diff --git a/dbms/src/DataStreams/MergingSortedBlockInputStream.h b/dbms/src/DataStreams/MergingSortedBlockInputStream.h
index 6521ef383ed..63d0ee9b2cc 100644
--- a/dbms/src/DataStreams/MergingSortedBlockInputStream.h
+++ b/dbms/src/DataStreams/MergingSortedBlockInputStream.h
@@ -157,7 +157,7 @@ protected:
     using QueueWithCollation = std::priority_queue<SortCursorWithCollation>;
     QueueWithCollation queue_with_collation;
 
-    /// Used in Vertical merge algorithm to gather non-PK columns (on next step)
+    /// Used in Vertical merge algorithm to gather non-PK/non-index columns (on next step)
     /// If it is not nullptr then it should be populated during execution
     WriteBuffer * out_row_sources_buf;
 
diff --git a/dbms/src/Storages/MergeTree/MergeTreeIndices.h b/dbms/src/Storages/MergeTree/MergeTreeIndices.h
index 7c7e83a743b..f47ba6a9121 100644
--- a/dbms/src/Storages/MergeTree/MergeTreeIndices.h
+++ b/dbms/src/Storages/MergeTree/MergeTreeIndices.h
@@ -64,11 +64,13 @@ public:
         ExpressionActionsPtr expr,
         const Names & columns,
         const DataTypes & data_types,
+        const Block & header,
         size_t granularity)
         : name(name)
         , expr(expr)
         , columns(columns)
         , data_types(data_types)
+        , header(header)
         , granularity(granularity) {}
 
     virtual ~MergeTreeIndex() = default;
@@ -85,6 +87,7 @@ public:
     ExpressionActionsPtr expr;
     Names columns;
     DataTypes data_types;
+    Block header;
     size_t granularity;
 };
 
diff --git a/dbms/src/Storages/MergeTree/MergeTreeMinMaxIndex.cpp b/dbms/src/Storages/MergeTree/MergeTreeMinMaxIndex.cpp
index 78195c0dea9..0970af3fa19 100644
--- a/dbms/src/Storages/MergeTree/MergeTreeMinMaxIndex.cpp
+++ b/dbms/src/Storages/MergeTree/MergeTreeMinMaxIndex.cpp
@@ -116,14 +116,14 @@ void MergeTreeMinMaxGranule::update(const Block & block, size_t * pos, size_t li
     LOG_DEBUG(log, "updated rows_read: " << rows_read);
 
     *pos += rows_read;
-};
+}
 
 
 MinMaxCondition::MinMaxCondition(
     const SelectQueryInfo &query,
     const Context &context,
     const MergeTreeMinMaxIndex &index)
-    : IndexCondition(), index(index), condition(query, context, index.columns, index.expr) {};
+    : IndexCondition(), index(index), condition(query, context, index.columns, index.expr) {}
 
 bool MinMaxCondition::alwaysUnknownOrTrue() const
 {
@@ -188,7 +188,7 @@ std::unique_ptr<MergeTreeIndex> MergeTreeMinMaxIndexCreator(
     }
 
     return std::make_unique<MergeTreeMinMaxIndex>(
-            node->name, std::move(minmax_expr), columns, data_types, node->granularity.get<size_t>());;
+            node->name, std::move(minmax_expr), columns, data_types, sample, node->granularity.get<size_t>());;
 }
 
 }
\ No newline at end of file
diff --git a/dbms/src/Storages/MergeTree/MergeTreeMinMaxIndex.h b/dbms/src/Storages/MergeTree/MergeTreeMinMaxIndex.h
index 9ccadaf1f0e..7c6fb41a3c4 100644
--- a/dbms/src/Storages/MergeTree/MergeTreeMinMaxIndex.h
+++ b/dbms/src/Storages/MergeTree/MergeTreeMinMaxIndex.h
@@ -12,6 +12,7 @@ namespace DB
 
 class MergeTreeMinMaxIndex;
 
+
 struct MergeTreeMinMaxGranule : public MergeTreeIndexGranule
 {
     explicit MergeTreeMinMaxGranule(const MergeTreeMinMaxIndex & index);
@@ -30,6 +31,7 @@ struct MergeTreeMinMaxGranule : public MergeTreeIndexGranule
     std::vector<Range> parallelogram;
 };
 
