diff --git a/src/AggregateFunctions/AggregateFunctionAggThrow.cpp b/src/AggregateFunctions/AggregateFunctionAggThrow.cpp
index ea3eb9b1a20..fada039e20a 100644
--- a/src/AggregateFunctions/AggregateFunctionAggThrow.cpp
+++ b/src/AggregateFunctions/AggregateFunctionAggThrow.cpp
@@ -93,7 +93,7 @@ public:
         buf.read(c);
     }
 
-    void insertResultInto(AggregateDataPtr, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr, IColumn & to, Arena *) const override
     {
         to.insertDefault();
     }
diff --git a/src/AggregateFunctions/AggregateFunctionArgMinMax.h b/src/AggregateFunctions/AggregateFunctionArgMinMax.h
index 9a0c428d75b..9470b1b8692 100644
--- a/src/AggregateFunctions/AggregateFunctionArgMinMax.h
+++ b/src/AggregateFunctions/AggregateFunctionArgMinMax.h
@@ -85,7 +85,7 @@ public:
         return Data::allocatesMemoryInArena();
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         this->data(place).result.insertResultInto(to);
     }
diff --git a/src/AggregateFunctions/AggregateFunctionArray.h b/src/AggregateFunctions/AggregateFunctionArray.h
index 4fe5e459ae1..24b07010707 100644
--- a/src/AggregateFunctions/AggregateFunctionArray.h
+++ b/src/AggregateFunctions/AggregateFunctionArray.h
@@ -119,9 +119,9 @@ public:
         nested_func->deserialize(place, buf, arena);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena * arena) const override
     {
-        nested_func->insertResultInto(place, to);
+        nested_func->insertResultInto(place, to, arena);
     }
 
     bool allocatesMemoryInArena() const override
diff --git a/src/AggregateFunctions/AggregateFunctionAvg.h b/src/AggregateFunctions/AggregateFunctionAvg.h
index d9ef8647b82..1f3426160cb 100644
--- a/src/AggregateFunctions/AggregateFunctionAvg.h
+++ b/src/AggregateFunctions/AggregateFunctionAvg.h
@@ -80,7 +80,7 @@ public:
         readBinary(this->data(place).denominator, buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         auto & column = static_cast<ColVecResult &>(to);
         column.getData().push_back(this->data(place).template result<ResultType>());
diff --git a/src/AggregateFunctions/AggregateFunctionBitwise.h b/src/AggregateFunctions/AggregateFunctionBitwise.h
index a4e5f7ddafa..6d9eb3c36e1 100644
--- a/src/AggregateFunctions/AggregateFunctionBitwise.h
+++ b/src/AggregateFunctions/AggregateFunctionBitwise.h
@@ -74,7 +74,7 @@ public:
         readBinary(this->data(place).value, buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         assert_cast<ColumnVector<T> &>(to).getData().push_back(this->data(place).value);
     }
diff --git a/src/AggregateFunctions/AggregateFunctionBoundingRatio.h b/src/AggregateFunctions/AggregateFunctionBoundingRatio.h
index 81846db4bac..9ceb7976f4a 100644
--- a/src/AggregateFunctions/AggregateFunctionBoundingRatio.h
+++ b/src/AggregateFunctions/AggregateFunctionBoundingRatio.h
@@ -150,7 +150,7 @@ public:
         data(place).deserialize(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         assert_cast<ColumnFloat64 &>(to).getData().push_back(getBoundingRatio(data(place)));
     }
diff --git a/src/AggregateFunctions/AggregateFunctionCategoricalInformationValue.h b/src/AggregateFunctions/AggregateFunctionCategoricalInformationValue.h
index 1c397c26631..aa205a71c97 100644
--- a/src/AggregateFunctions/AggregateFunctionCategoricalInformationValue.h
+++ b/src/AggregateFunctions/AggregateFunctionCategoricalInformationValue.h
@@ -119,8 +119,8 @@ public:
 
