 
         return [nested_function, from_nested_type, to_nested_type](
             Block & block, const ColumnNumbers & arguments, const size_t result, size_t /*input_rows_count*/)
@@ -1432,7 +1437,7 @@ private:
 
         /// Create conversion wrapper for each element in tuple
         for (const auto & idx_type : ext::enumerate(from_type->getElements()))
-            element_wrappers.push_back(prepare(idx_type.second, to_element_types[idx_type.first]));
+            element_wrappers.push_back(prepareUnpackDictionaries(idx_type.second, to_element_types[idx_type.first]));
 
         return [element_wrappers, from_element_types, to_element_types]
             (Block & block, const ColumnNumbers & arguments, const size_t result, size_t input_rows_count)
@@ -1576,16 +1581,11 @@ private:
         };
     }
 
-    WrapperType prepare(const DataTypePtr & from_type, const DataTypePtr & to_type) const
+    WrapperType prepareUnpackDictionaries(const DataTypePtr & from_type, const DataTypePtr & to_type) const
     {
-        /// Determine whether pre-processing and/or post-processing must take place during conversion.
-
-        bool source_is_nullable = from_type->isNullable();
-        bool result_is_nullable = to_type->isNullable();
-
         if (from_type->onlyNull())
         {
-            if (!result_is_nullable)
+            if (!to_type->isNullable())
                 throw Exception{"Cannot convert NULL to a non-nullable type", ErrorCodes::CANNOT_CONVERT_TYPE};
 
             return [](Block & block, const ColumnNumbers &, const size_t result, size_t input_rows_count)
@@ -1595,6 +1595,88 @@ private:
             };
         }
 
+        const auto * from_with_dict = typeid_cast<const DataTypeWithDictionary *>(from_type.get());
+        const auto * to_with_dict = typeid_cast<const DataTypeWithDictionary *>(to_type.get());
+        const auto & from_nested = from_with_dict ? from_with_dict->getDictionaryType() : from_type;
+        const auto & to_nested = to_with_dict ? to_with_dict->getDictionaryType() : to_type;
+
+        auto wrapper = prepareRemoveNullable(from_nested, to_nested);
+        if (!from_with_dict && !to_with_dict)
+            return wrapper;
+
+        return [wrapper, from_with_dict, to_with_dict]
+                (Block & block, const ColumnNumbers & arguments, const size_t result, size_t input_rows_count)
+        {
+            auto & arg = block.getByPosition(arguments[0]);
+            auto & res = block.getByPosition(result);
+
+            ColumnPtr res_indexes;
+            /// For some types default can't be casted (for example, String to Int). In that case convert column to full.
+            bool src_converted_to_full_column = false;
+
+            {
+                /// Replace argument and result columns (and types) to dictionary key columns (and types).
+                /// Call nested wrapper in order to cast dictionary keys. Then restore block.
+                auto prev_arg_col = arg.column;
+                auto prev_arg_type = arg.type;
+                auto prev_res_type = res.type;
+
+                auto tmp_rows_count = input_rows_count;
+
+                if (to_with_dict)
+                    res.type = to_with_dict->getDictionaryType();
+
+                if (from_with_dict)
+                {
+                    auto * col_with_dict = typeid_cast<const ColumnWithDictionary *>(prev_arg_col.get());
+                    arg.column = col_with_dict->getDictionary().getNestedColumn();
+                    arg.type = from_with_dict->getDictionaryType();
+
+                    /// TODO: Make map with defaults conversion.
+                    src_converted_to_full_column = !removeNullable(arg.type)->equals(*removeNullable(res.type));
+                    if (src_converted_to_full_column)
+                        arg.column = arg.column->index(col_with_dict->getIndexes(), 0);
+                    else
+                        res_indexes = col_with_dict->getIndexesPtr();
+
+                    tmp_rows_count = arg.column->size();
+                }
+
+                /// Perform the requested conversion.
+                wrapper(block, arguments, result, tmp_rows_count);
+
+                arg.column = prev_arg_col;
+                arg.type = prev_arg_type;
+                res.type = prev_res_type;
+            }
+
+            if (to_with_dict)
+            {
+                auto res_column = to_with_dict->createColumn();
+                auto * col_with_dict = typeid_cast<ColumnWithDictionary *>(res_column.get());
+
+                if (from_with_dict && !src_converted_to_full_column)
+                {
+                    auto res_keys = std::move(res.column);
+                    col_with_dict->insertRangeFromDictionaryEncodedColumn(*res_keys, *res_indexes);
+                }
+                else
+                    col_with_dict->insertRangeFromFullColumn(*res.column, 0, res.column->size());
+
+                res.column = std::move(res_column);
+            }
+            else if (!src_converted_to_full_column)
+                res.column = res.column->index(*res_indexes, 0);
+        };
+    }
+
+    WrapperType prepareRemoveNullable(const DataTypePtr & from_type, const DataTypePtr & to_type) const
+    {
+        /// Determine whether pre-processing and/or post-processing must take place during conversion.
+
+        bool source_is_nullable = from_type->isNullable();
+        bool result_is_nullable = to_type->isNullable();
+
         auto wrapper = prepareImpl(removeNullable(from_type), removeNullable(to_type), result_is_nullable);
 
         if (result_is_nullable)
@@ -1740,6 +1822,7 @@ protected:
     }
 
     bool useDefaultImplementationForNulls() const override { return false; }
+    bool useDefaultImplementationForColumnsWithDictionary() const override { return false; }
 
 private:
     template <typename DataType>

dbms/src/Functions/FunctionsDateTime.h

@@ -1206,7 +1206,7 @@ public:
         return std::make_shared<DataTypeDateTime>();
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers &, size_t result, size_t input_rows_count) override
     {
@@ -1235,7 +1235,7 @@ public:
         return std::make_shared<DataTypeDate>();
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers &, size_t result, size_t input_rows_count) override
     {
@@ -1264,7 +1264,7 @@ public:
         return std::make_shared<DataTypeDate>();
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers &, size_t result, size_t input_rows_count) override
     {

dbms/src/Functions/FunctionsEmbeddedDictionaries.h

@@ -220,7 +220,7 @@ public:
     bool useDefaultImplementationForConstants() const override { return true; }
     ColumnNumbers getArgumentsThatAreAlwaysConstant() const override { return {1}; }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t /*input_rows_count*/) override
     {
@@ -316,7 +316,7 @@ public:
         return std::make_shared<DataTypeUInt8>();
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t /*input_rows_count*/) override
     {
@@ -452,7 +452,7 @@ public:
     bool useDefaultImplementationForConstants() const override { return true; }
     ColumnNumbers getArgumentsThatAreAlwaysConstant() const override { return {1}; }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t /*input_rows_count*/) override
     {
@@ -728,7 +728,7 @@ public:
     bool useDefaultImplementationForConstants() const override { return true; }
     ColumnNumbers getArgumentsThatAreAlwaysConstant() const override { return {1}; }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t /*input_rows_count*/) override
     {

dbms/src/Functions/FunctionsExternalDictionaries.h

@@ -94,7 +94,7 @@ private:
         return std::make_shared<DataTypeUInt8>();
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t input_rows_count) override
     {
@@ -259,7 +259,7 @@ private:
         return std::make_shared<DataTypeString>();
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t input_rows_count) override
     {
@@ -489,7 +489,7 @@ private:
         return std::make_shared<DataTypeString>();
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t input_rows_count) override
     {
@@ -755,7 +755,7 @@ private:
         return std::make_shared<DataType>();
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t input_rows_count) override
     {
@@ -1030,7 +1030,7 @@ private:
         return std::make_shared<DataType>();
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t input_rows_count) override
     {
@@ -1263,7 +1263,7 @@ private:
         return std::make_shared<DataTypeArray>(std::make_shared<DataTypeUInt64>());
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t input_rows_count) override
     {
@@ -1423,7 +1423,7 @@ private:
         return std::make_shared<DataTypeUInt8>();
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t input_rows_count) override
     {

dbms/src/Functions/FunctionsExternalModels.h

@@ -23,7 +23,7 @@ public:
 
     bool isVariadic() const override { return true; }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     size_t getNumberOfArguments() const override { return 0; }
 

dbms/src/Functions/FunctionsMiscellaneous.cpp

@@ -11,6 +11,7 @@
 #include <Columns/ColumnString.h>
 #include <Columns/ColumnTuple.h>
 #include <Columns/ColumnArray.h>
+#include <Columns/ColumnWithDictionary.h>
 #include <Functions/FunctionHelpers.h>
 #include <Common/UnicodeBar.h>
 #include <Common/UTF8Helpers.h>
@@ -24,6 +25,7 @@
 #include <DataTypes/DataTypeTuple.h>
 #include <DataTypes/DataTypesNumber.h>
 #include <DataTypes/DataTypeEnum.h>
+#include <DataTypes/DataTypeWithDictionary.h>
 #include <DataTypes/NumberTraits.h>
 #include <Formats/FormatSettings.h>
 #include <Functions/FunctionFactory.h>
@@ -111,7 +113,7 @@ public:
         return std::make_shared<DataTypeString>();
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers &, size_t result, size_t input_rows_count) override
     {
@@ -135,9 +137,9 @@ public:
         return name;
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
-    bool isDeterministicInScopeOfQuery() override
+    bool isDeterministicInScopeOfQuery() const override
     {
         return false;
     }
@@ -213,6 +215,7 @@ public:
     }
 
     bool useDefaultImplementationForNulls() const override { return false; }
+    bool useDefaultImplementationForColumnsWithDictionary() const override { return false; }
 
     size_t getNumberOfArguments() const override
     {
@@ -404,9 +407,9 @@ public:
         return name;
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
-    bool isDeterministicInScopeOfQuery() override
+    bool isDeterministicInScopeOfQuery() const override
     {
         return false;
     }
@@ -448,9 +451,9 @@ public:
         return 0;
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
-    bool isDeterministicInScopeOfQuery() override
+    bool isDeterministicInScopeOfQuery() const override
     {
         return false;
     }
@@ -497,9 +500,9 @@ public:
         return 0;
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
-    bool isDeterministicInScopeOfQuery() override
+    bool isDeterministicInScopeOfQuery() const override
     {
         return false;
     }
@@ -541,9 +544,9 @@ public:
         return 0;
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
-    bool isDeterministicInScopeOfQuery() override
+    bool isDeterministicInScopeOfQuery() const override
     {
         return false;
     }
@@ -914,9 +917,9 @@ public:
     }
 
     /** It could return many different values for single argument. */
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
-    bool isDeterministicInScopeOfQuery() override
+    bool isDeterministicInScopeOfQuery() const override
     {
         return false;
     }
@@ -1317,7 +1320,7 @@ public:
         return std::make_shared<DataTypeUInt32>();
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers &, size_t result, size_t input_rows_count) override
     {
@@ -1354,7 +1357,7 @@ public:
         return std::make_shared<DataTypeString>();
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers &, size_t result, size_t input_rows_count) override
     {
@@ -1388,9 +1391,9 @@ public:
         return 1;
     }
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
-    bool isDeterministicInScopeOfQuery() override
+    bool isDeterministicInScopeOfQuery() const override
     {
         return false;
     }
@@ -1557,7 +1560,7 @@ public:
         return 1;
     }
 
-    bool isDeterministicInScopeOfQuery() override
+    bool isDeterministicInScopeOfQuery() const override
     {
         return false;
     }
@@ -1698,7 +1701,7 @@ public:
 
     DataTypePtr getReturnTypeImpl(const ColumnsWithTypeAndName & arguments) const override;
 