+
 class MinMaxCondition : public IndexCondition
 {
 public:
@@ -57,8 +59,9 @@ public:
         ExpressionActionsPtr expr,
         const Names & columns,
         const DataTypes & data_types,
+        const Block & header,
         size_t granularity)
-        : MergeTreeIndex(name, expr, columns, data_types, granularity) {}
+        : MergeTreeIndex(name, expr, columns, data_types, header, granularity) {}
 
     ~MergeTreeMinMaxIndex() override = default;
 
diff --git a/dbms/src/Storages/MergeTree/MergeTreeUniqueIndex.cpp b/dbms/src/Storages/MergeTree/MergeTreeUniqueIndex.cpp
new file mode 100644
index 00000000000..7d3c497e549
--- /dev/null
+++ b/dbms/src/Storages/MergeTree/MergeTreeUniqueIndex.cpp
@@ -0,0 +1,431 @@
+#include <Storages/MergeTree/MergeTreeUniqueIndex.h>
+
+#include <Interpreters/ExpressionActions.h>
+#include <Interpreters/ExpressionAnalyzer.h>
+#include <Interpreters/SyntaxAnalyzer.h>
+
+#include <Parsers/ASTIdentifier.h>
+#include <Parsers/ASTFunction.h>
+#include <Parsers/ASTLiteral.h>
+
+#include <Poco/Logger.h>
+
+
+namespace DB
+{
+
+namespace ErrorCodes
+{
+    extern const int INCORRECT_QUERY;
+}
+
+MergeTreeUniqueGranule::MergeTreeUniqueGranule(const MergeTreeUniqueIndex & index)
+        : MergeTreeIndexGranule(), index(index), set(new Set(SizeLimits{}, true))
+{
+    set->setHeader(index.header);
+}
+
+void MergeTreeUniqueGranule::serializeBinary(WriteBuffer & ostr) const
+{
+    if (empty())
+        throw Exception(
+                "Attempt to write empty unique index `" + index.name + "`", ErrorCodes::LOGICAL_ERROR);
+    Poco::Logger * log = &Poco::Logger::get("unique_idx");
+
+    LOG_DEBUG(log, "serializeBinary Granule");
+
+    const auto & columns = set->getSetElements();
+
+    const auto & size_type = DataTypePtr(std::make_shared<DataTypeUInt64>());
+
+    if (index.max_rows && size() > index.max_rows)
+    {
+        size_type->serializeBinary(0, ostr);
+        return;
+    }
+
+    size_type->serializeBinary(size(), ostr);
+
+    for (size_t i = 0; i < index.columns.size(); ++i)
+    {
+        const auto & type = index.data_types[i];
+
+        type->serializeBinaryBulk(*columns[i], ostr, 0, size());
+    }
+}
+
+void MergeTreeUniqueGranule::deserializeBinary(ReadBuffer & istr)
+{
+    Poco::Logger * log = &Poco::Logger::get("unique_idx");
+
+    LOG_DEBUG(log, "deserializeBinary Granule");
+    if (!set->empty())
+    {
+        auto new_set = std::make_unique<Set>(SizeLimits{}, true);
+        new_set->setHeader(index.header);
+        set.swap(new_set);
+    }
+
+    Block block;
+
+    Field field_rows;
+    const auto & size_type = DataTypePtr(std::make_shared<DataTypeUInt64>());
+    size_type->deserializeBinary(field_rows, istr);
+    size_t rows_to_read = field_rows.get<size_t>();
+
+    for (size_t i = 0; i < index.columns.size(); ++i)
+    {
+        const auto & type = index.data_types[i];
+
+        auto new_column = type->createColumn();
+        type->deserializeBinaryBulk(*new_column, istr, rows_to_read, 0);
+
+        block.insert(ColumnWithTypeAndName(new_column->getPtr(), type, index.columns[i]));
+    }
+
+    set->insertFromBlock(block);
+}
+
+String MergeTreeUniqueGranule::toString() const
+{
+    String res = "unique granule:\n";
+
+    const auto & columns = set->getSetElements();
+
+    for (size_t i = 0; i < index.columns.size(); ++i)
+    {
+        const auto & column = columns[i];
+        res += " [";
+        for (size_t j = 0; j < column->size(); ++j)
+        {
+            if (j != 0)
+                res += ", ";
+            Field field;