     void insertResultInto(
         AggregateDataPtr place,
-        IColumn & to
-    ) const override
+        IColumn & to,
+        Arena *) const override
     {
         auto & col = static_cast<ColumnArray &>(to);
         auto & data_col = static_cast<ColumnFloat64 &>(col.getData());
diff --git a/src/AggregateFunctions/AggregateFunctionCount.h b/src/AggregateFunctions/AggregateFunctionCount.h
index feb5725d9f1..29c5de0021c 100644
--- a/src/AggregateFunctions/AggregateFunctionCount.h
+++ b/src/AggregateFunctions/AggregateFunctionCount.h
@@ -57,7 +57,7 @@ public:
         readVarUInt(data(place).count, buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         assert_cast<ColumnUInt64 &>(to).getData().push_back(data(place).count);
     }
@@ -112,7 +112,7 @@ public:
         readVarUInt(data(place).count, buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         assert_cast<ColumnUInt64 &>(to).getData().push_back(data(place).count);
     }
diff --git a/src/AggregateFunctions/AggregateFunctionDistinct.cpp b/src/AggregateFunctions/AggregateFunctionDistinct.cpp
new file mode 100644
index 00000000000..4d89e8fb199
--- /dev/null
+++ b/src/AggregateFunctions/AggregateFunctionDistinct.cpp
@@ -0,0 +1,64 @@
+#include <AggregateFunctions/AggregateFunctionDistinct.h>
+#include <AggregateFunctions/AggregateFunctionCombinatorFactory.h>
+#include <AggregateFunctions/Helpers.h>
+#include <Common/typeid_cast.h>
+#include "registerAggregateFunctions.h"
+
+namespace DB
+{
+
+namespace ErrorCodes
+{
+    extern const int NUMBER_OF_ARGUMENTS_DOESNT_MATCH;
+}
+
+class AggregateFunctionCombinatorDistinct final : public IAggregateFunctionCombinator
+{
+public:
+    String getName() const override { return "Distinct"; }
+
+    DataTypes transformArguments(const DataTypes & arguments) const override
+    {
+        if (arguments.empty())
+            throw Exception("Incorrect number of arguments for aggregate function with " + getName() + " suffix",
+                            ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
+
+        return arguments;
+    }
+
+    AggregateFunctionPtr transformAggregateFunction(
+        const AggregateFunctionPtr & nested_function,
+        const AggregateFunctionProperties &,
+        const DataTypes & arguments,
+        const Array &) const override
+    {
+        AggregateFunctionPtr res;
+        if (arguments.size() == 1)
+        {
+            res.reset(createWithNumericType<
+                AggregateFunctionDistinct,
+                AggregateFunctionDistinctSingleNumericData>(*arguments[0], nested_function, arguments));
+
+            if (res)
+                return res;
+
+            if (arguments[0]->isValueUnambiguouslyRepresentedInContiguousMemoryRegion())
+                return std::make_shared<
+                    AggregateFunctionDistinct<
+                        AggregateFunctionDistinctSingleGenericData<true>>>(nested_function, arguments);
+            else
+                return std::make_shared<
+                    AggregateFunctionDistinct<
+                        AggregateFunctionDistinctSingleGenericData<false>>>(nested_function, arguments);
+        }
+
+        return std::make_shared<AggregateFunctionDistinct<AggregateFunctionDistinctMultipleGenericData>>(nested_function, arguments);
+    }
+};
+
+void registerAggregateFunctionCombinatorDistinct(AggregateFunctionCombinatorFactory & factory)
+{
+    factory.registerCombinator(std::make_shared<AggregateFunctionCombinatorDistinct>());
+}
+
+}
diff --git a/src/AggregateFunctions/AggregateFunctionDistinct.h b/src/AggregateFunctions/AggregateFunctionDistinct.h
new file mode 100644
index 00000000000..01a9c71d94f
--- /dev/null
+++ b/src/AggregateFunctions/AggregateFunctionDistinct.h
@@ -0,0 +1,240 @@
+#pragma once
+
+#include <AggregateFunctions/IAggregateFunction.h>
+#include <AggregateFunctions/KeyHolderHelpers.h>
+#include <Columns/ColumnArray.h>
+#include <Common/assert_cast.h>
+#include <DataTypes/DataTypeArray.h>
+#include <Interpreters/AggregationCommon.h>
+#include <Common/HashTable/HashSet.h>
+#include <Common/HashTable/HashMap.h>
+#include <Common/SipHash.h>
+
+#include <Common/FieldVisitors.h>
+
+namespace DB
+{
+
+
+template <typename T>
+struct AggregateFunctionDistinctSingleNumericData
+{
+    /// When creating, the hash table must be small.
+    using Set = HashSetWithStackMemory<T, DefaultHash<T>, 4>;
+    using Self = AggregateFunctionDistinctSingleNumericData<T>;
+    Set set;
+
+    void add(const IColumn ** columns, size_t /* columns_num */, size_t row_num, Arena *)
+    {
+        const auto & vec = assert_cast<const ColumnVector<T> &>(*columns[0]).getData();
+        set.insert(vec[row_num]);
+    }
+
+    void merge(const Self & rhs, Arena *)
+    {
+        set.merge(rhs.set);
+    }
+
+    void serialize(WriteBuffer & buf) const
+    {
+        set.write(buf);
+    }
+
+    void deserialize(ReadBuffer & buf, Arena *)
+    {
+        set.read(buf);
+    }
+
+    MutableColumns getArguments(const DataTypes & argument_types) const
+    {
+        MutableColumns argument_columns;
+        argument_columns.emplace_back(argument_types[0]->createColumn());
+        for (const auto & elem : set)
+            argument_columns[0]->insert(elem.getValue());
+
+        return argument_columns;
+    }
+};
+
+struct AggregateFunctionDistinctGenericData
+{
+    /// When creating, the hash table must be small.
+    using Set = HashSetWithSavedHashWithStackMemory<StringRef, StringRefHash, 4>;
+    using Self = AggregateFunctionDistinctGenericData;
+    Set set;
+
+    void merge(const Self & rhs, Arena * arena)
+    {
+        Set::LookupResult it;
+        bool inserted;
+        for (const auto & elem : rhs.set)
+            set.emplace(ArenaKeyHolder{elem.getValue(), *arena}, it, inserted);
+    }
+
+    void serialize(WriteBuffer & buf) const
+    {
+        writeVarUInt(set.size(), buf);
+        for (const auto & elem : set)
+            writeStringBinary(elem.getValue(), buf);
+    }
+
+    void deserialize(ReadBuffer & buf, Arena * arena)
+    {
+        size_t size;
+        readVarUInt(size, buf);
+        for (size_t i = 0; i < size; ++i)
+            set.insert(readStringBinaryInto(*arena, buf));
+    }
+};
+
+template <bool is_plain_column>
+struct AggregateFunctionDistinctSingleGenericData : public AggregateFunctionDistinctGenericData
+{
+    void add(const IColumn ** columns, size_t /* columns_num */, size_t row_num, Arena * arena)
+    {
+        Set::LookupResult it;
+        bool inserted;
+        auto key_holder = getKeyHolder<is_plain_column>(*columns[0], row_num, *arena);
+        set.emplace(key_holder, it, inserted);
+    }
+
+    MutableColumns getArguments(const DataTypes & argument_types) const
+    {
+        MutableColumns argument_columns;
+        argument_columns.emplace_back(argument_types[0]->createColumn());
+        for (const auto & elem : set)
+            deserializeAndInsert<is_plain_column>(elem.getValue(), *argument_columns[0]);
+
+        return argument_columns;
+    }
+};
+
+struct AggregateFunctionDistinctMultipleGenericData : public AggregateFunctionDistinctGenericData
+{
+    void add(const IColumn ** columns, size_t columns_num, size_t row_num, Arena * arena)
+    {
+        const char * begin = nullptr;
+        StringRef value(begin, 0);
+        for (size_t i = 0; i < columns_num; ++i)
+        {
+            auto cur_ref = columns[i]->serializeValueIntoArena(row_num, *arena, begin);
+            value.data = cur_ref.data - value.size;
+            value.size += cur_ref.size;
+        }
+
+        Set::LookupResult it;
+        bool inserted;
+        auto key_holder = SerializedKeyHolder{value, *arena};
+        set.emplace(key_holder, it, inserted);
+    }
+
+    MutableColumns getArguments(const DataTypes & argument_types) const
+    {
+        MutableColumns argument_columns(argument_types.size());
+        for (size_t i = 0; i < argument_types.size(); ++i)
+            argument_columns[i] = argument_types[i]->createColumn();
+
+        for (const auto & elem : set)
+        {
+            const char * begin = elem.getValue().data;
+            for (auto & column : argument_columns)
+                begin = column->deserializeAndInsertFromArena(begin);
+        }
+
+        return argument_columns;
+    }
+};
+
+/** Adaptor for aggregate functions.
+  * Adding -Distinct suffix to aggregate function
+**/
+template <typename Data>
+class AggregateFunctionDistinct : public IAggregateFunctionDataHelper<Data, AggregateFunctionDistinct<Data>>
+{
+private:
+    static constexpr auto prefix_size = sizeof(Data);
+    AggregateFunctionPtr nested_func;
+    size_t arguments_num;
+
+    AggregateDataPtr getNestedPlace(AggregateDataPtr place) const noexcept
+    {
+        return place + prefix_size;
+    }
+
+    ConstAggregateDataPtr getNestedPlace(ConstAggregateDataPtr place) const noexcept
+    {
+        return place + prefix_size;
+    }
+
+public:
+    AggregateFunctionDistinct(AggregateFunctionPtr nested_func_, const DataTypes & arguments)
+    : IAggregateFunctionDataHelper<Data, AggregateFunctionDistinct>(arguments, nested_func_->getParameters())
+    , nested_func(nested_func_)
+    , arguments_num(arguments.size()) {}
+
+    void add(AggregateDataPtr place, const IColumn ** columns, size_t row_num, Arena * arena) const override
+    {
+        this->data(place).add(columns, arguments_num, row_num, arena);
+    }
+
+    void merge(AggregateDataPtr place, ConstAggregateDataPtr rhs, Arena * arena) const override
+    {
+        this->data(place).merge(this->data(rhs), arena);
+    }
+
+    void serialize(ConstAggregateDataPtr place, WriteBuffer & buf) const override
+    {
+        this->data(place).serialize(buf);
+    }
+
+    void deserialize(AggregateDataPtr place, ReadBuffer & buf, Arena * arena) const override
+    {
+        this->data(place).deserialize(buf, arena);
+    }
+
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena * arena) const override
+    {
+        auto arguments = this->data(place).getArguments(this->argument_types);
+        ColumnRawPtrs arguments_raw(arguments.size());
+        for (size_t i = 0; i < arguments.size(); ++i)
+            arguments_raw[i] = arguments[i].get();
+
+        assert(!arguments.empty());
+        nested_func->addBatchSinglePlace(arguments[0]->size(), getNestedPlace(place), arguments_raw.data(), arena);
+        nested_func->insertResultInto(getNestedPlace(place), to, arena);
+    }
+
+    size_t sizeOfData() const override
+    {
+        return prefix_size + nested_func->sizeOfData();
+    }
+
+    void create(AggregateDataPtr place) const override
+    {
+        new (place) Data;
+        nested_func->create(getNestedPlace(place));
+    }
+
+    void destroy(AggregateDataPtr place) const noexcept override
+    {
+        this->data(place).~Data();
+        nested_func->destroy(getNestedPlace(place));
+    }
+
+    String getName() const override
+    {
+        return nested_func->getName() + "Distinct";
+    }
+
+    DataTypePtr getReturnType() const override
+    {
+        return nested_func->getReturnType();
+    }
+
+    bool allocatesMemoryInArena() const override
+    {
+        return true;
+    }
+};
+
+}
diff --git a/src/AggregateFunctions/AggregateFunctionEntropy.h b/src/AggregateFunctions/AggregateFunctionEntropy.h
index ff233a5ac93..656aca43f60 100644
--- a/src/AggregateFunctions/AggregateFunctionEntropy.h
+++ b/src/AggregateFunctions/AggregateFunctionEntropy.h
@@ -132,7 +132,7 @@ public:
         this->data(place).deserialize(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         auto & column = assert_cast<ColumnVector<Float64> &>(to);
         column.getData().push_back(this->data(place).get());
diff --git a/src/AggregateFunctions/AggregateFunctionForEach.h b/src/AggregateFunctions/AggregateFunctionForEach.h
index 23a3487de47..19f2994d3f1 100644
--- a/src/AggregateFunctions/AggregateFunctionForEach.h
+++ b/src/AggregateFunctions/AggregateFunctionForEach.h
@@ -225,7 +225,7 @@ public:
         }
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena * arena) const override
     {
         AggregateFunctionForEachData & state = data(place);
 