-    bool isDeterministic() override { return false; }
+    bool isDeterministic() const override { return false; }
 
     void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t input_rows_count) override;
 
@@ -1866,6 +1869,119 @@ public:
 };
 
 
+class FunctionToLowCardinality: public IFunction
+{
+public:
+    static constexpr auto name = "toLowCardinality";
+    static FunctionPtr create(const Context &) { return std::make_shared<FunctionToLowCardinality>(); }
+
+    String getName() const override { return name; }
+
+    size_t getNumberOfArguments() const override { return 1; }
+
+    bool useDefaultImplementationForNulls() const override { return false; }
+    bool useDefaultImplementationForConstants() const override { return true; }
+    bool useDefaultImplementationForColumnsWithDictionary() const override { return false; }
+
+    DataTypePtr getReturnTypeImpl(const DataTypes & arguments) const override
+    {
+        if (arguments[0]->withDictionary())
+            return arguments[0];
+
+        return std::make_shared<DataTypeWithDictionary>(arguments[0]);
+    }
+
+    void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t /*input_rows_count*/) override
+    {
+        auto arg_num = arguments[0];
+        const auto & arg = block.getByPosition(arg_num);
+        auto & res = block.getByPosition(result);
+
+        if (arg.type->withDictionary())
+            res.column = arg.column;
+        else
+        {
+            auto column = res.type->createColumn();
+            typeid_cast<ColumnWithDictionary &>(*column).insertRangeFromFullColumn(*arg.column, 0, arg.column->size());
+            res.column = std::move(column);
+        }
+    }
+};
+
+class FunctionLowCardinalityIndexes: public IFunction
+{
+public:
+    static constexpr auto name = "lowCardinalityIndexes";
+    static FunctionPtr create(const Context &) { return std::make_shared<FunctionLowCardinalityIndexes>(); }
+
+    String getName() const override { return name; }
+
+    size_t getNumberOfArguments() const override { return 1; }
+
+    bool useDefaultImplementationForNulls() const override { return false; }
+    bool useDefaultImplementationForConstants() const override { return true; }
+    bool useDefaultImplementationForColumnsWithDictionary() const override { return false; }
+
+    DataTypePtr getReturnTypeImpl(const DataTypes & arguments) const override
+    {
+        auto * type = typeid_cast<const DataTypeWithDictionary *>(arguments[0].get());
+        if (!type)
+            throw Exception("First first argument of function lowCardinalityIndexes must be ColumnWithDictionary, but got"
+                            + arguments[0]->getName(), ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
+
+        return std::make_shared<DataTypeUInt64>();
+    }
+
+    void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t /*input_rows_count*/) override
+    {
+        auto arg_num = arguments[0];
+        const auto & arg = block.getByPosition(arg_num);
+        auto & res = block.getByPosition(result);
+        auto indexes_col = typeid_cast<const ColumnWithDictionary *>(arg.column.get())->getIndexesPtr();
+        auto new_indexes_col = ColumnUInt64::create(indexes_col->size());
+        auto & data = new_indexes_col->getData();
+        for (size_t i = 0; i < data.size(); ++i)
+            data[i] = indexes_col->getUInt(i);
+
+        res.column = std::move(new_indexes_col);
+    }
+};
+
+class FunctionLowCardinalityKeys: public IFunction
+{
+public:
+    static constexpr auto name = "lowCardinalityKeys";
+    static FunctionPtr create(const Context &) { return std::make_shared<FunctionLowCardinalityKeys>(); }
+
+    String getName() const override { return name; }
+
+    size_t getNumberOfArguments() const override { return 1; }
+
+    bool useDefaultImplementationForNulls() const override { return false; }
+    bool useDefaultImplementationForConstants() const override { return true; }
+    bool useDefaultImplementationForColumnsWithDictionary() const override { return false; }
+
+    DataTypePtr getReturnTypeImpl(const DataTypes & arguments) const override
+    {
+        auto * type = typeid_cast<const DataTypeWithDictionary *>(arguments[0].get());
+        if (!type)
+            throw Exception("First first argument of function lowCardinalityKeys must be ColumnWithDictionary, but got"
+                            + arguments[0]->getName(), ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
+
+        return type->getDictionaryType();
+    }
+
+    void executeImpl(Block & block, const ColumnNumbers & arguments, size_t result, size_t /*input_rows_count*/) override
+    {
+        auto arg_num = arguments[0];
+        const auto & arg = block.getByPosition(arg_num);
+        auto & res = block.getByPosition(result);
+        const auto * column_with_dictionary = typeid_cast<const ColumnWithDictionary *>(arg.column.get());
+        res.column = column_with_dictionary->getDictionary().getNestedColumn()->cloneResized(arg.column->size());
+    }
+};
+
+
 std::string FunctionVersion::getVersion() const
 {
     return VERSION_STRING;
@@ -1915,5 +2031,9 @@ void registerFunctionsMiscellaneous(FunctionFactory & factory)
     factory.registerFunction<FunctionRunningDifference>();
     factory.registerFunction<FunctionRunningIncome>();
     factory.registerFunction<FunctionFinalizeAggregation>();
+
+    factory.registerFunction<FunctionToLowCardinality>();
+    factory.registerFunction<FunctionLowCardinalityIndexes>();
+    factory.registerFunction<FunctionLowCardinalityKeys>();
 }
 }

dbms/src/Functions/FunctionsMiscellaneous.h

@@ -6,6 +6,7 @@
 #include <IO/WriteBufferFromString.h>
 #include <IO/Operators.h>
 #include <Columns/ColumnFunction.h>
+#include <DataTypes/DataTypesNumber.h>
 
 namespace DB
 {

dbms/src/Functions/FunctionsRandom.h

@@ -138,7 +138,7 @@ public:
 
     bool isVariadic() const override { return true; }
     size_t getNumberOfArguments() const override { return 0; }
-    bool isDeterministicInScopeOfQuery() override { return false; }
+    bool isDeterministicInScopeOfQuery() const override { return false; }
 
     DataTypePtr getReturnTypeImpl(const DataTypes & arguments) const override
     {

dbms/src/Functions/IFunction.cpp

@@ -5,6 +5,9 @@
 #include <DataTypes/DataTypeNothing.h>
 #include <DataTypes/DataTypeNullable.h>
 #include <DataTypes/Native.h>
+#include <DataTypes/DataTypeWithDictionary.h>
+#include <DataTypes/getLeastSupertype.h>
+#include <Columns/ColumnWithDictionary.h>
 #include <Functions/FunctionHelpers.h>
 #include <Functions/IFunction.h>
 #include <Interpreters/ExpressionActions.h>
@@ -187,7 +190,7 @@ bool PreparedFunctionImpl::defaultImplementationForConstantArguments(Block & blo
     for (size_t i = 0; i < arguments_size; ++i)
         temporary_argument_numbers[i] = i;
 
-    execute(temporary_block, temporary_argument_numbers, arguments_size, temporary_block.rows());
+    executeWithoutColumnsWithDictionary(temporary_block, temporary_argument_numbers, arguments_size, temporary_block.rows());
 
     block.getByPosition(result).column = ColumnConst::create(temporary_block.getByPosition(arguments_size).column, input_rows_count);
     return true;
@@ -211,7 +214,7 @@ bool PreparedFunctionImpl::defaultImplementationForNulls(Block & block, const Co
     if (null_presence.has_nullable)
     {
         Block temporary_block = createBlockWithNestedColumns(block, args, result);
-        execute(temporary_block, args, result, temporary_block.rows());
+        executeWithoutColumnsWithDictionary(temporary_block, args, result, temporary_block.rows());
         block.getByPosition(result).column = wrapInNullable(temporary_block.getByPosition(result).column, block, args,
                                                             result, input_rows_count);
         return true;
@@ -220,7 +223,7 @@ bool PreparedFunctionImpl::defaultImplementationForNulls(Block & block, const Co
     return false;
 }
 
-void PreparedFunctionImpl::execute(Block & block, const ColumnNumbers & args, size_t result, size_t input_rows_count)
+void PreparedFunctionImpl::executeWithoutColumnsWithDictionary(Block & block, const ColumnNumbers & args, size_t result, size_t input_rows_count)
 {
     if (defaultImplementationForConstantArguments(block, args, result, input_rows_count))
         return;
@@ -231,6 +234,115 @@ void PreparedFunctionImpl::execute(Block & block, const ColumnNumbers & args, si
     executeImpl(block, args, result, input_rows_count);
 }
 
+static ColumnPtr replaceColumnsWithDictionaryByNestedAndGetDictionaryIndexes(Block & block, const ColumnNumbers & args, 
+                                                                             bool can_be_executed_on_default_arguments)
+{
+    size_t num_rows = 0;
+    ColumnPtr indexes;
+
+    for (auto arg : args)
+    {
+        ColumnWithTypeAndName & column = block.getByPosition(arg);
+        if (auto * column_with_dict = checkAndGetColumn<ColumnWithDictionary>(column.column.get()))
+        {
+            if (indexes)
+                throw Exception("Expected single dictionary argument for function.", ErrorCodes::LOGICAL_ERROR);
+
+            indexes = column_with_dict->getIndexesPtr();
+            num_rows = column_with_dict->getDictionary().size();
+        }
+    }
+
+    if (!indexes)
+        throw Exception("Expected column with dictionary for any function argument.", ErrorCodes::LOGICAL_ERROR);
+
+    for (auto arg : args)
+    {
+        ColumnWithTypeAndName & column = block.getByPosition(arg);
+        if (auto * column_const = checkAndGetColumn<ColumnConst>(column.column.get()))
+            column.column = column_const->cloneResized(num_rows);
+        else if (auto * column_with_dict = checkAndGetColumn<ColumnWithDictionary>(column.column.get()))
+        {
+            auto * type_with_dict = checkAndGetDataType<DataTypeWithDictionary>(column.type.get());
+
+            if (!type_with_dict)
+                throw Exception("Incompatible type for column with dictionary: " + column.type->getName(),
+                                ErrorCodes::LOGICAL_ERROR);
+
+            if (can_be_executed_on_default_arguments)
+                column.column = column_with_dict->getDictionary().getNestedColumn();
+            else
+            {
+                auto dict_encoded = column_with_dict->getMinimalDictionaryEncodedColumn(0, column_with_dict->size());
+                column.column = dict_encoded.dictionary;
+                indexes = dict_encoded.indexes;
+            }
+            column.type = type_with_dict->getDictionaryType();
+        }
+    }
+
+    return indexes;
+}
+
+static void convertColumnsWithDictionaryToFull(Block & block, const ColumnNumbers & args)
+{
+    for (auto arg : args)
+    {
+        ColumnWithTypeAndName & column = block.getByPosition(arg);
+        if (auto * column_with_dict = checkAndGetColumn<ColumnWithDictionary>(column.column.get()))
+        {
+            auto * type_with_dict = checkAndGetDataType<DataTypeWithDictionary>(column.type.get());
+
+            if (!type_with_dict)
+                throw Exception("Incompatible type for column with dictionary: " + column.type->getName(),
+                                ErrorCodes::LOGICAL_ERROR);
+
+            column.column = column_with_dict->convertToFullColumn();
+            column.type = type_with_dict->getDictionaryType();
+        }
+    }
+}
+
+void PreparedFunctionImpl::execute(Block & block, const ColumnNumbers & args, size_t result, size_t input_rows_count)
+{
+    if (useDefaultImplementationForColumnsWithDictionary())
+    {
+        auto & res = block.safeGetByPosition(result);
+        Block block_without_dicts = block.cloneWithoutColumns();
+
+        for (auto arg : args)
+            block_without_dicts.safeGetByPosition(arg).column = block.safeGetByPosition(arg).column;
+
+        if (auto * res_type_with_dict = typeid_cast<const DataTypeWithDictionary *>(res.type.get()))
+        {
+            block_without_dicts.safeGetByPosition(result).type = res_type_with_dict->getDictionaryType();
+            ColumnPtr indexes = replaceColumnsWithDictionaryByNestedAndGetDictionaryIndexes(
+                    block_without_dicts, args, canBeExecutedOnDefaultArguments());
+
+            executeWithoutColumnsWithDictionary(block_without_dicts, args, result, block_without_dicts.rows());
+
+            auto res_column = res.type->createColumn();
+            auto * column_with_dictionary = typeid_cast<ColumnWithDictionary *>(res_column.get());
+
+            if (!column_with_dictionary)
+                throw Exception("Expected ColumnWithDictionary, got" + res_column->getName(), ErrorCodes::LOGICAL_ERROR);
+
+            const auto & keys = block_without_dicts.safeGetByPosition(result).column;
+            column_with_dictionary->insertRangeFromDictionaryEncodedColumn(*keys, *indexes);
+
+            res.column = std::move(res_column);
+        }
+        else
+        {
+            convertColumnsWithDictionaryToFull(block_without_dicts, args);
+            executeWithoutColumnsWithDictionary(block_without_dicts, args, result, input_rows_count);
+            res.column = block_without_dicts.safeGetByPosition(result).column;
+        }
+    }
+    else
+        executeWithoutColumnsWithDictionary(block, args, result, input_rows_count);
+}
+
 void FunctionBuilderImpl::checkNumberOfArguments(size_t number_of_arguments) const
 {
     if (isVariadic())
@@ -244,19 +356,19 @@ void FunctionBuilderImpl::checkNumberOfArguments(size_t number_of_arguments) con
                         ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
 }
 