+            column->get(j, field);
+            res += applyVisitor(FieldVisitorToString(), field);
+        }
+        res += "]\n";
+    }
+
+    return res;
+}
+
+void MergeTreeUniqueGranule::update(const Block & new_block, size_t * pos, size_t limit)
+{
+    Poco::Logger * log = &Poco::Logger::get("unique_idx");
+
+    LOG_DEBUG(log, "update Granule " << new_block.columns()
+                                     << " pos: "<< *pos << " limit: " << limit << " rows: " << new_block.rows());
+    size_t rows_read = std::min(limit, new_block.rows() - *pos);
+
+    if (index.max_rows && size() > index.max_rows)
+    {
+        *pos += rows_read;
+        return;
+    }
+
+    Block key_block;
+    for (size_t i = 0; i < index.columns.size(); ++i)
+    {
+        const auto & name = index.columns[i];
+        const auto & type = index.data_types[i];
+        key_block.insert(
+                ColumnWithTypeAndName(
+                    new_block.getByName(name).column->cut(*pos, rows_read),
+                    type,
+                    name));
+    }
+
+    set->insertFromBlock(key_block);
+
+    LOG_DEBUG(log, "unique rows: " << set->getTotalRowCount());
+
+    *pos += rows_read;
+}
+
+Block MergeTreeUniqueGranule::getElementsBlock() const
+{
+    if (index.max_rows && size() > index.max_rows)
+        return index.header;
+    return index.header.cloneWithColumns(set->getSetElements());
+}
+
+
+UniqueCondition::UniqueCondition(
+        const SelectQueryInfo & query,
+        const Context & context,
+        const MergeTreeUniqueIndex &index)
+        : IndexCondition(), index(index)
+{
+    for (size_t i = 0, size = index.columns.size(); i < size; ++i)
+    {
+        std::string name = index.columns[i];
+        if (!key_columns.count(name))
+            key_columns.insert(name);
+    }
+
+    const ASTSelectQuery & select = typeid_cast<const ASTSelectQuery &>(*query.query);
+
+    /// Replace logical functions with bit functions.
+    /// Working with UInt8: last bit = can be true, previous = can be false.
+    ASTPtr new_expression;
+    if (select.where_expression && select.prewhere_expression)
+        new_expression = makeASTFunction(
+                "and",
+                select.where_expression->clone(),
+                select.prewhere_expression->clone());
+    else if (select.where_expression)
+        new_expression = select.where_expression->clone();
+    else if (select.prewhere_expression)
+        new_expression = select.prewhere_expression->clone();
+    else
+        /// 11_2 -- can be true and false at the same time
+        new_expression = std::make_shared<ASTLiteral>(Field(3));
+
+    useless = checkASTAlwaysUnknownOrTrue(new_expression);
+    /// Do not proceed if index is useless for this query.
+    if (useless)
+        return;
+
+    expression_ast = makeASTFunction(
+            "bitAnd",
+            new_expression,
+            std::make_shared<ASTLiteral>(Field(1)));
+
+    traverseAST(expression_ast);
+
+    auto syntax_analyzer_result = SyntaxAnalyzer(context, {}).analyze(
+            expression_ast, index.header.getNamesAndTypesList());
+    actions = ExpressionAnalyzer(expression_ast, syntax_analyzer_result, context).getActions(true);
+}
+
+bool UniqueCondition::alwaysUnknownOrTrue() const
+{
+    return useless;
+}
+
+bool UniqueCondition::mayBeTrueOnGranule(MergeTreeIndexGranulePtr idx_granule) const
+{
+    auto granule = std::dynamic_pointer_cast<MergeTreeUniqueGranule>(idx_granule);
+    if (!granule)
+        throw Exception(