@@ -236,7 +236,7 @@ public:
         char * nested_state = state.array_of_aggregate_datas;
         for (size_t i = 0; i < state.dynamic_array_size; ++i)
         {
-            nested_func->insertResultInto(nested_state, elems_to);
+            nested_func->insertResultInto(nested_state, elems_to, arena);
             nested_state += nested_size_of_data;
         }
 
diff --git a/src/AggregateFunctions/AggregateFunctionGroupArray.h b/src/AggregateFunctions/AggregateFunctionGroupArray.h
index b76efd9f6c2..f3d31eb599b 100644
--- a/src/AggregateFunctions/AggregateFunctionGroupArray.h
+++ b/src/AggregateFunctions/AggregateFunctionGroupArray.h
@@ -282,7 +282,7 @@ public:
         // if constexpr (Trait::sampler == Sampler::DETERMINATOR)
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         const auto & value = this->data(place).value;
         size_t size = value.size();
@@ -600,7 +600,7 @@ public:
         // if constexpr (Trait::sampler == Sampler::DETERMINATOR)
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         auto & column_array = assert_cast<ColumnArray &>(to);
 
@@ -815,7 +815,7 @@ public:
         data(place).last = prev;
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         auto & column_array = assert_cast<ColumnArray &>(to);
 
diff --git a/src/AggregateFunctions/AggregateFunctionGroupArrayInsertAt.h b/src/AggregateFunctions/AggregateFunctionGroupArrayInsertAt.h
index 0eec38c51a7..d84c99aec57 100644
--- a/src/AggregateFunctions/AggregateFunctionGroupArrayInsertAt.h
+++ b/src/AggregateFunctions/AggregateFunctionGroupArrayInsertAt.h
@@ -179,7 +179,7 @@ public:
         }
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         ColumnArray & to_array = assert_cast<ColumnArray &>(to);
         IColumn & to_data = to_array.getData();
diff --git a/src/AggregateFunctions/AggregateFunctionGroupArrayMoving.h b/src/AggregateFunctions/AggregateFunctionGroupArrayMoving.h
index 8f93a7eb25a..19562b37a12 100644
--- a/src/AggregateFunctions/AggregateFunctionGroupArrayMoving.h
+++ b/src/AggregateFunctions/AggregateFunctionGroupArrayMoving.h
@@ -158,7 +158,7 @@ public:
         this->data(place).sum = value.back();
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         const auto & data = this->data(place);
         size_t size = data.value.size();
diff --git a/src/AggregateFunctions/AggregateFunctionGroupBitmap.h b/src/AggregateFunctions/AggregateFunctionGroupBitmap.h
index 766479cc08d..a6470aa6943 100644
--- a/src/AggregateFunctions/AggregateFunctionGroupBitmap.h
+++ b/src/AggregateFunctions/AggregateFunctionGroupBitmap.h
@@ -48,7 +48,7 @@ public:
         this->data(place).rbs.read(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         assert_cast<ColumnVector<T> &>(to).getData().push_back(this->data(place).rbs.size());
     }
@@ -113,7 +113,7 @@ public:
         this->data(place).rbs.read(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         assert_cast<ColumnVector<T> &>(to).getData().push_back(this->data(place).rbs.size());
     }
diff --git a/src/AggregateFunctions/AggregateFunctionGroupUniqArray.h b/src/AggregateFunctions/AggregateFunctionGroupUniqArray.h
index 88b1c87f526..2ee9d0f6e1c 100644
--- a/src/AggregateFunctions/AggregateFunctionGroupUniqArray.h
+++ b/src/AggregateFunctions/AggregateFunctionGroupUniqArray.h
@@ -16,6 +16,7 @@
 #include <Common/assert_cast.h>
 
 #include <AggregateFunctions/IAggregateFunction.h>
+#include <AggregateFunctions/KeyHolderHelpers.h>
 
 #define AGGREGATE_FUNCTION_GROUP_ARRAY_UNIQ_MAX_SIZE 0xFFFFFF
 
@@ -97,7 +98,7 @@ public:
         this->data(place).value.read(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         ColumnArray & arr_to = assert_cast<ColumnArray &>(to);
         ColumnArray::Offsets & offsets_to = arr_to.getOffsets();
@@ -147,26 +148,6 @@ class AggregateFunctionGroupUniqArrayGeneric
 
     using State = AggregateFunctionGroupUniqArrayGenericData;
 
-    static auto getKeyHolder(const IColumn & column, size_t row_num, Arena & arena)
-    {
-        if constexpr (is_plain_column)
-        {
-            return ArenaKeyHolder{column.getDataAt(row_num), arena};
-        }
-        else
-        {
-            const char * begin = nullptr;
-            StringRef serialized = column.serializeValueIntoArena(row_num, arena, begin);
-            assert(serialized.data != nullptr);
-            return SerializedKeyHolder{serialized, arena};
-        }
-    }
-
-    static void deserializeAndInsert(StringRef str, IColumn & data_to)
-    {
-        return deserializeAndInsertImpl<is_plain_column>(str, data_to);
-    }
-
 public:
     AggregateFunctionGroupUniqArrayGeneric(const DataTypePtr & input_data_type_, UInt64 max_elems_ = std::numeric_limits<UInt64>::max())
         : IAggregateFunctionDataHelper<AggregateFunctionGroupUniqArrayGenericData, AggregateFunctionGroupUniqArrayGeneric<is_plain_column, Tlimit_num_elem>>({input_data_type_}, {})
@@ -215,7 +196,7 @@ public:
 
         bool inserted;
         State::Set::LookupResult it;
-        auto key_holder = getKeyHolder(*columns[0], row_num, *arena);
+        auto key_holder = getKeyHolder<is_plain_column>(*columns[0], row_num, *arena);
         set.emplace(key_holder, it, inserted);
     }
 
@@ -237,7 +218,7 @@ public:
         }
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         ColumnArray & arr_to = assert_cast<ColumnArray &>(to);
         ColumnArray::Offsets & offsets_to = arr_to.getOffsets();
@@ -247,22 +228,10 @@ public:
         offsets_to.push_back(offsets_to.back() + set.size());
 
         for (auto & elem : set)
-            deserializeAndInsert(elem.getValue(), data_to);
+            deserializeAndInsert<is_plain_column>(elem.getValue(), data_to);
     }
 };
 
-template <>
-inline void deserializeAndInsertImpl<false>(StringRef str, IColumn & data_to)
-{
-    data_to.deserializeAndInsertFromArena(str.data);
-}
-
-template <>
-inline void deserializeAndInsertImpl<true>(StringRef str, IColumn & data_to)
-{
-    data_to.insertData(str.data, str.size);
-}
-
 #undef AGGREGATE_FUNCTION_GROUP_ARRAY_UNIQ_MAX_SIZE
 
 }
diff --git a/src/AggregateFunctions/AggregateFunctionHistogram.h b/src/AggregateFunctions/AggregateFunctionHistogram.h
index 8eaa42fdac4..bc9c95ecf2a 100644
--- a/src/AggregateFunctions/AggregateFunctionHistogram.h
+++ b/src/AggregateFunctions/AggregateFunctionHistogram.h
@@ -353,7 +353,7 @@ public:
         this->data(place).read(buf, max_bins);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         auto & data = this->data(place);
 
diff --git a/src/AggregateFunctions/AggregateFunctionIf.h b/src/AggregateFunctions/AggregateFunctionIf.h
index bf4f0b24de3..f04450c9142 100644
--- a/src/AggregateFunctions/AggregateFunctionIf.h
+++ b/src/AggregateFunctions/AggregateFunctionIf.h
@@ -95,9 +95,9 @@ public:
         nested_func->deserialize(place, buf, arena);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena * arena) const override
     {
-        nested_func->insertResultInto(place, to);
+        nested_func->insertResultInto(place, to, arena);
     }
 