-DataTypePtr FunctionBuilderImpl::getReturnType(const ColumnsWithTypeAndName & arguments) const
+DataTypePtr FunctionBuilderImpl::getReturnTypeWithoutDictionary(const ColumnsWithTypeAndName & arguments) const
 {
     checkNumberOfArguments(arguments.size());
 
     if (!arguments.empty() && useDefaultImplementationForNulls())
     {
-        NullPresence null_presense = getNullPresense(arguments);
+        NullPresence null_presence = getNullPresense(arguments);
 
-        if (null_presense.has_null_constant)
+        if (null_presence.has_null_constant)
         {
             return makeNullable(std::make_shared<DataTypeNothing>());
         }
-        if (null_presense.has_nullable)
+        if (null_presence.has_nullable)
         {
             Block nested_block = createBlockWithNestedColumns(Block(arguments), ext::collection_cast<ColumnNumbers>(ext::range(0, arguments.size())));
             auto return_type = getReturnTypeImpl(ColumnsWithTypeAndName(nested_block.begin(), nested_block.end()));
@@ -334,4 +446,39 @@ llvm::Value * IFunction::compile(llvm::IRBuilderBase & builder, const DataTypes
 
 #endif
 
+
+DataTypePtr FunctionBuilderImpl::getReturnType(const ColumnsWithTypeAndName & arguments) const
+{
+    if (useDefaultImplementationForColumnsWithDictionary())
+    {
+        bool has_type_with_dictionary = false;
+        bool can_run_function_on_dictionary = true;
+
+        ColumnsWithTypeAndName args_without_dictionary(arguments);
+
+        for (ColumnWithTypeAndName & arg : args_without_dictionary)
+        {
+            if (arg.column && arg.column->isColumnConst())
+                continue;
+
+            if (auto * type_with_dictionary = typeid_cast<const DataTypeWithDictionary *>(arg.type.get()))
+            {
+                if (has_type_with_dictionary)
+                    can_run_function_on_dictionary = false;
+
+                has_type_with_dictionary = true;
+                arg.type = type_with_dictionary->getDictionaryType();
+            }
+            else
+                can_run_function_on_dictionary = false;
+        }
+
+        if (canBeExecutedOnLowCardinalityDictionary() && has_type_with_dictionary && can_run_function_on_dictionary)
+            return std::make_shared<DataTypeWithDictionary>(getReturnTypeWithoutDictionary(args_without_dictionary));
+        else
+            return getReturnTypeWithoutDictionary(args_without_dictionary);
+    }
+
+    return getReturnTypeWithoutDictionary(arguments);
+}
 }

dbms/src/Functions/IFunction.h

@@ -68,15 +68,28 @@ protected:
       */
     virtual bool useDefaultImplementationForConstants() const { return false; }
 
+    /** If function arguments has single column with dictionary and all other arguments are constants, call function on nested column.
+      * Otherwise, convert all columns with dictionary to ordinary columns.
+      * Returns ColumnWithDictionary if at least one argument is ColumnWithDictionary.
+      */
+    virtual bool useDefaultImplementationForColumnsWithDictionary() const { return true; }
+
     /** Some arguments could remain constant during this implementation.
       */
     virtual ColumnNumbers getArgumentsThatAreAlwaysConstant() const { return {}; }
 
+    /** True if function can be called on default arguments (include Nullable's) and won't throw.
+      * Counterexample: modulo(0, 0)
+      */
+    virtual bool canBeExecutedOnDefaultArguments() const { return true; }
+
 private:
     bool defaultImplementationForNulls(Block & block, const ColumnNumbers & args, size_t result,
                                            size_t input_rows_count);
     bool defaultImplementationForConstantArguments(Block & block, const ColumnNumbers & args, size_t result,
                                                        size_t input_rows_count);
+    void executeWithoutColumnsWithDictionary(Block & block, const ColumnNumbers & arguments, size_t result,
+                                             size_t input_rows_count);
 };
 
 using ValuePlaceholders = std::vector<std::function<llvm::Value * ()>>;
@@ -159,9 +172,9 @@ public:
       * Example: now(). Another example: functions that work with periodically updated dictionaries.
       */
 
-    virtual bool isDeterministic() { return true; }
+    virtual bool isDeterministic() const { return true; }
 
-    virtual bool isDeterministicInScopeOfQuery() { return true; }
+    virtual bool isDeterministicInScopeOfQuery() const { return true; }
 
     /** Lets you know if the function is monotonic in a range of values.
       * This is used to work with the index in a sorted chunk of data.
@@ -266,12 +279,25 @@ protected:
       */
     virtual bool useDefaultImplementationForNulls() const { return true; }
 
+    /** If useDefaultImplementationForNulls() is true, than change arguments for getReturnType() and buildImpl().
+      * If function arguments has types with dictionary, convert them to ordinary types.
+      * getReturnType returns ColumnWithDictionary if at least one argument type is ColumnWithDictionary.
+      */
+    virtual bool useDefaultImplementationForColumnsWithDictionary() const { return true; }
+
+    /// If it isn't, will convert all ColumnWithDictionary arguments to full columns.
+    virtual bool canBeExecutedOnLowCardinalityDictionary() const { return true; }
+
     virtual FunctionBasePtr buildImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr & return_type) const = 0;
 
     virtual void getLambdaArgumentTypesImpl(DataTypes & /*arguments*/) const
     {
         throw Exception("Function " + getName() + " can't have lambda-expressions as arguments", ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
     }
+
+private:
+
+    DataTypePtr getReturnTypeWithoutDictionary(const ColumnsWithTypeAndName & arguments) const;
 };
 
 /// Previous function interface.
@@ -286,7 +312,10 @@ public:
     /// Override this functions to change default implementation behavior. See details in IMyFunction.
     bool useDefaultImplementationForNulls() const override { return true; }
     bool useDefaultImplementationForConstants() const override { return false; }
+    bool useDefaultImplementationForColumnsWithDictionary() const override { return true; }
     ColumnNumbers getArgumentsThatAreAlwaysConstant() const override { return {}; }
+    bool canBeExecutedOnDefaultArguments() const override { return true; }
+    bool canBeExecutedOnLowCardinalityDictionary() const override { return isDeterministicInScopeOfQuery(); }
 
     using PreparedFunctionImpl::execute;
     using FunctionBuilderImpl::getReturnTypeImpl;
@@ -365,7 +394,9 @@ protected:
     }
     bool useDefaultImplementationForNulls() const final { return function->useDefaultImplementationForNulls(); }
     bool useDefaultImplementationForConstants() const final { return function->useDefaultImplementationForConstants(); }
+    bool useDefaultImplementationForColumnsWithDictionary() const final { return function->useDefaultImplementationForColumnsWithDictionary(); }
     ColumnNumbers getArgumentsThatAreAlwaysConstant() const final { return function->getArgumentsThatAreAlwaysConstant(); }
+    bool canBeExecutedOnDefaultArguments() const override { return function->canBeExecutedOnDefaultArguments(); }
 
 private:
     std::shared_ptr<IFunction> function;
@@ -396,9 +427,9 @@ public:
 
     bool isInjective(const Block & sample_block) override { return function->isInjective(sample_block); }
 
-    bool isDeterministic() override { return function->isDeterministic(); }
+    bool isDeterministic() const override { return function->isDeterministic(); }
 
-    bool isDeterministicInScopeOfQuery() override { return function->isDeterministicInScopeOfQuery(); }
+    bool isDeterministicInScopeOfQuery() const override { return function->isDeterministicInScopeOfQuery(); }
 
     bool hasInformationAboutMonotonicity() const override { return function->hasInformationAboutMonotonicity(); }
 
@@ -431,6 +462,8 @@ protected:
     DataTypePtr getReturnTypeImpl(const ColumnsWithTypeAndName & arguments) const override { return function->getReturnTypeImpl(arguments); }
 
     bool useDefaultImplementationForNulls() const override { return function->useDefaultImplementationForNulls(); }
+    bool useDefaultImplementationForColumnsWithDictionary() const override { return function->useDefaultImplementationForColumnsWithDictionary(); }
+    bool canBeExecutedOnLowCardinalityDictionary() const override { return function->canBeExecutedOnLowCardinalityDictionary(); }
 
     FunctionBasePtr buildImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr & return_type) const override
     {

dbms/src/Interpreters/Context.cpp

@@ -618,6 +618,13 @@ void Context::checkDatabaseAccessRights(const std::string & database_name) const
     checkDatabaseAccessRightsImpl(database_name);
 }
 
+bool Context::hasDatabaseAccessRights(const String & database_name) const
+{
+    auto lock = getLock();
+    return client_info.current_user.empty() || (database_name == "system") ||
+        shared->security_manager->hasAccessToDatabase(client_info.current_user, database_name);
+}
+
 void Context::checkDatabaseAccessRightsImpl(const std::string & database_name) const
 {
     if (client_info.current_user.empty() || (database_name == "system"))
@@ -1793,6 +1800,7 @@ std::shared_ptr<ActionLocksManager> Context::getActionLocksManager()
     return shared->action_locks_manager;
 }
 
+
 SessionCleaner::~SessionCleaner()
 {
     try

dbms/src/Interpreters/Context.h

@@ -180,6 +180,7 @@ public:
     bool isTableExist(const String & database_name, const String & table_name) const;
     bool isDatabaseExist(const String & database_name) const;
     bool isExternalTableExist(const String & table_name) const;
+    bool hasDatabaseAccessRights(const String & database_name) const;
     void assertTableExists(const String & database_name, const String & table_name) const;
 
     /** The parameter check_database_access_rights exists to not check the permissions of the database again,

dbms/src/Interpreters/ExpressionJIT.cpp

@@ -476,7 +476,7 @@ public:
 
     PreparedFunctionPtr prepare(const Block &) const override { return std::make_shared<LLVMPreparedFunction>(name, context); }
 