+                "Unique index condition got wrong granule", ErrorCodes::LOGICAL_ERROR);
+
+    if (useless)
+        return true;
+
+    if (index.max_rows && granule->size() > index.max_rows)
+        return true;
+
+    Block result = granule->getElementsBlock();
+    actions->execute(result);
+
+
+    const auto & column = result.getByName(expression_ast->getColumnName()).column;
+
+    for (size_t i = 0; i < column->size(); ++i)
+        if (column->getBool(i))
+            return true;
+
+    return false;
+}
+
+void UniqueCondition::traverseAST(ASTPtr & node) const
+{
+    if (operatorFromAST(node))
+    {
+        auto * func = typeid_cast<ASTFunction *>(&*node);
+        auto & args = typeid_cast<ASTExpressionList &>(*func->arguments).children;
+
+        for (auto & arg : args)
+            traverseAST(arg);
+        return;
+    }
+
+    if (!atomFromAST(node))
+        node = std::make_shared<ASTLiteral>(Field(3)); /// can_be_true=1 can_be_false=1
+}
+
+bool UniqueCondition::atomFromAST(ASTPtr & node) const
+{
+    /// Function, literal or column
+
+    if (typeid_cast<const ASTLiteral *>(node.get()))
+        return true;
+
+    if (const auto * identifier = typeid_cast<const ASTIdentifier *>(node.get()))
+        return key_columns.count(identifier->getColumnName()) != 0;
+
+    if (auto * func = typeid_cast<ASTFunction *>(node.get()))
+    {
+        if (key_columns.count(func->getColumnName()))
+        {
+            /// Function is already calculated.
+            node = std::make_shared<ASTIdentifier>(func->getColumnName());
+            return true;
+        }
+
+        ASTs & args = typeid_cast<ASTExpressionList &>(*func->arguments).children;
+
+        for (auto & arg : args)
+            if (!atomFromAST(arg))
+                return false;
+
+        return true;
+    }
+
+    return false;
+}
+
+bool UniqueCondition::operatorFromAST(ASTPtr & node) const
+{
+    /// Functions AND, OR, NOT. Replace with bit*.
+    auto * func = typeid_cast<ASTFunction *>(&*node);
+    if (!func)
+        return false;
+
+    const ASTs & args = typeid_cast<const ASTExpressionList &>(*func->arguments).children;
+
+    if (func->name == "not")
+    {
+        if (args.size() != 1)
+            return false;
+
+        const auto one = std::make_shared<ASTLiteral>(Field(1));
+        const auto two = std::make_shared<ASTLiteral>(Field(2));
+
+        node = makeASTFunction(
+                "bitOr",
+                makeASTFunction(
+                        "bitShiftLeft",
+                        makeASTFunction(
+                                "bitAnd",
+                                node->clone(),
+                                one->clone()),
+                        one->clone()),
+                makeASTFunction(
+                        "bitShiftRight",
+                        makeASTFunction(
+                                "bitAnd",
+                                node->clone(),
+                                two->clone()),
+                        one->clone()));
+    }
+    else if (func->name == "and" || func->name == "indexHint")
+        func->name = "bitAnd";
+    else if (func->name == "or")
+        func->name = "bitOr";
+    else
+        return false;
+
+    return true;
+}
+
+bool checkAtomName(const String & name)
+{
+    static std::set<String> atoms = {
+            "notEquals",
+            "equals",
+            "less",
+            "greater",
+            "lessOrEquals",
+            "greaterOrEquals",
+            "in",
+            "notIn",
+            "like"
+            };
+    return atoms.find(name) != atoms.end();