     bool allocatesMemoryInArena() const override
diff --git a/src/AggregateFunctions/AggregateFunctionMLMethod.h b/src/AggregateFunctions/AggregateFunctionMLMethod.h
index a11ca9032a5..8a93b66ab3b 100644
--- a/src/AggregateFunctions/AggregateFunctionMLMethod.h
+++ b/src/AggregateFunctions/AggregateFunctionMLMethod.h
@@ -388,7 +388,7 @@ public:
     /** This function is called if aggregate function without State modifier is selected in a query.
      *  Inserts all weights of the model into the column 'to', so user may use such information if needed
      */
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         this->data(place).returnWeights(to);
     }
diff --git a/src/AggregateFunctions/AggregateFunctionMaxIntersections.h b/src/AggregateFunctions/AggregateFunctionMaxIntersections.h
index 050c5fd78ea..b8a4dd63eea 100644
--- a/src/AggregateFunctions/AggregateFunctionMaxIntersections.h
+++ b/src/AggregateFunctions/AggregateFunctionMaxIntersections.h
@@ -129,7 +129,7 @@ public:
         buf.read(reinterpret_cast<char *>(value.data()), size * sizeof(value[0]));
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         Int64 current_intersections = 0;
         Int64 max_intersections = 0;
diff --git a/src/AggregateFunctions/AggregateFunctionMerge.h b/src/AggregateFunctions/AggregateFunctionMerge.h
index 51a3c11118f..066f7a762f8 100644
--- a/src/AggregateFunctions/AggregateFunctionMerge.h
+++ b/src/AggregateFunctions/AggregateFunctionMerge.h
@@ -93,9 +93,9 @@ public:
         nested_func->deserialize(place, buf, arena);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena * arena) const override
     {
-        nested_func->insertResultInto(place, to);
+        nested_func->insertResultInto(place, to, arena);
     }
 
     bool allocatesMemoryInArena() const override
diff --git a/src/AggregateFunctions/AggregateFunctionMinMaxAny.h b/src/AggregateFunctions/AggregateFunctionMinMaxAny.h
index 69504f7b249..a21a64af9a4 100644
--- a/src/AggregateFunctions/AggregateFunctionMinMaxAny.h
+++ b/src/AggregateFunctions/AggregateFunctionMinMaxAny.h
@@ -746,7 +746,7 @@ public:
         return Data::allocatesMemoryInArena();
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         this->data(place).insertResultInto(to);
     }
diff --git a/src/AggregateFunctions/AggregateFunctionNothing.h b/src/AggregateFunctions/AggregateFunctionNothing.h
index b3206f6db6e..f373b3b55b0 100644
--- a/src/AggregateFunctions/AggregateFunctionNothing.h
+++ b/src/AggregateFunctions/AggregateFunctionNothing.h
@@ -67,7 +67,7 @@ public:
     {
     }
 
-    void insertResultInto(AggregateDataPtr, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr, IColumn & to, Arena *) const override
     {
         to.insertDefault();
     }
diff --git a/src/AggregateFunctions/AggregateFunctionNull.h b/src/AggregateFunctions/AggregateFunctionNull.h
index d6f0079232c..2f2c23fdc8b 100644
--- a/src/AggregateFunctions/AggregateFunctionNull.h
+++ b/src/AggregateFunctions/AggregateFunctionNull.h
@@ -150,14 +150,14 @@ public:
         }
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena * arena) const override
     {
         if constexpr (result_is_nullable)
         {
             ColumnNullable & to_concrete = assert_cast<ColumnNullable &>(to);
             if (getFlag(place))
             {
-                nested_function->insertResultInto(nestedPlace(place), to_concrete.getNestedColumn());
+                nested_function->insertResultInto(nestedPlace(place), to_concrete.getNestedColumn(), arena);
                 to_concrete.getNullMapData().push_back(0);
             }
             else
@@ -167,7 +167,7 @@ public:
         }
         else
         {
-            nested_function->insertResultInto(nestedPlace(place), to);
+            nested_function->insertResultInto(nestedPlace(place), to, arena);
         }
     }
 
diff --git a/src/AggregateFunctions/AggregateFunctionOrFill.h b/src/AggregateFunctions/AggregateFunctionOrFill.h
index 1bbf2ea3135..333f07d5e33 100644
--- a/src/AggregateFunctions/AggregateFunctionOrFill.h
+++ b/src/AggregateFunctions/AggregateFunctionOrFill.h
@@ -148,7 +148,8 @@ public:
 
     void insertResultInto(
         AggregateDataPtr place,
-        IColumn & to) const override
+        IColumn & to,
+        Arena * arena) const override
     {
         if (place[size_of_data])
         {
@@ -157,20 +158,20 @@ public:
                 // -OrNull
 
                 if (inner_nullable)
-                    nested_function->insertResultInto(place, to);
+                    nested_function->insertResultInto(place, to, arena);
                 else
                 {
                     ColumnNullable & col = typeid_cast<ColumnNullable &>(to);
 
                     col.getNullMapColumn().insertDefault();
-                    nested_function->insertResultInto(place, col.getNestedColumn());
+                    nested_function->insertResultInto(place, col.getNestedColumn(), arena);
                 }
             }
             else
             {
                 // -OrDefault
 
-                nested_function->insertResultInto(place, to);
+                nested_function->insertResultInto(place, to, arena);
             }
         }
         else
diff --git a/src/AggregateFunctions/AggregateFunctionQuantile.h b/src/AggregateFunctions/AggregateFunctionQuantile.h
index 7bdfc13295c..536d9d5683f 100644
--- a/src/AggregateFunctions/AggregateFunctionQuantile.h
+++ b/src/AggregateFunctions/AggregateFunctionQuantile.h
@@ -138,7 +138,7 @@ public:
         this->data(place).deserialize(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         /// const_cast is required because some data structures apply finalizaton (like sorting) for obtain a result.
         auto & data = this->data(place);
diff --git a/src/AggregateFunctions/AggregateFunctionResample.h b/src/AggregateFunctions/AggregateFunctionResample.h
index 49cc312287e..043e094a688 100644
--- a/src/AggregateFunctions/AggregateFunctionResample.h
+++ b/src/AggregateFunctions/AggregateFunctionResample.h
@@ -174,13 +174,14 @@ public:
 
     void insertResultInto(
         AggregateDataPtr place,
-        IColumn & to) const override
+        IColumn & to,
+        Arena * arena) const override
     {
         auto & col = assert_cast<ColumnArray &>(to);
         auto & col_offsets = assert_cast<ColumnArray::ColumnOffsets &>(col.getOffsetsColumn());
 
         for (size_t i = 0; i < total; ++i)
-            nested_function->insertResultInto(place + i * size_of_data, col.getData());
+            nested_function->insertResultInto(place + i * size_of_data, col.getData(), arena);
 
         col_offsets.getData().push_back(col.getData().size());
     }
diff --git a/src/AggregateFunctions/AggregateFunctionRetention.h b/src/AggregateFunctions/AggregateFunctionRetention.h
index 3a76ba9f055..b742dcdf77f 100644
--- a/src/AggregateFunctions/AggregateFunctionRetention.h
+++ b/src/AggregateFunctions/AggregateFunctionRetention.h
@@ -123,7 +123,7 @@ public:
         this->data(place).deserialize(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         auto & data_to = assert_cast<ColumnUInt8 &>(assert_cast<ColumnArray &>(to).getData()).getData();
         auto & offsets_to = assert_cast<ColumnArray &>(to).getOffsets();
diff --git a/src/AggregateFunctions/AggregateFunctionSequenceMatch.h b/src/AggregateFunctions/AggregateFunctionSequenceMatch.h
index 416786f8fcb..79463e890e4 100644
--- a/src/AggregateFunctions/AggregateFunctionSequenceMatch.h
+++ b/src/AggregateFunctions/AggregateFunctionSequenceMatch.h
@@ -560,7 +560,7 @@ public:
 
     DataTypePtr getReturnType() const override { return std::make_shared<DataTypeUInt8>(); }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         this->data(place).sort();
 
@@ -588,7 +588,7 @@ public:
 