-    bool isDeterministic() override
+    bool isDeterministic() const override
     {
         for (const auto & f : originals)
             if (!f->isDeterministic())
@@ -484,7 +484,7 @@ public:
         return true;
     }
 
-    bool isDeterministicInScopeOfQuery() override
+    bool isDeterministicInScopeOfQuery() const override
     {
         for (const auto & f : originals)
             if (!f->isDeterministicInScopeOfQuery())

dbms/src/Interpreters/InterpreterCreateQuery.cpp

@@ -53,6 +53,7 @@ namespace ErrorCodes
     extern const int UNKNOWN_DATABASE_ENGINE;
     extern const int DUPLICATE_COLUMN;
     extern const int READONLY;
+    extern const int ILLEGAL_COLUMN;
 }
 
 
@@ -348,6 +349,33 @@ ColumnsDescription InterpreterCreateQuery::getColumnsDescription(const ASTExpres
 }
 
 
+void InterpreterCreateQuery::checkSupportedTypes(const ColumnsDescription & columns, const Context & context)
+{
+    const auto & settings = context.getSettingsRef();
+    bool allow_low_cardinality = settings.allow_experimental_low_cardinality_type != 0;
+
+    if (allow_low_cardinality)
+        return;
+
+    auto check_types = [&](const NamesAndTypesList & list)
+    {
+        for (const auto & column : list)
+        {
+            if (!allow_low_cardinality && column.type && column.type->withDictionary())
+            {
+                String message = "Cannot create table with column " + column.name + " which type is "
+                                 + column.type->getName() + " because LowCardinality type is not allowed. "
+                                 + "Set setting allow_experimental_low_cardinality_type = 1 in order to allow it.";
+                throw Exception(message, ErrorCodes::ILLEGAL_COLUMN);
+            }
+        }
+    };
+
+    check_types(columns.ordinary);
+    check_types(columns.materialized);
+}
+
+
 ColumnsDescription InterpreterCreateQuery::setColumns(
     ASTCreateQuery & create, const Block & as_select_sample, const StoragePtr & as_storage) const
 {
@@ -487,6 +515,10 @@ BlockIO InterpreterCreateQuery::createTable(ASTCreateQuery & create)
     /// Set and retrieve list of columns.
     ColumnsDescription columns = setColumns(create, as_select_sample, as_storage);
 
+    /// Some column types may be not allowed according to settings.
+    if (!create.attach)
+        checkSupportedTypes(columns, context);
+
     /// Set the table engine if it was not specified explicitly.
     setEngine(create);
 

dbms/src/Interpreters/InterpreterCreateQuery.h

@@ -46,6 +46,8 @@ public:
 
     /// Obtain information about columns, their types and default values, for case when columns in CREATE query is specified explicitly.
     static ColumnsDescription getColumnsDescription(const ASTExpressionList & columns, const Context & context);
+    /// Check that column types are allowed for usage in table according to settings.
+    static void checkSupportedTypes(const ColumnsDescription & columns, const Context & context);
 
 private:
     BlockIO createDatabase(ASTCreateQuery & create);

dbms/src/Interpreters/InterpreterSelectQuery.cpp

@@ -18,6 +18,7 @@
 #include <DataStreams/CreatingSetsBlockInputStream.h>
 #include <DataStreams/MaterializingBlockInputStream.h>
 #include <DataStreams/ConcatBlockInputStream.h>
+#include <DataStreams/ConvertColumnWithDictionaryToFullBlockInputStream.h>
 
 #include <Parsers/ASTSelectQuery.h>
 #include <Parsers/ASTSelectWithUnionQuery.h>
@@ -775,7 +776,8 @@ void InterpreterSelectQuery::executeAggregation(Pipeline & pipeline, const Expre
 {
     pipeline.transform([&](auto & stream)
     {
-        stream = std::make_shared<ExpressionBlockInputStream>(stream, expression);
+        stream = std::make_shared<ConvertColumnWithDictionaryToFullBlockInputStream>(
+                std::make_shared<ExpressionBlockInputStream>(stream, expression));
     });
 
     Names key_names;

dbms/src/Interpreters/Settings.h

@@ -268,6 +268,11 @@ struct Settings
     M(SettingUInt64, enable_conditional_computation, 0, "Enable conditional computations") \
     \
     M(SettingDateTimeInputFormat, date_time_input_format, FormatSettings::DateTimeInputFormat::Basic, "Method to read DateTime from text input formats. Possible values: 'basic' and 'best_effort'.") \
+    \
+    M(SettingUInt64, low_cardinality_max_dictionary_size, 8192, "Maximum size (in rows) of shared global dictionary for LowCardinality type.") \
+    M(SettingBool, low_cardinality_use_single_dictionary_for_part, false, "LowCardinality type serialization setting. If is true, than will use additional keys when global dictionary overflows. Otherwise, will create several shared dictionaries.") \
+    M(SettingBool, allow_experimental_low_cardinality_type, false, "Allows to create table with LowCardinality types.") \
+    \
     M(SettingBool, prefer_localhost_replica, 1, "1 - always send query to local replica, if it exists. 0 - choose replica to send query between local and remote ones according to load_balancing") \
     M(SettingUInt64, max_fetch_partition_retries_count, 5, "Amount of retries while fetching partition from another host.") \
 

dbms/src/Interpreters/convertFieldToType.cpp

@@ -18,6 +18,7 @@
 #include <Common/typeid_cast.h>
 #include <Common/NaNUtils.h>
 #include <DataTypes/DataTypeUUID.h>
+#include <DataTypes/DataTypeWithDictionary.h>
 
 
 namespace DB
@@ -217,12 +218,10 @@ Field convertFieldToType(const Field & from_value, const IDataType & to_type, co
     if (from_type_hint && from_type_hint->equals(to_type))
         return from_value;
 
-    if (to_type.isNullable())
-    {
-        const DataTypeNullable & nullable_type = static_cast<const DataTypeNullable &>(to_type);
-        const DataTypePtr & nested_type = nullable_type.getNestedType();
-        return convertFieldToTypeImpl(from_value, *nested_type);
-    }
+    if (auto * with_dict_type = typeid_cast<const DataTypeWithDictionary *>(&to_type))
+        return convertFieldToType(from_value, *with_dict_type->getDictionaryType(), from_type_hint);
+    else if (auto * nullable_type = typeid_cast<const DataTypeNullable *>(&to_type))
+        return convertFieldToTypeImpl(from_value, *nullable_type->getNestedType());
     else
         return convertFieldToTypeImpl(from_value, to_type);
 }

dbms/src/Parsers/ParserSelectQuery.cpp

@@ -17,6 +17,7 @@ namespace DB
 namespace ErrorCodes
 {
     extern const int SYNTAX_ERROR;
+    extern const int TOP_AND_LIMIT_TOGETHER;
 }
 
 
@@ -38,6 +39,8 @@ bool ParserSelectQuery::parseImpl(Pos & pos, ASTPtr & node, Expected & expected)
     ParserKeyword s_limit("LIMIT");
     ParserKeyword s_settings("SETTINGS");
     ParserKeyword s_by("BY");
+    ParserKeyword s_top("TOP");
+    ParserKeyword s_offset("OFFSET");
 
     ParserNotEmptyExpressionList exp_list(false);
     ParserNotEmptyExpressionList exp_list_for_with_clause(false, true); /// Set prefer_alias_to_column_name for each alias.
@@ -62,6 +65,26 @@ bool ParserSelectQuery::parseImpl(Pos & pos, ASTPtr & node, Expected & expected)
         if (s_distinct.ignore(pos, expected))
             select_query->distinct = true;
 
+        if (s_top.ignore(pos, expected))
+        {
+            ParserToken open_bracket(TokenType::OpeningRoundBracket);
+            ParserToken close_bracket(TokenType::ClosingRoundBracket);
+            ParserNumber num;
+
+            if (open_bracket.ignore(pos, expected))
+            {
+                if (!num.parse(pos, select_query->limit_length, expected))
+                    return false;
+                if (!close_bracket.ignore(pos, expected))
+                    return false;
+            }
+            else
+            {
+                if (!num.parse(pos, select_query->limit_length, expected))
+                    return false;
+            }
+        }
+
         if (!exp_list_for_select_clause.parse(pos, select_query->select_expression_list, expected))
             return false;
     }
@@ -120,6 +143,9 @@ bool ParserSelectQuery::parseImpl(Pos & pos, ASTPtr & node, Expected & expected)
     /// LIMIT length | LIMIT offset, length | LIMIT count BY expr-list
     if (s_limit.ignore(pos, expected))
     {
+        if (select_query->limit_length)
+            throw Exception("Can not use TOP and LIMIT together", ErrorCodes::TOP_AND_LIMIT_TOGETHER);
+
         ParserToken s_comma(TokenType::Comma);
         ParserNumber num;
 
@@ -140,6 +166,11 @@ bool ParserSelectQuery::parseImpl(Pos & pos, ASTPtr & node, Expected & expected)
             if (!exp_list.parse(pos, select_query->limit_by_expression_list, expected))
                 return false;
         }
+        else if (s_offset.ignore(pos, expected))
+        {
+            if (!num.parse(pos, select_query->limit_offset, expected))
+                return false;
+        }
     }
 
     /// LIMIT length | LIMIT offset, length

dbms/src/Storages/MergeTree/MergeTreeDataPart.cpp

@@ -586,12 +586,13 @@ void MergeTreeDataPart::accumulateColumnSizes(ColumnToSize & column_to_size) con
 
     for (const NameAndTypePair & name_type : storage.getColumns().getAllPhysical())
     {
+        IDataType::SubstreamPath path;
         name_type.type->enumerateStreams([&](const IDataType::SubstreamPath & substream_path)
         {
             Poco::File bin_file(getFullPath() + IDataType::getFileNameForStream(name_type.name, substream_path) + ".bin");
             if (bin_file.exists())
                 column_to_size[name_type.name] += bin_file.getSize();
-        }, {});
+        }, path);
     }
 }
 
@@ -637,6 +638,7 @@ void MergeTreeDataPart::checkConsistency(bool require_part_metadata)
         {
             for (const NameAndTypePair & name_type : columns)
             {
+                IDataType::SubstreamPath stream_path;
                 name_type.type->enumerateStreams([&](const IDataType::SubstreamPath & substream_path)
                 {
                     String file_name = IDataType::getFileNameForStream(name_type.name, substream_path);
@@ -648,7 +650,7 @@ void MergeTreeDataPart::checkConsistency(bool require_part_metadata)
                     if (!checksums.files.count(bin_file_name))
                         throw Exception("No " + bin_file_name + " file checksum for column " + name + " in part " + path,
                             ErrorCodes::NO_FILE_IN_DATA_PART);
-                }, {});
+                }, stream_path);
             }
         }
 
@@ -721,7 +723,7 @@ void MergeTreeDataPart::checkConsistency(bool require_part_metadata)
                         throw Exception("Part " + path + " is broken: marks have different sizes.",
                             ErrorCodes::BAD_SIZE_OF_FILE_IN_DATA_PART);
                 }
-            }, {});
+            });
         }
     }
 }

dbms/src/Storages/MergeTree/MergeTreeReader.cpp

@@ -42,7 +42,7 @@ MergeTreeReader::MergeTreeReader(const String & path,
     clockid_t clock_type)
     : avg_value_size_hints(avg_value_size_hints), path(path), data_part(data_part), columns(columns)
     , uncompressed_cache(uncompressed_cache), mark_cache(mark_cache), save_marks_in_cache(save_marks_in_cache), storage(storage)
-    , all_mark_ranges(all_mark_ranges), aio_threshold(aio_threshold), max_read_buffer_size(max_read_buffer_size)
+    , all_mark_ranges(all_mark_ranges), aio_threshold(aio_threshold), max_read_buffer_size(max_read_buffer_size), index_granularity(storage.index_granularity)
 {
     try
     {
@@ -200,6 +200,11 @@ MergeTreeReader::Stream::Stream(
             getMark(right).offset_in_compressed_file - getMark(all_mark_ranges[i].begin).offset_in_compressed_file);
     }
 