+}
+
+bool UniqueCondition::checkASTAlwaysUnknownOrTrue(const ASTPtr & node, bool atomic) const
+{
+    if (const auto * func = typeid_cast<const ASTFunction *>(node.get()))
+    {
+        if (key_columns.count(func->getColumnName()))
+            return false;
+
+        const ASTs & args = typeid_cast<const ASTExpressionList &>(*func->arguments).children;
+
+        if (func->name == "and" || func->name == "indexHint")
+            return checkASTAlwaysUnknownOrTrue(args[0], atomic) && checkASTAlwaysUnknownOrTrue(args[1], atomic);
+        else if (func->name == "or")
+            return checkASTAlwaysUnknownOrTrue(args[0], atomic) || checkASTAlwaysUnknownOrTrue(args[1], atomic);
+        else if (func->name == "not")
+            return checkASTAlwaysUnknownOrTrue(args[0], atomic);
+        else if (!atomic && checkAtomName(func->name))
+            return checkASTAlwaysUnknownOrTrue(node, true);
+        else
+            return std::any_of(args.begin(), args.end(),
+                    [this, &atomic](const auto & arg) { return checkASTAlwaysUnknownOrTrue(arg, atomic); });
+    }
+    else if (const auto * literal = typeid_cast<const ASTLiteral *>(node.get()))
+        return !atomic && literal->value.get<bool>();
+    else if (const auto * identifier = typeid_cast<const ASTIdentifier *>(node.get()))
+        return key_columns.find(identifier->getColumnName()) == key_columns.end();
+    else
+        return true;
+}
+
+
+MergeTreeIndexGranulePtr MergeTreeUniqueIndex::createIndexGranule() const
+{
+    return std::make_shared<MergeTreeUniqueGranule>(*this);
+}
+
+IndexConditionPtr MergeTreeUniqueIndex::createIndexCondition(
+        const SelectQueryInfo & query, const Context & context) const
+{
+    return std::make_shared<UniqueCondition>(query, context, *this);
+};
+
+
+std::unique_ptr<MergeTreeIndex> MergeTreeUniqueIndexCreator(
+        const MergeTreeData & data,
+        std::shared_ptr<ASTIndexDeclaration> node,
+        const Context & context)
+{
+    if (node->name.empty())
+        throw Exception("Index must have unique name", ErrorCodes::INCORRECT_QUERY);
+
+    size_t max_rows = 0;
+    if (node->type->arguments)
+    {
+        if (node->type->arguments->children.size() > 1)
+            throw Exception("Unique index cannot have only 0 or 1 argument", ErrorCodes::INCORRECT_QUERY);
+        else if (node->type->arguments->children.size() == 1)
+            max_rows = typeid_cast<const ASTLiteral &>(
+                    *node->type->arguments->children[0]).value.get<size_t>();
+    }
+
+
+    ASTPtr expr_list = MergeTreeData::extractKeyExpressionList(node->expr->clone());
+    auto syntax = SyntaxAnalyzer(context, {}).analyze(
+            expr_list, data.getColumns().getAllPhysical());
+    auto unique_expr = ExpressionAnalyzer(expr_list, syntax, context).getActions(false);
+
+    auto sample = ExpressionAnalyzer(expr_list, syntax, context)
+            .getActions(true)->getSampleBlock();
+
+    Block header;
+
+    Names columns;
+    DataTypes data_types;
+
+    Poco::Logger * log = &Poco::Logger::get("unique_idx");
+    LOG_DEBUG(log, "new unique index" << node->name);
+    for (size_t i = 0; i < expr_list->children.size(); ++i)
+    {
+        const auto & column = sample.getByPosition(i);
+
+        columns.emplace_back(column.name);
+        data_types.emplace_back(column.type);
+
+        header.insert(ColumnWithTypeAndName(column.type->createColumn(), column.type, column.name));
+        LOG_DEBUG(log, ">" << column.name << " " << column.type->getName());
+    }
+