     DataTypePtr getReturnType() const override { return std::make_shared<DataTypeUInt64>(); }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         const_cast<Data &>(this->data(place)).sort();
         assert_cast<ColumnUInt64 &>(to).getData().push_back(count(place));
diff --git a/src/AggregateFunctions/AggregateFunctionSimpleLinearRegression.h b/src/AggregateFunctions/AggregateFunctionSimpleLinearRegression.h
index d1405172e27..8c029855a26 100644
--- a/src/AggregateFunctions/AggregateFunctionSimpleLinearRegression.h
+++ b/src/AggregateFunctions/AggregateFunctionSimpleLinearRegression.h
@@ -170,8 +170,8 @@ public:
 
     void insertResultInto(
         AggregateDataPtr place,
-        IColumn & to
-    ) const override
+        IColumn & to,
+        Arena *) const override
     {
         Ret k = this->data(place).getK();
         Ret b = this->data(place).getB(k);
diff --git a/src/AggregateFunctions/AggregateFunctionState.h b/src/AggregateFunctions/AggregateFunctionState.h
index 126d63573af..51a31677723 100644
--- a/src/AggregateFunctions/AggregateFunctionState.h
+++ b/src/AggregateFunctions/AggregateFunctionState.h
@@ -80,7 +80,7 @@ public:
         nested_func->deserialize(place, buf, arena);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         assert_cast<ColumnAggregateFunction &>(to).getData().push_back(place);
     }
diff --git a/src/AggregateFunctions/AggregateFunctionStatistics.h b/src/AggregateFunctions/AggregateFunctionStatistics.h
index 7f6de43f5e1..b0ff57665da 100644
--- a/src/AggregateFunctions/AggregateFunctionStatistics.h
+++ b/src/AggregateFunctions/AggregateFunctionStatistics.h
@@ -143,7 +143,7 @@ public:
         this->data(place).deserialize(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         this->data(place).publish(to);
     }
@@ -395,7 +395,7 @@ public:
         this->data(place).deserialize(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         this->data(place).publish(to);
     }
diff --git a/src/AggregateFunctions/AggregateFunctionStatisticsSimple.h b/src/AggregateFunctions/AggregateFunctionStatisticsSimple.h
index 96c07cc3d41..7962453cb35 100644
--- a/src/AggregateFunctions/AggregateFunctionStatisticsSimple.h
+++ b/src/AggregateFunctions/AggregateFunctionStatisticsSimple.h
@@ -455,7 +455,7 @@ public:
         this->data(place).read(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         const auto & data = this->data(place);
         auto & dst = static_cast<ColVecResult &>(to).getData();
diff --git a/src/AggregateFunctions/AggregateFunctionSum.h b/src/AggregateFunctions/AggregateFunctionSum.h
index 9d3d559ecee..6f921dbb78b 100644
--- a/src/AggregateFunctions/AggregateFunctionSum.h
+++ b/src/AggregateFunctions/AggregateFunctionSum.h
@@ -305,7 +305,7 @@ public:
         this->data(place).read(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         auto & column = static_cast<ColVecResult &>(to);
         column.getData().push_back(this->data(place).get());
diff --git a/src/AggregateFunctions/AggregateFunctionSumMap.h b/src/AggregateFunctions/AggregateFunctionSumMap.h
index 1f4be4e806e..ab17da1b490 100644
--- a/src/AggregateFunctions/AggregateFunctionSumMap.h
+++ b/src/AggregateFunctions/AggregateFunctionSumMap.h
@@ -246,7 +246,7 @@ public:
         }
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         // Final step does compaction of keys that have zero values, this mutates the state
         auto & merged_maps = this->data(place).merged_maps;
diff --git a/src/AggregateFunctions/AggregateFunctionTimeSeriesGroupSum.h b/src/AggregateFunctions/AggregateFunctionTimeSeriesGroupSum.h
index ad83324e483..3ec40455cf3 100644
--- a/src/AggregateFunctions/AggregateFunctionTimeSeriesGroupSum.h
+++ b/src/AggregateFunctions/AggregateFunctionTimeSeriesGroupSum.h
@@ -253,7 +253,7 @@ public:
 
     void deserialize(AggregateDataPtr place, ReadBuffer & buf, Arena *) const override { this->data(place).deserialize(buf); }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         const auto & value = this->data(place).result;
         size_t size = value.size();
diff --git a/src/AggregateFunctions/AggregateFunctionTopK.h b/src/AggregateFunctions/AggregateFunctionTopK.h
index 23eb0e7ff09..68317d0bdf0 100644
--- a/src/AggregateFunctions/AggregateFunctionTopK.h
+++ b/src/AggregateFunctions/AggregateFunctionTopK.h
@@ -79,7 +79,7 @@ public:
         set.read(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         ColumnArray & arr_to = assert_cast<ColumnArray &>(to);
         ColumnArray::Offsets & offsets_to = arr_to.getOffsets();
@@ -200,7 +200,7 @@ public:
         this->data(place).value.merge(this->data(rhs).value);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         ColumnArray & arr_to = assert_cast<ColumnArray &>(to);
         ColumnArray::Offsets & offsets_to = arr_to.getOffsets();
diff --git a/src/AggregateFunctions/AggregateFunctionUniq.h b/src/AggregateFunctions/AggregateFunctionUniq.h
index 334e809ebe7..920232ee92c 100644
--- a/src/AggregateFunctions/AggregateFunctionUniq.h
+++ b/src/AggregateFunctions/AggregateFunctionUniq.h
@@ -240,7 +240,7 @@ public:
         this->data(place).set.read(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         assert_cast<ColumnUInt64 &>(to).getData().push_back(this->data(place).set.size());
     }
@@ -294,7 +294,7 @@ public:
         this->data(place).set.read(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         assert_cast<ColumnUInt64 &>(to).getData().push_back(this->data(place).set.size());
     }
diff --git a/src/AggregateFunctions/AggregateFunctionUniqCombined.h b/src/AggregateFunctions/AggregateFunctionUniqCombined.h
index a92caa4a551..e34cc602ccd 100644
--- a/src/AggregateFunctions/AggregateFunctionUniqCombined.h
+++ b/src/AggregateFunctions/AggregateFunctionUniqCombined.h
@@ -167,7 +167,7 @@ public:
         this->data(place).set.read(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         assert_cast<ColumnUInt64 &>(to).getData().push_back(this->data(place).set.size());
     }
@@ -229,7 +229,7 @@ public:
         this->data(place).set.read(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         assert_cast<ColumnUInt64 &>(to).getData().push_back(this->data(place).set.size());
     }
diff --git a/src/AggregateFunctions/AggregateFunctionUniqUpTo.h b/src/AggregateFunctions/AggregateFunctionUniqUpTo.h
index 4c71215141c..2a48e0fb182 100644
--- a/src/AggregateFunctions/AggregateFunctionUniqUpTo.h
+++ b/src/AggregateFunctions/AggregateFunctionUniqUpTo.h
@@ -180,7 +180,7 @@ public:
         this->data(place).read(buf, threshold);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         assert_cast<ColumnUInt64 &>(to).getData().push_back(this->data(place).size());
     }
@@ -242,7 +242,7 @@ public:
         this->data(place).read(buf, threshold);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         assert_cast<ColumnUInt64 &>(to).getData().push_back(this->data(place).size());
     }
diff --git a/src/AggregateFunctions/AggregateFunctionWindowFunnel.h b/src/AggregateFunctions/AggregateFunctionWindowFunnel.h
index b5704203ade..3f41046c20e 100644
--- a/src/AggregateFunctions/AggregateFunctionWindowFunnel.h
+++ b/src/AggregateFunctions/AggregateFunctionWindowFunnel.h
@@ -280,7 +280,7 @@ public:
         this->data(place).deserialize(buf);
     }
 
-    void insertResultInto(AggregateDataPtr place, IColumn & to) const override
+    void insertResultInto(AggregateDataPtr place, IColumn & to, Arena *) const override
     {
         assert_cast<ColumnUInt8 &>(to).getData().push_back(getEventLevel(this->data(place)));
     }
diff --git a/src/AggregateFunctions/Helpers.h b/src/AggregateFunctions/Helpers.h
index 6c03d25e0b1..bc24e53a763 100644
--- a/src/AggregateFunctions/Helpers.h
+++ b/src/AggregateFunctions/Helpers.h
@@ -33,6 +33,19 @@ static IAggregateFunction * createWithNumericType(const IDataType & argument_typ
     return nullptr;
 }
 