+    /// Avoid empty buffer. May happen while reading dictionary for DataTypeWithDictionary.
+    /// For example: part has single dictionary and all marks point to the same position.
+    if (max_mark_range == 0)
+        max_mark_range = max_read_buffer_size;
+
     size_t buffer_size = std::min(max_read_buffer_size, max_mark_range);
 
     /// Estimate size of the data to be read.
@@ -329,6 +334,26 @@ void MergeTreeReader::Stream::seekToMark(size_t index)
 }
 
 
+void MergeTreeReader::Stream::seekToStart()
+{
+    try
+    {
+        if (cached_buffer)
+            cached_buffer->seek(0, 0);
+        if (non_cached_buffer)
+            non_cached_buffer->seek(0, 0);
+    }
+    catch (Exception & e)
+    {
+        /// Better diagnostics.
+        if (e.code() == ErrorCodes::ARGUMENT_OUT_OF_BOUND)
+            e.addMessage("(while seeking to start of column " + path_prefix + ")");
+
+        throw;
+    }
+}
+
+
 void MergeTreeReader::addStreams(const String & name, const IDataType & type, const MarkRanges & all_mark_ranges,
     const ReadBufferFromFileBase::ProfileCallback & profile_callback, clockid_t clock_type)
 {
@@ -353,7 +378,8 @@ void MergeTreeReader::addStreams(const String & name, const IDataType & type, co
             uncompressed_cache, aio_threshold, max_read_buffer_size, profile_callback, clock_type));
     };
 
-    type.enumerateStreams(callback, {});
+    IDataType::SubstreamPath path;
+    type.enumerateStreams(callback, path);
 }
 
 
@@ -362,7 +388,9 @@ void MergeTreeReader::readData(
     size_t from_mark, bool continue_reading, size_t max_rows_to_read,
     bool with_offsets)
 {
-    IDataType::InputStreamGetter stream_getter = [&] (const IDataType::SubstreamPath & path) -> ReadBuffer *
+    auto get_stream_getter = [&](bool stream_for_prefix) -> IDataType::InputStreamGetter
+    {
+        return [&, stream_for_prefix](const IDataType::SubstreamPath & path) -> ReadBuffer *
         {
             /// If offsets for arrays have already been read.
             if (!with_offsets && path.size() == 1 && path[0].type == IDataType::Substream::ArraySizes)
@@ -376,14 +404,31 @@ void MergeTreeReader::readData(
 
             Stream & stream = *it->second;
 
-        if (!continue_reading)
+            if (stream_for_prefix)
+            {
+                stream.seekToStart();
+                continue_reading = false;
+            }
+            else if (!continue_reading)
                 stream.seekToMark(from_mark);
 
             return stream.data_buffer;
         };
+    };
 
     double & avg_value_size_hint = avg_value_size_hints[name];
-    type.deserializeBinaryBulkWithMultipleStreams(column, stream_getter, max_rows_to_read, avg_value_size_hint, true, {});
+    IDataType::DeserializeBinaryBulkSettings settings;
+    settings.avg_value_size_hint = avg_value_size_hint;
+
+    if (deserialize_binary_bulk_state_map.count(name) == 0)
+    {
+        settings.getter = get_stream_getter(true);
+        type.deserializeBinaryBulkStatePrefix(settings, deserialize_binary_bulk_state_map[name]);
+    }
+
+    settings.getter = get_stream_getter(false);
+    auto & deserialize_state = deserialize_binary_bulk_state_map[name];
+    type.deserializeBinaryBulkWithMultipleStreams(column, max_rows_to_read, settings, deserialize_state);
     IDataType::updateAvgValueSizeHint(column, avg_value_size_hint);
 }
 

dbms/src/Storages/MergeTree/MergeTreeReader.h

@@ -23,6 +23,7 @@ class MergeTreeReader : private boost::noncopyable
 {
 public:
     using ValueSizeMap = std::map<std::string, double>;
+    using DeserializeBinaryBulkStateMap = std::map<std::string, IDataType::DeserializeBinaryBulkStatePtr>;
 
     MergeTreeReader(const String & path, /// Path to the directory containing the part
         const MergeTreeData::DataPartPtr & data_part, const NamesAndTypesList & columns,
@@ -63,6 +64,7 @@ private:
             const ReadBufferFromFileBase::ProfileCallback & profile_callback, clockid_t clock_type);
 
         void seekToMark(size_t index);
+        void seekToStart();
 
         ReadBuffer * data_buffer;
 
@@ -91,6 +93,8 @@ private:
 
     /// avg_value_size_hints are used to reduce the number of reallocations when creating columns of variable size.
     ValueSizeMap avg_value_size_hints;
+    /// Stores states for IDataType::deserializeBinaryBulk
+    DeserializeBinaryBulkStateMap deserialize_binary_bulk_state_map;
     String path;
     MergeTreeData::DataPartPtr data_part;
 
@@ -108,6 +112,7 @@ private:
     MarkRanges all_mark_ranges;
     size_t aio_threshold;
     size_t max_read_buffer_size;
+    size_t index_granularity;
 
     void addStreams(const String & name, const IDataType & type, const MarkRanges & all_mark_ranges,
         const ReadBufferFromFileBase::ProfileCallback & profile_callback, clockid_t clock_type);

dbms/src/Storages/MergeTree/MergedBlockOutputStream.cpp

@@ -64,7 +64,28 @@ void IMergedBlockOutputStream::addStreams(
             aio_threshold);
     };
 
-    type.enumerateStreams(callback, {});
+    IDataType::SubstreamPath stream_path;
+    type.enumerateStreams(callback, stream_path);
+}
+
+
+IDataType::OutputStreamGetter IMergedBlockOutputStream::createStreamGetter(
+        const String & name, OffsetColumns & offset_columns, bool skip_offsets)
+{
+    return [&, skip_offsets] (const IDataType::SubstreamPath & substream_path) -> WriteBuffer *
+    {
+        bool is_offsets = !substream_path.empty() && substream_path.back().type == IDataType::Substream::ArraySizes;
+        if (is_offsets && skip_offsets)
+            return nullptr;
+
+        String stream_name = IDataType::getFileNameForStream(name, substream_path);
+
+        /// Don't write offsets more than one time for Nested type.
+        if (is_offsets && offset_columns.count(stream_name))
+            return nullptr;
+
+        return &column_streams[stream_name]->compressed;
+    };
 }
 
 
@@ -73,8 +94,15 @@ void IMergedBlockOutputStream::writeData(
     const IDataType & type,
     const IColumn & column,
     OffsetColumns & offset_columns,
-    bool skip_offsets)
+    bool skip_offsets,
+    IDataType::SerializeBinaryBulkStatePtr & serialization_state)
 {
+    auto & settings = storage.context.getSettingsRef();
+    IDataType::SerializeBinaryBulkSettings serialize_settings;
+    serialize_settings.getter = createStreamGetter(name, offset_columns, skip_offsets);
+    serialize_settings.low_cardinality_max_dictionary_size = settings.low_cardinality_max_dictionary_size;
+    serialize_settings.low_cardinality_use_single_dictionary_for_part = settings.low_cardinality_use_single_dictionary_for_part != 0;
+
     size_t size = column.size();
     size_t prev_mark = 0;
     while (prev_mark < size)
@@ -109,25 +137,10 @@ void IMergedBlockOutputStream::writeData(
 
                 writeIntBinary(stream.plain_hashing.count(), stream.marks);
                 writeIntBinary(stream.compressed.offset(), stream.marks);
-            }, {});
+            }, serialize_settings.path);
         }
 
-        IDataType::OutputStreamGetter stream_getter = [&] (const IDataType::SubstreamPath & substream_path) -> WriteBuffer *
-        {
-            bool is_offsets = !substream_path.empty() && substream_path.back().type == IDataType::Substream::ArraySizes;
-            if (is_offsets && skip_offsets)
-                return nullptr;
-
-            String stream_name = IDataType::getFileNameForStream(name, substream_path);
-
-            /// Don't write offsets more than one time for Nested type.
-            if (is_offsets && offset_columns.count(stream_name))
-                return nullptr;
-
-            return &column_streams[stream_name]->compressed;
-        };
-
-        type.serializeBinaryBulkWithMultipleStreams(column, stream_getter, prev_mark, limit, true, {});
+        type.serializeBinaryBulkWithMultipleStreams(column, prev_mark, limit, serialize_settings, serialization_state);
 
         /// So that instead of the marks pointing to the end of the compressed block, there were marks pointing to the beginning of the next one.
         type.enumerateStreams([&] (const IDataType::SubstreamPath & substream_path)
@@ -143,7 +156,7 @@ void IMergedBlockOutputStream::writeData(
                 return;
 
             column_streams[stream_name]->compressed.nextIfAtEnd();
-        }, {});
+        }, serialize_settings.path);
 
         prev_mark += limit;
     }
@@ -157,7 +170,7 @@ void IMergedBlockOutputStream::writeData(
             String stream_name = IDataType::getFileNameForStream(name, substream_path);
             offset_columns.insert(stream_name);
         }
-    }, {});
+    }, serialize_settings.path);
 }
 
 
@@ -284,6 +297,19 @@ void MergedBlockOutputStream::writeSuffixAndFinalizePart(
         const NamesAndTypesList * total_column_list,
         MergeTreeData::DataPart::Checksums * additional_column_checksums)
 {
+    /// Finish columns serialization.
+    auto & settings = storage.context.getSettingsRef();
+    IDataType::SerializeBinaryBulkSettings serialize_settings;
+    serialize_settings.low_cardinality_max_dictionary_size = settings.low_cardinality_max_dictionary_size;
+    serialize_settings.low_cardinality_use_single_dictionary_for_part = settings.low_cardinality_use_single_dictionary_for_part != 0;
+    OffsetColumns offset_columns;
+    auto it = columns_list.begin();
+    for (size_t i = 0; i < columns_list.size(); ++i, ++it)
+    {
+        serialize_settings.getter = createStreamGetter(it->name, offset_columns, false);
+        it->type->serializeBinaryBulkStateSuffix(serialize_settings, serialization_states[i]);
+    }
+
     if (!total_column_list)
         total_column_list = &columns_list;
 
@@ -395,28 +421,44 @@ void MergedBlockOutputStream::writeImpl(const Block & block, const IColumn::Perm
             index_columns[i] = primary_columns[i].column->cloneEmpty();
     }
 