+    return std::make_unique<MergeTreeUniqueIndex>(
+            node->name, std::move(unique_expr), columns, data_types, header, node->granularity.get<size_t>(), max_rows);;
+}
+
+}
diff --git a/dbms/src/Storages/MergeTree/MergeTreeUniqueIndex.h b/dbms/src/Storages/MergeTree/MergeTreeUniqueIndex.h
new file mode 100644
index 00000000000..c73fb2643a8
--- /dev/null
+++ b/dbms/src/Storages/MergeTree/MergeTreeUniqueIndex.h
@@ -0,0 +1,93 @@
+#pragma once
+
+#include <Storages/MergeTree/MergeTreeIndices.h>
+#include <Storages/MergeTree/MergeTreeData.h>
+
+#include <Interpreters/Set.h>
+
+#include <memory>
+#include <set>
+
+
+namespace DB
+{
+
+class MergeTreeUniqueIndex;
+
+struct MergeTreeUniqueGranule : public MergeTreeIndexGranule
+{
+    explicit MergeTreeUniqueGranule(const MergeTreeUniqueIndex & index);
+
+    void serializeBinary(WriteBuffer & ostr) const override;
+    void deserializeBinary(ReadBuffer & istr) override;
+
+    String toString() const override;
+    size_t size() const { return set->getTotalRowCount(); }
+    bool empty() const override { return !size(); }
+
+    void update(const Block & block, size_t * pos, size_t limit) override;
+    Block getElementsBlock() const;
+
+    ~MergeTreeUniqueGranule() override = default;
+
+    const MergeTreeUniqueIndex & index;
+    std::unique_ptr<Set> set;
+};
+
+
+class UniqueCondition : public IndexCondition
+{
+public:
+    UniqueCondition(
+            const SelectQueryInfo & query,
+            const Context & context,
+            const MergeTreeUniqueIndex & index);
+
+    bool alwaysUnknownOrTrue() const override;
+
+    bool mayBeTrueOnGranule(MergeTreeIndexGranulePtr idx_granule) const override;
+
+    ~UniqueCondition() override = default;
+private:
+    void traverseAST(ASTPtr & node) const;
+    bool atomFromAST(ASTPtr & node) const;
+    bool operatorFromAST(ASTPtr & node) const;
+
+    bool checkASTAlwaysUnknownOrTrue(const ASTPtr & node, bool atomic = false) const;
+
+    const MergeTreeUniqueIndex & index;
+
+    bool useless;
+    std::set<String> key_columns;
+    ASTPtr expression_ast;
+    ExpressionActionsPtr actions;
+};
+
+
+class MergeTreeUniqueIndex : public MergeTreeIndex
+{
+public:
+    MergeTreeUniqueIndex(
+            String name,
+            ExpressionActionsPtr expr,
+            const Names & columns,
+            const DataTypes & data_types,
+            const Block & header,
+            size_t granularity,
+            size_t _max_rows)
+            : MergeTreeIndex(std::move(name), std::move(expr), columns, data_types, header, granularity), max_rows(_max_rows) {}
+
+    ~MergeTreeUniqueIndex() override = default;
+
+    MergeTreeIndexGranulePtr createIndexGranule() const override;
+
+    IndexConditionPtr createIndexCondition(
+            const SelectQueryInfo & query, const Context & context) const override;
+
+    size_t max_rows = 0;
+};
+
+std::unique_ptr<MergeTreeIndex> MergeTreeUniqueIndexCreator(
+        const MergeTreeData & data, std::shared_ptr<ASTIndexDeclaration> node, const Context & context);
+
+}
diff --git a/dbms/src/Storages/MergeTree/registerStorageMergeTree.cpp b/dbms/src/Storages/MergeTree/registerStorageMergeTree.cpp
index ebaba506864..ed0a859e495 100644
--- a/dbms/src/Storages/MergeTree/registerStorageMergeTree.cpp
+++ b/dbms/src/Storages/MergeTree/registerStorageMergeTree.cpp
@@ -3,6 +3,7 @@
 #include <Storages/StorageReplicatedMergeTree.h>
 #include <Storages/MergeTree/MergeTreeIndices.h>
 #include <Storages/MergeTree/MergeTreeMinMaxIndex.h>
+#include <Storages/MergeTree/MergeTreeUniqueIndex.h>
 