+template <template <typename> class AggregateFunctionTemplate, template <typename> class Data, typename... TArgs>
+static IAggregateFunction * createWithNumericType(const IDataType & argument_type, TArgs && ... args)
+{
+    WhichDataType which(argument_type);
+#define DISPATCH(TYPE) \
+    if (which.idx == TypeIndex::TYPE) return new AggregateFunctionTemplate<Data<TYPE>>(std::forward<TArgs>(args)...);
+    FOR_NUMERIC_TYPES(DISPATCH)
+#undef DISPATCH
+    if (which.idx == TypeIndex::Enum8) return new AggregateFunctionTemplate<Data<Int8>>(std::forward<TArgs>(args)...);
+    if (which.idx == TypeIndex::Enum16) return new AggregateFunctionTemplate<Data<Int16>>(std::forward<TArgs>(args)...);
+    return nullptr;
+}
+
 template <template <typename, bool> class AggregateFunctionTemplate, bool bool_param, typename... TArgs>
 static IAggregateFunction * createWithNumericType(const IDataType & argument_type, TArgs && ... args)
 {
diff --git a/src/AggregateFunctions/IAggregateFunction.h b/src/AggregateFunctions/IAggregateFunction.h
index 5f4291dd21d..40d589f773d 100644
--- a/src/AggregateFunctions/IAggregateFunction.h
+++ b/src/AggregateFunctions/IAggregateFunction.h
@@ -106,7 +106,7 @@ public:
     /// Inserts results into a column.
     /// This method must be called once, from single thread.
     /// After this method was called for state, you can't do anything with state but destroy.
-    virtual void insertResultInto(AggregateDataPtr place, IColumn & to) const = 0;
+    virtual void insertResultInto(AggregateDataPtr place, IColumn & to, Arena * arena) const = 0;
 
     /// Used for machine learning methods. Predict result from trained model.
     /// Will insert result into `to` column for rows in range [offset, offset + limit).
diff --git a/src/AggregateFunctions/KeyHolderHelpers.h b/src/AggregateFunctions/KeyHolderHelpers.h
new file mode 100644
index 00000000000..999e2518f89
--- /dev/null
+++ b/src/AggregateFunctions/KeyHolderHelpers.h
@@ -0,0 +1,34 @@
+#pragma once
+
+#include <Common/HashTable/HashTableKeyHolder.h>
+#include <Columns/IColumn.h>
+
+namespace DB
+{
+
+template <bool is_plain_column = false>
+static auto getKeyHolder(const IColumn & column, size_t row_num, Arena & arena)
+{
+    if constexpr (is_plain_column)
+    {
+        return ArenaKeyHolder{column.getDataAt(row_num), arena};
+    }
+    else
+    {
+        const char * begin = nullptr;
+        StringRef serialized = column.serializeValueIntoArena(row_num, arena, begin);
+        assert(serialized.data != nullptr);
+        return SerializedKeyHolder{serialized, arena};
+    }
+}
+
+template <bool is_plain_column>
+static void deserializeAndInsert(StringRef str, IColumn & data_to)
+{
+    if constexpr (is_plain_column)
+        data_to.insertData(str.data, str.size);
+    else
+        data_to.deserializeAndInsertFromArena(str.data);
+}
+
+}
diff --git a/src/AggregateFunctions/registerAggregateFunctions.cpp b/src/AggregateFunctions/registerAggregateFunctions.cpp
index a9ab1d4f8ea..a8d0cf6e37c 100644
--- a/src/AggregateFunctions/registerAggregateFunctions.cpp
+++ b/src/AggregateFunctions/registerAggregateFunctions.cpp
@@ -58,6 +58,7 @@ void registerAggregateFunctions()
         registerAggregateFunctionCombinatorNull(factory);
         registerAggregateFunctionCombinatorOrFill(factory);
         registerAggregateFunctionCombinatorResample(factory);
+        registerAggregateFunctionCombinatorDistinct(factory);
     }
 }
 
diff --git a/src/AggregateFunctions/registerAggregateFunctions.h b/src/AggregateFunctions/registerAggregateFunctions.h
index 88cdf4a504d..981273141f9 100644
--- a/src/AggregateFunctions/registerAggregateFunctions.h
+++ b/src/AggregateFunctions/registerAggregateFunctions.h
@@ -45,6 +45,7 @@ void registerAggregateFunctionCombinatorMerge(AggregateFunctionCombinatorFactory
 void registerAggregateFunctionCombinatorNull(AggregateFunctionCombinatorFactory &);
 void registerAggregateFunctionCombinatorOrFill(AggregateFunctionCombinatorFactory &);
 void registerAggregateFunctionCombinatorResample(AggregateFunctionCombinatorFactory &);
+void registerAggregateFunctionCombinatorDistinct(AggregateFunctionCombinatorFactory &);
 
 void registerAggregateFunctions();
 
diff --git a/src/AggregateFunctions/ya.make b/src/AggregateFunctions/ya.make
index bfa32b6dd78..39f863d9471 100644
--- a/src/AggregateFunctions/ya.make
+++ b/src/AggregateFunctions/ya.make
@@ -14,6 +14,7 @@ SRCS(
     AggregateFunctionCategoricalInformationValue.cpp
     AggregateFunctionCombinatorFactory.cpp
     AggregateFunctionCount.cpp
+    AggregateFunctionDistinct.cpp
     AggregateFunctionEntropy.cpp
     AggregateFunctionFactory.cpp
     AggregateFunctionForEach.cpp
diff --git a/src/Columns/ColumnAggregateFunction.cpp b/src/Columns/ColumnAggregateFunction.cpp
index d4021b45f0e..3374d171059 100644
--- a/src/Columns/ColumnAggregateFunction.cpp
+++ b/src/Columns/ColumnAggregateFunction.cpp
@@ -135,7 +135,7 @@ MutableColumnPtr ColumnAggregateFunction::convertToValues(MutableColumnPtr colum
     res->reserve(data.size());
 
     for (auto * val : data)
-        func->insertResultInto(val, *res);
+        func->insertResultInto(val, *res, &column_aggregate_func.createOrGetArena());
 
     return res;
 }
diff --git a/src/Functions/array/arrayReduce.cpp b/src/Functions/array/arrayReduce.cpp
index 2b37965260f..82d93643e86 100644
--- a/src/Functions/array/arrayReduce.cpp
+++ b/src/Functions/array/arrayReduce.cpp
@@ -188,7 +188,7 @@ void FunctionArrayReduce::executeImpl(Block & block, const ColumnNumbers & argum
 
     for (size_t i = 0; i < input_rows_count; ++i)
         if (!res_col_aggregate_function)
-            agg_func.insertResultInto(places[i], res_col);
+            agg_func.insertResultInto(places[i], res_col, arena.get());
         else
             res_col_aggregate_function->insertFrom(places[i]);
     block.getByPosition(result).column = std::move(result_holder);
diff --git a/src/Functions/array/arrayReduceInRanges.cpp b/src/Functions/array/arrayReduceInRanges.cpp
index c3c65c4d9e5..ce2995a02d1 100644
--- a/src/Functions/array/arrayReduceInRanges.cpp
+++ b/src/Functions/array/arrayReduceInRanges.cpp
@@ -377,7 +377,7 @@ void FunctionArrayReduceInRanges::executeImpl(Block & block, const ColumnNumbers
             }
 
             if (!res_col_aggregate_function)
-                agg_func.insertResultInto(place, result_data);
+                agg_func.insertResultInto(place, result_data, arena.get());
             else
                 res_col_aggregate_function->insertFrom(place);
         }
diff --git a/src/Functions/runningAccumulate.cpp b/src/Functions/runningAccumulate.cpp
index 275259e1209..bf109654bc2 100644
--- a/src/Functions/runningAccumulate.cpp
+++ b/src/Functions/runningAccumulate.cpp
@@ -124,7 +124,7 @@ public:
             }
 
             agg_func.merge(place.data(), state_to_add, arena.get());
-            agg_func.insertResultInto(place.data(), result_column);
+            agg_func.insertResultInto(place.data(), result_column, arena.get());
 