-    /// Now write the data.
-    for (const auto & it : columns_list)
+    if (serialization_states.empty())
     {
-        const ColumnWithTypeAndName & column = block.getByName(it.name);
+        serialization_states.reserve(columns_list.size());
+        OffsetColumns tmp_offset_columns;
+        IDataType::SerializeBinaryBulkSettings settings;
+
+        for (const auto & col : columns_list)
+        {
+            settings.getter = createStreamGetter(col.name, tmp_offset_columns, false);
+            serialization_states.emplace_back(nullptr);
+            col.type->serializeBinaryBulkStatePrefix(settings, serialization_states.back());
+        }
+    }
+
+    /// Now write the data.
+    auto it = columns_list.begin();
+    for (size_t i = 0; i < columns_list.size(); ++i, ++it)
+    {
+        const ColumnWithTypeAndName & column = block.getByName(it->name);
 
         if (permutation)
         {
-            auto primary_column_it = primary_columns_name_to_position.find(it.name);
+            auto primary_column_it = primary_columns_name_to_position.find(it->name);
             if (primary_columns_name_to_position.end() != primary_column_it)
             {
-                writeData(column.name, *column.type, *primary_columns[primary_column_it->second].column, offset_columns, false);
+                auto & primary_column = *primary_columns[primary_column_it->second].column;
+                writeData(column.name, *column.type, primary_column, offset_columns, false, serialization_states[i]);
             }
             else
             {
                 /// We rearrange the columns that are not included in the primary key here; Then the result is released - to save RAM.
-                ColumnPtr permutted_column = column.column->permute(*permutation, 0);
-                writeData(column.name, *column.type, *permutted_column, offset_columns, false);
+                ColumnPtr permuted_column = column.column->permute(*permutation, 0);
+                writeData(column.name, *column.type, *permuted_column, offset_columns, false, serialization_states[i]);
             }
         }
         else
         {
-            writeData(column.name, *column.type, *column.column, offset_columns, false);
+            writeData(column.name, *column.type, *column.column, offset_columns, false, serialization_states[i]);
         }
     }
 
@@ -470,11 +512,21 @@ void MergedColumnOnlyOutputStream::write(const Block & block)
     if (!initialized)
     {
         column_streams.clear();
+        serialization_states.clear();
+        serialization_states.reserve(block.columns());
+        OffsetColumns tmp_offset_columns;
+        IDataType::SerializeBinaryBulkSettings settings;
+
         for (size_t i = 0; i < block.columns(); ++i)
         {
-            addStreams(part_path, block.safeGetByPosition(i).name,
-                *block.safeGetByPosition(i).type, 0, skip_offsets);
+            const auto & col = block.safeGetByPosition(i);
+
+            addStreams(part_path, col.name, *col.type, 0, skip_offsets);
+            serialization_states.emplace_back(nullptr);
+            settings.getter = createStreamGetter(col.name, tmp_offset_columns, false);
+            col.type->serializeBinaryBulkStatePrefix(settings, serialization_states.back());
         }
+
         initialized = true;
     }
 
@@ -484,7 +536,7 @@ void MergedColumnOnlyOutputStream::write(const Block & block)
     for (size_t i = 0; i < block.columns(); ++i)
     {
         const ColumnWithTypeAndName & column = block.safeGetByPosition(i);
-        writeData(column.name, *column.type, *column.column, offset_columns, skip_offsets);
+        writeData(column.name, *column.type, *column.column, offset_columns, skip_offsets, serialization_states[i]);
     }
 
     size_t written_for_last_mark = (storage.index_granularity - index_offset + rows) % storage.index_granularity;
@@ -498,6 +550,19 @@ void MergedColumnOnlyOutputStream::writeSuffix()
 
 MergeTreeData::DataPart::Checksums MergedColumnOnlyOutputStream::writeSuffixAndGetChecksums()
 {
+    /// Finish columns serialization.
+    auto & settings = storage.context.getSettingsRef();
+    IDataType::SerializeBinaryBulkSettings serialize_settings;
+    serialize_settings.low_cardinality_max_dictionary_size = settings.low_cardinality_max_dictionary_size;
+    serialize_settings.low_cardinality_use_single_dictionary_for_part = settings.low_cardinality_use_single_dictionary_for_part != 0;
+    OffsetColumns offset_columns;
+    for (size_t i = 0; i < header.columns(); ++i)
+    {
+        auto & column = header.safeGetByPosition(i);
+        serialize_settings.getter = createStreamGetter(column.name, offset_columns, skip_offsets);
+        column.type->serializeBinaryBulkStateSuffix(serialize_settings, serialization_states[i]);
+    }
+
     MergeTreeData::DataPart::Checksums checksums;
 
     for (auto & column_stream : column_streams)
@@ -510,6 +575,7 @@ MergeTreeData::DataPart::Checksums MergedColumnOnlyOutputStream::writeSuffixAndG
     }
 
     column_streams.clear();
+    serialization_states.clear();
     initialized = false;
 
     return checksums;

dbms/src/Storages/MergeTree/MergedBlockOutputStream.h

@@ -25,6 +25,8 @@ public:
 
 protected:
     using OffsetColumns = std::set<std::string>;
+    using SerializationState = IDataType::SerializeBinaryBulkStatePtr;
+    using SerializationStates = std::vector<SerializationState>;
 
     struct ColumnStream
     {
@@ -64,8 +66,12 @@ protected:
 
     void addStreams(const String & path, const String & name, const IDataType & type, size_t estimated_size, bool skip_offsets);
 
+
+    IDataType::OutputStreamGetter createStreamGetter(const String & name, OffsetColumns & offset_columns, bool skip_offsets);
+
     /// Write data of one column.
-    void writeData(const String & name, const IDataType & type, const IColumn & column, OffsetColumns & offset_columns, bool skip_offsets);
+    void writeData(const String & name, const IDataType & type, const IColumn & column, OffsetColumns & offset_columns,
+                   bool skip_offsets, IDataType::SerializeBinaryBulkStatePtr & serialization_state);
 
     MergeTreeData & storage;
 
@@ -132,6 +138,7 @@ private:
 
 private:
     NamesAndTypesList columns_list;
+    SerializationStates serialization_states;
     String part_path;
 
     size_t rows_count = 0;
@@ -158,6 +165,7 @@ public:
 
 private:
     Block header;
+    SerializationStates serialization_states;
     String part_path;
 
     bool initialized = false;

dbms/src/Storages/MergeTree/checkDataPart.cpp

@@ -249,6 +249,8 @@ MergeTreeData::DataPart::Checksums checkDataPart(
 
         while (true)
         {
+            IDataType::DeserializeBinaryBulkSettings settings;
+
             /// Check that mark points to current position in file.
             bool marks_eof = false;
             name_type.type->enumerateStreams([&](const IDataType::SubstreamPath & substream_path)
@@ -270,7 +272,7 @@ MergeTreeData::DataPart::Checksums checkDataPart(
                             + ", mrk file offset: " + toString(stream.mrk_hashing_buf.count()));
                         throw;
                     }
-                }, {});
+                }, settings.path);
 
             ++mark_num;
 
@@ -278,18 +280,18 @@ MergeTreeData::DataPart::Checksums checkDataPart(
             /// NOTE Shared array sizes of Nested columns are read more than once. That's Ok.
 
             MutableColumnPtr tmp_column = name_type.type->createColumn();
-            name_type.type->deserializeBinaryBulkWithMultipleStreams(
-                *tmp_column,
-                [&](const IDataType::SubstreamPath & substream_path)
+            settings.getter = [&](const IDataType::SubstreamPath & substream_path)
             {
                 String file_name = IDataType::getFileNameForStream(name_type.name, substream_path);
                 auto stream_it = streams.find(file_name);
                 if (stream_it == streams.end())
                     throw Exception("Logical error: cannot find stream " + file_name);
                 return &stream_it->second.uncompressed_hashing_buf;
-                },
-                index_granularity,
-                0, true, {});
+            };
+
+            IDataType::DeserializeBinaryBulkStatePtr state;
+            name_type.type->deserializeBinaryBulkStatePrefix(settings, state);
+            name_type.type->deserializeBinaryBulkWithMultipleStreams(*tmp_column, index_granularity, settings, state);
 
             size_t read_size = tmp_column->size();
             column_size += read_size;

dbms/src/Storages/StorageLog.cpp

@@ -76,6 +76,7 @@ protected:
     Block readImpl() override;
 
 private:
+
     size_t block_size;
     NamesAndTypesList columns;
     StorageLog & storage;
@@ -101,6 +102,10 @@ private:
     using FileStreams = std::map<std::string, Stream>;
     FileStreams streams;
 
+    using DeserializeState = IDataType::DeserializeBinaryBulkStatePtr;
+    using DeserializeStates = std::map<String, DeserializeState>;
+    DeserializeStates deserialize_states;
+
     void readData(const String & name, const IDataType & type, IColumn & column, size_t max_rows_to_read);
 
 };
@@ -168,6 +173,12 @@ private:
 
     WriteBufferFromFile marks_stream; /// Declared below `lock` to make the file open when rwlock is captured.
 
+    using SerializeState = IDataType::SerializeBinaryBulkStatePtr;
+    using SerializeStates = std::map<String, SerializeState>;
+    SerializeStates serialize_states;
+
+    IDataType::OutputStreamGetter createStreamGetter(const String & name, WrittenStreams & written_streams);
+
     void writeData(const String & name, const IDataType & type, const IColumn & column,
         MarksForColumns & out_marks,
         WrittenStreams & written_streams);
@@ -226,7 +237,11 @@ Block LogBlockInputStream::readImpl()
 
 void LogBlockInputStream::readData(const String & name, const IDataType & type, IColumn & column, size_t max_rows_to_read)
 {
-    IDataType::InputStreamGetter stream_getter = [&] (const IDataType::SubstreamPath & path) -> ReadBuffer *
+    IDataType::DeserializeBinaryBulkSettings settings; /// TODO Use avg_value_size_hint.
+
+    auto createStringGetter = [&](bool stream_for_prefix)
+    {
+        return [&] (const IDataType::SubstreamPath & path) -> ReadBuffer *
         {
             String stream_name = IDataType::getFileNameForStream(name, path);
 
@@ -234,17 +249,24 @@ void LogBlockInputStream::readData(const String & name, const IDataType & type,
             if (storage.files.end() == file_it)
                 throw Exception("Logical error: no information about file " + stream_name + " in StorageLog", ErrorCodes::LOGICAL_ERROR);
 
-        auto it = streams.try_emplace(stream_name,
-            file_it->second.data_file.path(),
-            mark_number
-                ? file_it->second.marks[mark_number].offset
-                : 0,
-            max_read_buffer_size).first;
+            UInt64 offset = 0;
+            if (!stream_for_prefix && mark_number)
+                offset = file_it->second.marks[mark_number].offset;
 
+            auto & data_file_path = file_it->second.data_file.path();
+            auto it = streams.try_emplace(stream_name, data_file_path, offset, max_read_buffer_size).first;
             return &it->second.compressed;
         };
+    };
 
-    type.deserializeBinaryBulkWithMultipleStreams(column, stream_getter, max_rows_to_read, 0, true, {}); /// TODO Use avg_value_size_hint.
+    if (deserialize_states.count(name) == 0)
+    {
+        settings.getter = createStringGetter(true);
+        type.deserializeBinaryBulkStatePrefix(settings, deserialize_states[name]);
+    }
+
+    settings.getter = createStringGetter(false);
+    type.deserializeBinaryBulkWithMultipleStreams(column, max_rows_to_read, settings, deserialize_states[name]);
 }
 
 
@@ -274,6 +296,18 @@ void LogBlockOutputStream::writeSuffix()
         return;
     done = true;
 
+    WrittenStreams written_streams;
+    IDataType::SerializeBinaryBulkSettings settings;
+    for (const auto & column : getHeader())
+    {
+        auto it = serialize_states.find(column.name);
+        if (it != serialize_states.end())
+        {
+            settings.getter = createStreamGetter(column.name, written_streams);
+            column.type->serializeBinaryBulkStateSuffix(settings, it->second);
+        }
+    }
+
     /// Finish write.
     marks_stream.next();
 