             ++row_number;
         }
diff --git a/src/Interpreters/Aggregator.cpp b/src/Interpreters/Aggregator.cpp
index c8165632896..5f7414b774b 100644
--- a/src/Interpreters/Aggregator.cpp
+++ b/src/Interpreters/Aggregator.cpp
@@ -831,10 +831,11 @@ Block Aggregator::convertOneBucketToBlock(
             MutableColumns & key_columns,
             AggregateColumnsData & aggregate_columns,
             MutableColumns & final_aggregate_columns,
+            Arena * arena,
             bool final_)
         {
             convertToBlockImpl(method, method.data.impls[bucket],
-                key_columns, aggregate_columns, final_aggregate_columns, final_);
+                key_columns, aggregate_columns, final_aggregate_columns, arena, final_);
         });
 
     block.info.bucket_num = bucket;
@@ -992,6 +993,7 @@ void Aggregator::convertToBlockImpl(
     MutableColumns & key_columns,
     AggregateColumnsData & aggregate_columns,
     MutableColumns & final_aggregate_columns,
+    Arena * arena,
     bool final) const
 {
     if (data.empty())
@@ -1001,7 +1003,7 @@ void Aggregator::convertToBlockImpl(
         throw Exception{"Aggregate. Unexpected key columns size.", ErrorCodes::LOGICAL_ERROR};
 
     if (final)
-        convertToBlockImplFinal(method, data, key_columns, final_aggregate_columns);
+        convertToBlockImplFinal(method, data, key_columns, final_aggregate_columns, arena);
     else
         convertToBlockImplNotFinal(method, data, key_columns, aggregate_columns);
     /// In order to release memory early.
@@ -1012,7 +1014,8 @@ void Aggregator::convertToBlockImpl(
 template <typename Mapped>
 inline void Aggregator::insertAggregatesIntoColumns(
     Mapped & mapped,
-    MutableColumns & final_aggregate_columns) const
+    MutableColumns & final_aggregate_columns,
+    Arena * arena) const
 {
     /** Final values of aggregate functions are inserted to columns.
       * Then states of aggregate functions, that are not longer needed, are destroyed.
@@ -1043,7 +1046,8 @@ inline void Aggregator::insertAggregatesIntoColumns(
         for (; insert_i < params.aggregates_size; ++insert_i)
             aggregate_functions[insert_i]->insertResultInto(
                 mapped + offsets_of_aggregate_states[insert_i],
-                *final_aggregate_columns[insert_i]);
+                *final_aggregate_columns[insert_i],
+                arena);
     }
     catch (...)
     {
@@ -1080,21 +1084,22 @@ void NO_INLINE Aggregator::convertToBlockImplFinal(
     Method & method,
     Table & data,
     MutableColumns & key_columns,
-    MutableColumns & final_aggregate_columns) const
+    MutableColumns & final_aggregate_columns,
+    Arena * arena) const
 {
     if constexpr (Method::low_cardinality_optimization)
     {
         if (data.hasNullKeyData())
         {
             key_columns[0]->insertDefault();
-            insertAggregatesIntoColumns(data.getNullKeyData(), final_aggregate_columns);
+            insertAggregatesIntoColumns(data.getNullKeyData(), final_aggregate_columns, arena);
         }
     }
 
     data.forEachValue([&](const auto & key, auto & mapped)
     {
         method.insertKeyIntoColumns(key, key_columns, key_sizes);
-        insertAggregatesIntoColumns(mapped, final_aggregate_columns);
+        insertAggregatesIntoColumns(mapped, final_aggregate_columns, arena);
     });
 }
 
@@ -1183,7 +1188,7 @@ Block Aggregator::prepareBlockAndFill(
         }
     }
 
-    filler(key_columns, aggregate_columns_data, final_aggregate_columns, final);
+    filler(key_columns, aggregate_columns_data, final_aggregate_columns, data_variants.aggregates_pool, final);
 
     Block res = header.cloneEmpty();
 
@@ -1207,6 +1212,7 @@ Block Aggregator::prepareBlockAndFill(
 
     return res;
 }
+
 void Aggregator::fillAggregateColumnsWithSingleKey(
     AggregatedDataVariants & data_variants,
     MutableColumns & final_aggregate_columns)
@@ -1249,6 +1255,7 @@ Block Aggregator::prepareBlockAndFillWithoutKey(AggregatedDataVariants & data_va
         MutableColumns & key_columns,
         AggregateColumnsData & aggregate_columns,
         MutableColumns & final_aggregate_columns,
+        Arena * arena,
         bool final_)
     {
         if (data_variants.type == AggregatedDataVariants::Type::without_key || params.overflow_row)
@@ -1263,7 +1270,7 @@ Block Aggregator::prepareBlockAndFillWithoutKey(AggregatedDataVariants & data_va
             }
             else
             {
-                insertAggregatesIntoColumns(data, final_aggregate_columns);
+                insertAggregatesIntoColumns(data, final_aggregate_columns, arena);
             }
 
             if (params.overflow_row)
@@ -1291,12 +1298,13 @@ Block Aggregator::prepareBlockAndFillSingleLevel(AggregatedDataVariants & data_v
         MutableColumns & key_columns,
         AggregateColumnsData & aggregate_columns,
         MutableColumns & final_aggregate_columns,
+        Arena * arena,
         bool final_)
     {
     #define M(NAME) \
         else if (data_variants.type == AggregatedDataVariants::Type::NAME) \
             convertToBlockImpl(*data_variants.NAME, data_variants.NAME->data, \
-                key_columns, aggregate_columns, final_aggregate_columns, final_);
+                key_columns, aggregate_columns, final_aggregate_columns, arena, final_);
 
         if (false) {} // NOLINT
         APPLY_FOR_VARIANTS_SINGLE_LEVEL(M)
diff --git a/src/Interpreters/Aggregator.h b/src/Interpreters/Aggregator.h
index 0546271873b..7eb30bef787 100644
--- a/src/Interpreters/Aggregator.h
+++ b/src/Interpreters/Aggregator.h
@@ -1176,19 +1176,22 @@ protected:
         MutableColumns & key_columns,
         AggregateColumnsData & aggregate_columns,
         MutableColumns & final_aggregate_columns,
+        Arena * arena,
         bool final) const;
 
     template <typename Mapped>
     void insertAggregatesIntoColumns(
         Mapped & mapped,
-        MutableColumns & final_aggregate_columns) const;
+        MutableColumns & final_aggregate_columns,
+        Arena * arena) const;
 
     template <typename Method, typename Table>
     void convertToBlockImplFinal(
         Method & method,
         Table & data,
         MutableColumns & key_columns,
-        MutableColumns & final_aggregate_columns) const;
+        MutableColumns & final_aggregate_columns,
+        Arena * arena) const;
 
     template <typename Method, typename Table>
     void convertToBlockImplNotFinal(
diff --git a/src/Interpreters/SyntaxAnalyzer.cpp b/src/Interpreters/SyntaxAnalyzer.cpp
index b8d6d171401..5dec563eebb 100644
--- a/src/Interpreters/SyntaxAnalyzer.cpp
+++ b/src/Interpreters/SyntaxAnalyzer.cpp
@@ -89,7 +89,10 @@ struct CustomizeFunctionsData
 };
 
 char countdistinct[] = "countdistinct";
-using CustomizeFunctionsVisitor = InDepthNodeVisitor<OneTypeMatcher<CustomizeFunctionsData<countdistinct>>, true>;
+using CustomizeCountDistinctVisitor = InDepthNodeVisitor<OneTypeMatcher<CustomizeFunctionsData<countdistinct>>, true>;
+
+char countifdistinct[] = "countifdistinct";
+using CustomizeCountIfDistinctVisitor = InDepthNodeVisitor<OneTypeMatcher<CustomizeFunctionsData<countifdistinct>>, true>;
 
 char in[] = "in";
 using CustomizeInVisitor = InDepthNodeVisitor<OneTypeMatcher<CustomizeFunctionsData<in>>, true>;
@@ -103,6 +106,26 @@ using CustomizeGlobalInVisitor = InDepthNodeVisitor<OneTypeMatcher<CustomizeFunc
 char globalNotIn[] = "globalnotin";
 using CustomizeGlobalNotInVisitor = InDepthNodeVisitor<OneTypeMatcher<CustomizeFunctionsData<globalNotIn>>, true>;
 
+template <char const * func_suffix>
+struct CustomizeFunctionsSuffixData
+{
+    using TypeToVisit = ASTFunction;
+
+    const String & customized_func_suffix;
+
+    void visit(ASTFunction & func, ASTPtr &)
+    {
+        if (endsWith(Poco::toLower(func.name), func_suffix))
+        {
+            size_t prefix_len = func.name.length() - strlen(func_suffix);
+            func.name = func.name.substr(0, prefix_len) + customized_func_suffix;
+        }
+    }
+};
+
+/// Swap 'if' and 'distinct' suffixes to make execution more optimal.
+char ifDistinct[] = "ifdistinct";
+using CustomizeIfDistinctVisitor = InDepthNodeVisitor<OneTypeMatcher<CustomizeFunctionsSuffixData<ifDistinct>>, true>;
 
 /// Translate qualified names such as db.table.column, table.column, table_alias.column to names' normal form.
 /// Expand asterisks and qualified asterisks with column names.
@@ -1044,8 +1067,14 @@ SyntaxAnalyzerResultPtr SyntaxAnalyzer::analyze(
 
 void SyntaxAnalyzer::normalize(ASTPtr & query, Aliases & aliases, const Settings & settings)
 {
-    CustomizeFunctionsVisitor::Data data{settings.count_distinct_implementation};
-    CustomizeFunctionsVisitor(data).visit(query);
+    CustomizeCountDistinctVisitor::Data data_count_distinct{settings.count_distinct_implementation};
+    CustomizeCountDistinctVisitor(data_count_distinct).visit(query);
+
+    CustomizeCountIfDistinctVisitor::Data data_count_if_distinct{settings.count_distinct_implementation.toString() + "If"};
+    CustomizeCountIfDistinctVisitor(data_count_if_distinct).visit(query);
+
+    CustomizeIfDistinctVisitor::Data data_distinct_if{"DistinctIf"};
+    CustomizeIfDistinctVisitor(data_distinct_if).visit(query);
 
     if (settings.transform_null_in)
     {
diff --git a/src/Processors/Merges/Algorithms/AggregatingSortedAlgorithm.cpp b/src/Processors/Merges/Algorithms/AggregatingSortedAlgorithm.cpp
index 34e32cdfe3a..b834ed82729 100644
--- a/src/Processors/Merges/Algorithms/AggregatingSortedAlgorithm.cpp
+++ b/src/Processors/Merges/Algorithms/AggregatingSortedAlgorithm.cpp
@@ -223,7 +223,7 @@ void AggregatingSortedAlgorithm::AggregatingMergedData::finishGroup()
     /// Write the simple aggregation result for the current group.
     for (auto & desc : def.columns_to_simple_aggregate)
     {
-        desc.function->insertResultInto(desc.state.data(), *desc.column);
+        desc.function->insertResultInto(desc.state.data(), *desc.column, arena.get());
         desc.destroyState();
     }
 
diff --git a/src/Processors/Merges/Algorithms/GraphiteRollupSortedAlgorithm.cpp b/src/Processors/Merges/Algorithms/GraphiteRollupSortedAlgorithm.cpp
index e364a452797..0e704e5a05b 100644
--- a/src/Processors/Merges/Algorithms/GraphiteRollupSortedAlgorithm.cpp
+++ b/src/Processors/Merges/Algorithms/GraphiteRollupSortedAlgorithm.cpp
@@ -308,7 +308,7 @@ void GraphiteRollupSortedAlgorithm::GraphiteRollupMergedData::insertRow(
     const Graphite::AggregationPattern * aggregation_pattern = std::get<1>(current_rule);
     if (aggregate_state_created)
     {
-        aggregation_pattern->function->insertResultInto(place_for_aggregate_state.data(), *value_column);
+        aggregation_pattern->function->insertResultInto(place_for_aggregate_state.data(), *value_column, nullptr);
         aggregation_pattern->function->destroy(place_for_aggregate_state.data());
         aggregate_state_created = false;
     }
diff --git a/src/Processors/Merges/Algorithms/SummingSortedAlgorithm.cpp b/src/Processors/Merges/Algorithms/SummingSortedAlgorithm.cpp
index 78ee848c377..3833e3288fd 100644
--- a/src/Processors/Merges/Algorithms/SummingSortedAlgorithm.cpp
+++ b/src/Processors/Merges/Algorithms/SummingSortedAlgorithm.cpp
@@ -498,7 +498,7 @@ void SummingSortedAlgorithm::SummingMergedData::finishGroup()
             {
                 try
                 {
-                    desc.function->insertResultInto(desc.state.data(), *desc.merged_column);
+                    desc.function->insertResultInto(desc.state.data(), *desc.merged_column, nullptr);
 
                     /// Update zero status of current row
                     if (desc.column_numbers.size() == 1)
diff --git a/tests/performance/distinct_combinator.xml b/tests/performance/distinct_combinator.xml
new file mode 100644
index 00000000000..2261dc2f86d
--- /dev/null
+++ b/tests/performance/distinct_combinator.xml
@@ -0,0 +1,6 @@
+<test>
+    <query>SELECT x, sum(DISTINCT y) from (SELECT number % 12 AS x, number % 12321 AS y FROM numbers(10000000)) GROUP BY x</query>
+    <query>SELECT x, sum(y) from (SELECT DISTINCT number % 12 AS x, number % 12321 AS y FROM numbers(10000000)) GROUP BY x</query>
+    <query>SELECT sumIf(DISTINCT number % 10, number % 5 = 2) FROM numbers(100000000)</query>
+    <query>SELECT countIf(DISTINCT number % 10, number % 5 = 2) FROM numbers(100000000)</query>
+</test>
diff --git a/tests/queries/0_stateless/01259_combinator_distinct.reference b/tests/queries/0_stateless/01259_combinator_distinct.reference
new file mode 100644
index 00000000000..94e4f7d0d17
--- /dev/null
+++ b/tests/queries/0_stateless/01259_combinator_distinct.reference
@@ -0,0 +1,15 @@
+4999950000
+78
+[0,1,2,3,4,5,6,7,8,9,10,11,12]
+20
+0.49237
+0.49237
+15
+18
+21
+24
+27
+2
+SELECT uniqExactIf(number % 10, (number % 5) = 2)\nFROM numbers(10000)
+9
+SELECT sumDistinctIf(number % 10, (number % 5) = 2)\nFROM numbers(10000)
diff --git a/tests/queries/0_stateless/01259_combinator_distinct.sql b/tests/queries/0_stateless/01259_combinator_distinct.sql
new file mode 100644
index 00000000000..1b6f887ff70
--- /dev/null
+++ b/tests/queries/0_stateless/01259_combinator_distinct.sql
@@ -0,0 +1,15 @@
+SELECT sum(DISTINCT number) FROM numbers_mt(100000);
+SELECT sum(DISTINCT number % 13) FROM numbers_mt(100000);
+SELECT arraySort(groupArray(DISTINCT number % 13)) FROM numbers_mt(100000);
+SELECT finalizeAggregation(countState(DISTINCT toString(number % 20))) FROM numbers_mt(100000);
+SELECT round(corrStable(DISTINCT x, y), 5) FROM (SELECT number % 10 AS x, number % 5 AS y FROM numbers(1000));
+SELECT round(corrStable(x, y), 5) FROM (SELECT DISTINCT number % 10 AS x, number % 5 AS y FROM numbers(1000));
+
+SELECT sum(DISTINCT y) FROM (SELECT number % 5 AS x, number % 15 AS y FROM numbers(1000)) GROUP BY x;
+
+SET enable_debug_queries = 1;
+SELECT countIf(DISTINCT number % 10, number % 5 = 2) FROM numbers(10000);
+ANALYZE SELECT countIf(DISTINCT number % 10, number % 5 = 2) FROM numbers(10000);
+
+SELECT sumIf(DISTINCT number % 10, number % 5 = 2) FROM numbers(10000);
+ANALYZE SELECT sumIf(DISTINCT number % 10, number % 5 = 2) FROM numbers(10000);
diff --git a/tests/queries/0_stateless/01259_combinator_distinct_distributed.reference b/tests/queries/0_stateless/01259_combinator_distinct_distributed.reference
new file mode 100644
index 00000000000..096d5703292
--- /dev/null
+++ b/tests/queries/0_stateless/01259_combinator_distinct_distributed.reference
@@ -0,0 +1,4 @@
+78
+[0,1,2,3,4,5,6,7,8,9,10,11,12]
+20
+0.49237
diff --git a/tests/queries/0_stateless/01259_combinator_distinct_distributed.sql b/tests/queries/0_stateless/01259_combinator_distinct_distributed.sql
new file mode 100644
index 00000000000..f851e64dbcb
--- /dev/null
+++ b/tests/queries/0_stateless/01259_combinator_distinct_distributed.sql
@@ -0,0 +1,4 @@
+SELECT sum(DISTINCT number % 13) FROM remote('127.0.0.{1,2}', numbers_mt(100000));
+SELECT arraySort(groupArray(DISTINCT number % 13)) FROM remote('127.0.0.{1,2}', numbers_mt(100000));
+SELECT finalizeAggregation(countState(DISTINCT toString(number % 20))) FROM remote('127.0.0.{1,2}', numbers_mt(100000));
+SELECT round(corrStable(DISTINCT x, y), 5) FROM (SELECT number % 10 AS x, number % 5 AS y FROM remote('127.0.0.{1,2}', numbers(1000)));