@@ -291,27 +325,10 @@ void LogBlockOutputStream::writeSuffix()
 }
 
 
-void LogBlockOutputStream::writeData(const String & name, const IDataType & type, const IColumn & column,
-    MarksForColumns & out_marks,
+IDataType::OutputStreamGetter LogBlockOutputStream::createStreamGetter(const String & name,
                                                                        WrittenStreams & written_streams)
 {
-    type.enumerateStreams([&] (const IDataType::SubstreamPath & path)
-    {
-        String stream_name = IDataType::getFileNameForStream(name, path);
-        if (written_streams.count(stream_name))
-            return;
-
-        const auto & file = storage.files[stream_name];
-        const auto stream_it = streams.try_emplace(stream_name, storage.files[stream_name].data_file.path(), storage.max_compress_block_size).first;
-
-        Mark mark;
-        mark.rows = (file.marks.empty() ? 0 : file.marks.back().rows) + column.size();
-        mark.offset = stream_it->second.plain_offset + stream_it->second.plain.count();
-
-        out_marks.emplace_back(file.column_index, mark);
-    }, {});
-
-    IDataType::OutputStreamGetter stream_getter = [&] (const IDataType::SubstreamPath & path) -> WriteBuffer *
+    return [&] (const IDataType::SubstreamPath & path) -> WriteBuffer *
     {
         String stream_name = IDataType::getFileNameForStream(name, path);
         if (written_streams.count(stream_name))
@@ -319,11 +336,50 @@ void LogBlockOutputStream::writeData(const String & name, const IDataType & type
 
         auto it = streams.find(stream_name);
         if (streams.end() == it)
-            throw Exception("Logical error: stream was not created when writing data in LogBlockOutputStream", ErrorCodes::LOGICAL_ERROR);
+            throw Exception("Logical error: stream was not created when writing data in LogBlockOutputStream",
+                            ErrorCodes::LOGICAL_ERROR);
         return &it->second.compressed;
     };
+}
 
-    type.serializeBinaryBulkWithMultipleStreams(column, stream_getter, 0, 0, true, {});
+
+void LogBlockOutputStream::writeData(const String & name, const IDataType & type, const IColumn & column,
+    MarksForColumns & out_marks,
+    WrittenStreams & written_streams)
+{
+    IDataType::SerializeBinaryBulkSettings settings;
+
+    type.enumerateStreams([&] (const IDataType::SubstreamPath & path)
+    {
+        String stream_name = IDataType::getFileNameForStream(name, path);
+        if (written_streams.count(stream_name))
+            return;
+
+        streams.try_emplace(stream_name, storage.files[stream_name].data_file.path(), storage.max_compress_block_size);
+    }, settings.path);
+
+    settings.getter = createStreamGetter(name, written_streams);
+
+    if (serialize_states.count(name) == 0)
+         type.serializeBinaryBulkStatePrefix(settings, serialize_states[name]);
+
+    type.enumerateStreams([&] (const IDataType::SubstreamPath & path)
+    {
+        String stream_name = IDataType::getFileNameForStream(name, path);
+        if (written_streams.count(stream_name))
+            return;
+
+        const auto & file = storage.files[stream_name];
+        const auto stream_it = streams.find(stream_name);
+
+        Mark mark;
+        mark.rows = (file.marks.empty() ? 0 : file.marks.back().rows) + column.size();
+        mark.offset = stream_it->second.plain_offset + stream_it->second.plain.count();
+
+        out_marks.emplace_back(file.column_index, mark);
+    }, settings.path);
+
+    type.serializeBinaryBulkWithMultipleStreams(column, 0, 0, settings, serialize_states[name]);
 
     type.enumerateStreams([&] (const IDataType::SubstreamPath & path)
     {
@@ -335,7 +391,7 @@ void LogBlockOutputStream::writeData(const String & name, const IDataType & type
         if (streams.end() == it)
             throw Exception("Logical error: stream was not created when writing data in LogBlockOutputStream", ErrorCodes::LOGICAL_ERROR);
         it->second.compressed.next();
-    }, {});
+    }, settings.path);
 }
 
 
@@ -401,7 +457,8 @@ void StorageLog::addFiles(const String & column_name, const IDataType & type)
         }
     };
 
-    type.enumerateStreams(stream_callback, {});
+    IDataType::SubstreamPath path;
+    type.enumerateStreams(stream_callback, path);
 }
 
 
@@ -497,11 +554,12 @@ const StorageLog::Marks & StorageLog::getMarksWithRealRowCount() const
       * If this is a data type with multiple stream, get the first stream, that we assume have real row count.
       * (Example: for Array data type, first stream is array sizes; and number of array sizes is the number of arrays).
       */
+    IDataType::SubstreamPath path;
     column_type.enumerateStreams([&](const IDataType::SubstreamPath & substream_path)
     {
         if (filename.empty())
             filename = IDataType::getFileNameForStream(column_name, substream_path);
-    },  {});
+    }, path);
 
     Files_t::const_iterator it = files.find(filename);
     if (files.end() == it)

dbms/src/Storages/StorageTinyLog.cpp

@@ -95,6 +95,10 @@ private:
     using FileStreams = std::map<std::string, std::unique_ptr<Stream>>;
     FileStreams streams;
 
+    using DeserializeState = IDataType::DeserializeBinaryBulkStatePtr;
+    using DeserializeStates = std::map<String, DeserializeState>;
+    DeserializeStates deserialize_states;
+
     void readData(const String & name, const IDataType & type, IColumn & column, size_t limit);
 };
 
@@ -149,8 +153,13 @@ private:
     using FileStreams = std::map<std::string, std::unique_ptr<Stream>>;
     FileStreams streams;
 
+    using SerializeState = IDataType::SerializeBinaryBulkStatePtr;
+    using SerializeStates = std::map<String, SerializeState>;
+    SerializeStates serialize_states;
+
     using WrittenStreams = std::set<std::string>;
 
+    IDataType::OutputStreamGetter createStreamGetter(const String & name, WrittenStreams & written_streams);
     void writeData(const String & name, const IDataType & type, const IColumn & column, WrittenStreams & written_streams);
 };
 
@@ -206,7 +215,8 @@ Block TinyLogBlockInputStream::readImpl()
 
 void TinyLogBlockInputStream::readData(const String & name, const IDataType & type, IColumn & column, size_t limit)
 {
-    IDataType::InputStreamGetter stream_getter = [&] (const IDataType::SubstreamPath & path) -> ReadBuffer *
+    IDataType::DeserializeBinaryBulkSettings settings; /// TODO Use avg_value_size_hint.
+    settings.getter = [&] (const IDataType::SubstreamPath & path) -> ReadBuffer *
     {
         String stream_name = IDataType::getFileNameForStream(name, path);
 
@@ -216,13 +226,17 @@ void TinyLogBlockInputStream::readData(const String & name, const IDataType & ty
         return &streams[stream_name]->compressed;
     };
 
-    type.deserializeBinaryBulkWithMultipleStreams(column, stream_getter, limit, 0, true, {}); /// TODO Use avg_value_size_hint.
+    if (deserialize_states.count(name) == 0)
+         type.deserializeBinaryBulkStatePrefix(settings, deserialize_states[name]);
+
+    type.deserializeBinaryBulkWithMultipleStreams(column, limit, settings, deserialize_states[name]);
 }
 
 
-void TinyLogBlockOutputStream::writeData(const String & name, const IDataType & type, const IColumn & column, WrittenStreams & written_streams)
+IDataType::OutputStreamGetter TinyLogBlockOutputStream::createStreamGetter(const String & name,
+                                                                           WrittenStreams & written_streams)
 {
-    IDataType::OutputStreamGetter stream_getter = [&] (const IDataType::SubstreamPath & path) -> WriteBuffer *
+    return [&] (const IDataType::SubstreamPath & path) -> WriteBuffer *
     {
         String stream_name = IDataType::getFileNameForStream(name, path);
 
@@ -230,12 +244,23 @@ void TinyLogBlockOutputStream::writeData(const String & name, const IDataType &
             return nullptr;
 
         if (!streams.count(stream_name))
-            streams[stream_name] = std::make_unique<Stream>(storage.files[stream_name].data_file.path(), storage.max_compress_block_size);
+            streams[stream_name] = std::make_unique<Stream>(storage.files[stream_name].data_file.path(),
+                                                            storage.max_compress_block_size);
 
         return &streams[stream_name]->compressed;
     };
+}
 
-    type.serializeBinaryBulkWithMultipleStreams(column, stream_getter, 0, 0, true, {});
+
+void TinyLogBlockOutputStream::writeData(const String & name, const IDataType & type, const IColumn & column, WrittenStreams & written_streams)
+{
+    IDataType::SerializeBinaryBulkSettings settings;
+    settings.getter = createStreamGetter(name, written_streams);
+
+    if (serialize_states.count(name) == 0)
+        type.serializeBinaryBulkStatePrefix(settings, serialize_states[name]);
+
+    type.serializeBinaryBulkWithMultipleStreams(column, 0, 0, settings, serialize_states[name]);
 }
 
 
@@ -249,6 +274,18 @@ void TinyLogBlockOutputStream::writeSuffix()
     if (streams.empty())
         return;
 
+    WrittenStreams written_streams;
+    IDataType::SerializeBinaryBulkSettings settings;
+    for (const auto & column : getHeader())
+    {
+        auto it = serialize_states.find(column.name);
+        if (it != serialize_states.end())
+        {
+            settings.getter = createStreamGetter(column.name, written_streams);
+            column.type->serializeBinaryBulkStateSuffix(settings, it->second);
+        }
+    }
+
     /// Finish write.
     for (auto & stream : streams)
         stream.second->finalize();
@@ -324,7 +361,8 @@ void StorageTinyLog::addFiles(const String & column_name, const IDataType & type
         }
     };
 
-    type.enumerateStreams(stream_callback, {});
+    IDataType::SubstreamPath path;
+    type.enumerateStreams(stream_callback, path);
 }
 
 

dbms/src/Storages/System/StorageSystemColumns.cpp

@@ -42,7 +42,11 @@ void StorageSystemColumns::fillData(MutableColumns & res_columns, const Context
         /// Add `database` column.
         MutableColumnPtr database_column_mut = ColumnString::create();
         for (const auto & database : databases)
+        {
+            if (context.hasDatabaseAccessRights(database.first))
                 database_column_mut->insert(database.first);
+        }
+
         block_to_filter.insert(ColumnWithTypeAndName(std::move(database_column_mut), std::make_shared<DataTypeString>(), "database"));
 
         /// Filter block with `database` column.

dbms/src/Storages/System/StorageSystemDatabases.cpp

@@ -21,6 +21,8 @@ void StorageSystemDatabases::fillData(MutableColumns & res_columns, const Contex
 {
     auto databases = context.getDatabases();
     for (const auto & database : databases)
+    {
+        if (context.hasDatabaseAccessRights(database.first))
         {
             res_columns[0]->insert(database.first);
             res_columns[1]->insert(database.second->getEngineName());
@@ -28,5 +30,6 @@ void StorageSystemDatabases::fillData(MutableColumns & res_columns, const Contex
             res_columns[3]->insert(database.second->getMetadataPath());
         }
     }
+}
 
 }

dbms/src/Storages/System/StorageSystemMerges.cpp

@@ -32,6 +32,8 @@ NamesAndTypesList StorageSystemMerges::getNamesAndTypes()
 void StorageSystemMerges::fillData(MutableColumns & res_columns, const Context & context, const SelectQueryInfo &) const
 {
     for (const auto & merge : context.getMergeList().get())
+    {
+        if (context.hasDatabaseAccessRights(merge.database))
         {
             size_t i = 0;
             res_columns[i++]->insert(merge.database);
@@ -52,5 +54,6 @@ void StorageSystemMerges::fillData(MutableColumns & res_columns, const Context &
             res_columns[i++]->insert(merge.thread_number);
         }
     }
+}
 
 }

dbms/src/Storages/System/StorageSystemMutations.cpp

@@ -37,6 +37,8 @@ void StorageSystemMutations::fillData(MutableColumns & res_columns, const Contex
     /// Collect a set of *MergeTree tables.
     std::map<String, std::map<String, StoragePtr>> merge_tree_tables;
     for (const auto & db : context.getDatabases())
+    {
+        if (context.hasDatabaseAccessRights(db.first))
         {
             for (auto iterator = db.second->getIterator(context); iterator->isValid(); iterator->next())
             {
@@ -47,6 +49,7 @@ void StorageSystemMutations::fillData(MutableColumns & res_columns, const Contex
                 }
             }
         }
+    }
 
     MutableColumnPtr col_database_mut = ColumnString::create();
     MutableColumnPtr col_table_mut = ColumnString::create();

dbms/src/Storages/System/StorageSystemPartsBase.cpp

@@ -59,7 +59,10 @@ public:
             /// Add column 'database'.
             MutableColumnPtr database_column_mut = ColumnString::create();
             for (const auto & database : databases)
+            {
+                if (context.hasDatabaseAccessRights(database.first))
                     database_column_mut->insert(database.first);
+            }
             block_to_filter.insert(ColumnWithTypeAndName(
                     std::move(database_column_mut), std::make_shared<DataTypeString>(), "database"));
 

dbms/src/Storages/System/StorageSystemReplicas.cpp

@@ -65,9 +65,15 @@ BlockInputStreams StorageSystemReplicas::read(
     /// We collect a set of replicated tables.
     std::map<String, std::map<String, StoragePtr>> replicated_tables;
     for (const auto & db : context.getDatabases())
+    {
+        if (context.hasDatabaseAccessRights(db.first))
+        {
             for (auto iterator = db.second->getIterator(context); iterator->isValid(); iterator->next())
                 if (dynamic_cast<const StorageReplicatedMergeTree *>(iterator->table().get()))
                     replicated_tables[db.first][iterator->name()] = iterator->table();
+        }
+    }
+
 
     /// Do you need columns that require a walkthrough in ZooKeeper to compute.
     bool with_zk_fields = false;

dbms/src/Storages/System/StorageSystemReplicationQueue.cpp

@@ -50,9 +50,15 @@ void StorageSystemReplicationQueue::fillData(MutableColumns & res_columns, const
 {
     std::map<String, std::map<String, StoragePtr>> replicated_tables;
     for (const auto & db : context.getDatabases())
+    {
+        if (context.hasDatabaseAccessRights(db.first))
+        {
             for (auto iterator = db.second->getIterator(context); iterator->isValid(); iterator->next())
                 if (dynamic_cast<const StorageReplicatedMergeTree *>(iterator->table().get()))
                     replicated_tables[db.first][iterator->name()] = iterator->table();
+        }
+    }
+
 
     MutableColumnPtr col_database_mut = ColumnString::create();
     MutableColumnPtr col_table_mut = ColumnString::create();

dbms/src/Storages/System/StorageSystemTables.cpp

@@ -95,7 +95,7 @@ BlockInputStreams StorageSystemTables::read(
 
         auto database = context.tryGetDatabase(database_name);
 
-        if (!database)
+        if (!database || !context.hasDatabaseAccessRights(database_name))
         {
             /// Database was deleted just now.
             continue;

dbms/tests/integration/helpers/client.py

@@ -11,11 +11,11 @@ class Client:
         self.command = [command, '--host', self.host, '--port', str(self.port), '--stacktrace']
 
 
-    def query(self, sql, stdin=None, timeout=None, settings=None):
-        return self.get_query_request(sql, stdin=stdin, timeout=timeout, settings=settings).get_answer()
+    def query(self, sql, stdin=None, timeout=None, settings=None, user=None):
+        return self.get_query_request(sql, stdin=stdin, timeout=timeout, settings=settings, user=user).get_answer()
 
 
-    def get_query_request(self, sql, stdin=None, timeout=None, settings=None):
+    def get_query_request(self, sql, stdin=None, timeout=None, settings=None, user=None):
         command = self.command[:]
 
         if stdin is None:
@@ -28,6 +28,9 @@ class Client:
             for setting, value in settings.iteritems():
                 command += ['--' + setting, str(value)]
 
+        if user is not None:
+            command += ['--user', user]
+
         return CommandRequest(command, stdin, timeout)
 
 

dbms/tests/integration/test_config_substitutions/configs/config_allow_databases.xml

@@ -0,0 +1,20 @@
+<yandex>
+    <users>
+        <default>
+            <password></password>
+            <profile>default</profile>
+            <quota>default</quota>
+        </default>
+        <test_allow>
+            <password></password>
+            <profile>default</profile>
+            <quota>default</quota>
+            <networks>
+                <ip>::/0</ip>
+            </networks>
+            <allow_databases>
+                <database>default</database>
+            </allow_databases>
+        </test_allow>
+    </users>
+</yandex>

dbms/tests/integration/test_config_substitutions/test.py

@@ -8,6 +8,7 @@ node1 = cluster.add_instance('node1', main_configs=['configs/config_no_substs.xm
 node2 = cluster.add_instance('node2', main_configs=['configs/config_env.xml'], env_variables={"MAX_QUERY_SIZE": "55555"})
 node3 = cluster.add_instance('node3', main_configs=['configs/config_zk.xml'], with_zookeeper=True)
 node4 = cluster.add_instance('node4', main_configs=['configs/config_incl.xml', 'configs/max_query_size.xml']) # include value 77777
+node5 = cluster.add_instance('node5', main_configs=['configs/config_allow_databases.xml'])
 
 @pytest.fixture(scope="module")
 def start_cluster():
@@ -26,3 +27,19 @@ def test_config(start_cluster):
    assert node2.query("select value from system.settings where name = 'max_query_size'") == "55555\n"
    assert node3.query("select value from system.settings where name = 'max_query_size'") == "77777\n"
    assert node4.query("select value from system.settings where name = 'max_query_size'") == "99999\n"
+
+def test_allow_databases(start_cluster):
+    node5.query("CREATE DATABASE db1")
+    node5.query("CREATE TABLE db1.test_table(date Date, k1 String, v1 Int32) ENGINE = MergeTree(date, (k1, date), 8192)")
+    node5.query("INSERT INTO db1.test_table VALUES('2000-01-01', 'test_key', 1)")
+    assert node5.query("SELECT name FROM system.databases WHERE name = 'db1'") == "db1\n"
+    assert node5.query("SELECT name FROM system.tables WHERE database = 'db1' AND name = 'test_table' ") == "test_table\n"
+    assert node5.query("SELECT name FROM system.columns WHERE database = 'db1' AND table = 'test_table'") == "date\nk1\nv1\n"
+    assert node5.query("SELECT name FROM system.parts WHERE database = 'db1' AND table = 'test_table'") == "20000101_20000101_1_1_0\n"
+    assert node5.query("SELECT name FROM system.parts_columns WHERE database = 'db1' AND table = 'test_table'") == "20000101_20000101_1_1_0\n20000101_20000101_1_1_0\n20000101_20000101_1_1_0\n"
+
+    assert node5.query("SELECT name FROM system.databases WHERE name = 'db1'", user="test_allow") == "\n"
+    assert node5.query("SELECT name FROM system.tables WHERE database = 'db1' AND name = 'test_table'", user="test_allow") == "\n"
+    assert node5.query("SELECT name FROM system.columns WHERE database = 'db1' AND table = 'test_table'", user="test_allow") == "\n"
+    assert node5.query("SELECT name FROM system.parts WHERE database = 'db1' AND table = 'test_table'", user="test_allow") == "\n"
+    assert node5.query("SELECT name FROM system.parts_columns WHERE database = 'db1' AND table = 'test_table'", user="test_allow") == "\n"

dbms/tests/queries/0_stateless/00606_quantiles_and_nans.sql

@@ -1 +1,16 @@
-SELECT DISTINCT eq FROM (WITH range(number % 10) AS arr, arrayMap(x -> x = intDiv(number, 10) ? nan : x, arr) AS arr_with_nan, arrayFilter(x -> x != intDiv(number, 10), arr) AS arr_filtered SELECT number, arrayReduce('quantileExact', arr_with_nan) AS q1, arrayReduce('quantileExact', arr_filtered) AS q2, q1 = q2 AS eq FROM numbers(100));
+SELECT DISTINCT
+    eq
+FROM
+    (
+        WITH
+            range(2 + number % 10) AS arr, -- minimum two elements, to avoid nan result --
+            arrayMap(x -> x = intDiv(number, 10) ? nan : x, arr) AS arr_with_nan,
+            arrayFilter(x -> x != intDiv(number, 10), arr) AS arr_filtered
+        SELECT
+            number,
+            arrayReduce('quantileExact', arr_with_nan) AS q1,
+            arrayReduce('quantileExact', arr_filtered) AS q2,
+            q1 = q2 AS eq
+        FROM
+            numbers(100)
+    );

dbms/tests/queries/0_stateless/00672_arrayDistinct.reference

@@ -10,3 +10,4 @@
 ['123']
 [['1212'],['sef'],['343r4']]
 [(1,2),(1,3),(1,5)]
+0

dbms/tests/queries/0_stateless/00672_arrayDistinct.sql

@@ -19,3 +19,4 @@ DROP TABLE arrayDistinct_test;
 
 SELECT arrayDistinct([['1212'], ['sef'], ['343r4'], ['1212']]);
 SELECT arrayDistinct([(1, 2), (1, 3), (1, 2), (1, 2), (1, 2), (1, 5)]);
+SELECT length(arrayDistinct([NULL, NULL, NULL]));

dbms/tests/queries/0_stateless/00687_top_and_offset.reference

@@ -0,0 +1,6 @@
+1
+2
+1
+2
+3
+4

dbms/tests/queries/0_stateless/00687_top_and_offset.sql

@@ -0,0 +1,21 @@
+DROP TABLE IF EXISTS test.test;
+
+CREATE TABLE test.test(val Int64) engine = Memory;
+
+INSERT INTO test.test VALUES (1);
+INSERT INTO test.test VALUES (2);
+INSERT INTO test.test VALUES (3);
+INSERT INTO test.test VALUES (4);
+INSERT INTO test.test VALUES (5);
+INSERT INTO test.test VALUES (6);
+INSERT INTO test.test VALUES (7);
+INSERT INTO test.test VALUES (8);
+INSERT INTO test.test VALUES (9);
+
+SELECT TOP 2 * FROM test.test ORDER BY val;
+SELECT TOP (2) * FROM test.test ORDER BY val;
+SELECT * FROM test.test ORDER BY val LIMIT 2 OFFSET 2;
+SELECT TOP 2 * FROM test.test ORDER BY val LIMIT 2; -- { clientError 406 }
+SELECT * FROM test.test ORDER BY val LIMIT 2,3 OFFSET 2; -- { clientError 62 }
+
+DROP TABLE test.test;

dbms/tests/queries/0_stateless/00688_low_cardinality_defaults.reference

@@ -0,0 +1,4 @@
+1
+1
+0
+1

dbms/tests/queries/0_stateless/00688_low_cardinality_defaults.sql

@@ -0,0 +1,4 @@
+select CAST(toLowCardinality(val) as UInt64) from (select arrayJoin(['1']) as val);
+select toUInt64(toLowCardinality(val)) from (select arrayJoin(['1']) as val);
+select 1 % toLowCardinality(val) from (select arrayJoin([1]) as val);
+select gcd(1, toLowCardinality(val)) from (select arrayJoin([1]) as val);