 #include <Common/typeid_cast.h>
 #include <Common/OptimizedRegularExpression.h>
@@ -627,6 +628,7 @@ static void registerMergeTreeSkipIndices()
 {
     auto & factory = MergeTreeIndexFactory::instance();
     factory.registerIndex("minmax", MergeTreeMinMaxIndexCreator);
+    factory.registerIndex("unique", MergeTreeUniqueIndexCreator);
 }
 
 
diff --git a/dbms/tests/queries/0_stateless/00825_minmax_index.sql b/dbms/tests/queries/0_stateless/00825_minmax_index.sql
index fa233a15819..39271dc18e6 100644
--- a/dbms/tests/queries/0_stateless/00825_minmax_index.sql
+++ b/dbms/tests/queries/0_stateless/00825_minmax_index.sql
@@ -9,12 +9,9 @@ CREATE TABLE test.minmax_idx
     s String,
     e Enum8('a' = 1, 'b' = 2, 'c' = 3),
     dt Date,
-    INDEX
-      idx_all (i32, i32 + f64, d, s, e, dt) TYPE minmax GRANULARITY 4,
-    INDEX
-      idx_all2 (i32, i32 + f64, d, s, e, dt) TYPE minmax GRANULARITY 2,
-    INDEX
-      idx_2 (u64 + toYear(dt), substring(s, 2, 4)) TYPE minmax GRANULARITY 3
+    INDEX idx_all (i32, i32 + f64, d, s, e, dt) TYPE minmax GRANULARITY 4,
+    INDEX idx_all2 (i32, i32 + f64, d, s, e, dt) TYPE minmax GRANULARITY 2,
+    INDEX idx_2 (u64 + toYear(dt), substring(s, 2, 4)) TYPE minmax GRANULARITY 3
 ) ENGINE = MergeTree()
 ORDER BY u64
 SETTINGS index_granularity = 2;
diff --git a/dbms/tests/queries/0_stateless/00832_unique_index.reference b/dbms/tests/queries/0_stateless/00832_unique_index.reference
new file mode 100644
index 00000000000..046f6502f65
--- /dev/null
+++ b/dbms/tests/queries/0_stateless/00832_unique_index.reference
@@ -0,0 +1,8 @@
+0	5	4.7	6.50	cba	b	2014-01-04
+0	5	4.7	6.50	cba	b	2014-03-11
+2	5	4.7	6.50	cba	b	2014-06-11
+2	5	4.7	6.50	cba	b	2015-01-01
+0	5	4.7	6.50	cba	b	2014-01-04
+0	5	4.7	6.50	cba	b	2014-03-11
+2	5	4.7	6.50	cba	b	2014-06-11
+2	5	4.7	6.50	cba	b	2015-01-01
diff --git a/dbms/tests/queries/0_stateless/00832_unique_index.sql b/dbms/tests/queries/0_stateless/00832_unique_index.sql
new file mode 100644
index 00000000000..65568255c86
--- /dev/null
+++ b/dbms/tests/queries/0_stateless/00832_unique_index.sql
@@ -0,0 +1,41 @@
+DROP TABLE IF EXISTS test.unique_idx;
+
+CREATE TABLE test.unique_idx
+(
+    u64 UInt64,
+    i32 Int32,
+    f64 Float64,
+    d Decimal(10, 2),
+    s String,
+    e Enum8('a' = 1, 'b' = 2, 'c' = 3),
+    dt Date,
+    INDEX idx_all (i32, i32 + f64, d, s, e, dt) TYPE unique GRANULARITY 4,
+    INDEX idx_all2 (i32, i32 + f64, d, s, e, dt) TYPE unique GRANULARITY 2,
+    INDEX idx_2 (u64 + toYear(dt), substring(s, 2, 4)) TYPE unique GRANULARITY 3
+) ENGINE = MergeTree()
+ORDER BY u64
+SETTINGS index_granularity = 2;
+
+
+/* many small inserts => table will make merges */
+INSERT INTO test.unique_idx VALUES (1, 2, 4.5, 2.5, 'abc', 'a', '2014-01-01');
+INSERT INTO test.unique_idx VALUES (0, 5, 4.7, 6.5, 'cba', 'b', '2014-01-04');
+INSERT INTO test.unique_idx VALUES (1, 5, 6.9, 1.57, 'bac', 'c', '2017-01-01');
+INSERT INTO test.unique_idx VALUES (1, 2, 4.5, 2.5, 'abc', 'a', '2016-01-01');
+INSERT INTO test.unique_idx VALUES (2, 5, 4.7, 6.5, 'cba', 'b', '2015-01-01');
+INSERT INTO test.unique_idx VALUES (1, 5, 6.9, 1.57, 'bac', 'c', '2014-11-11');
+
+INSERT INTO test.unique_idx VALUES (1, 2, 4.5, 2.5, 'abc', 'a', '2014-02-11');
+INSERT INTO test.unique_idx VALUES (0, 5, 4.7, 6.5, 'cba', 'b', '2014-03-11');
+INSERT INTO test.unique_idx VALUES (1, 5, 6.9, 1.57, 'bac', 'c', '2014-04-11');
+INSERT INTO test.unique_idx VALUES (1, 2, 4.5, 2.5, 'abc', 'a', '2014-05-11');
+INSERT INTO test.unique_idx VALUES (2, 5, 4.7, 6.5, 'cba', 'b', '2014-06-11');
+INSERT INTO test.unique_idx VALUES (1, 5, 6.9, 1.57, 'bac', 'c', '2014-07-11');
+
+/* simple select */
+SELECT * FROM test.unique_idx WHERE i32 = 5 AND i32 + f64 < 12 AND 3 < d AND d < 7 AND (s = 'bac' OR s = 'cba') ORDER BY dt;
+
+/* select with hole made by primary key */
+SELECT * FROM test.unique_idx WHERE u64 != 1 AND e = 'b' ORDER BY dt;
+
+DROP TABLE test.unique_idx;
\ No newline at end of file
diff --git a/docs/en/operations/table_engines/mergetree.md b/docs/en/operations/table_engines/mergetree.md
index 52d34ea059b..473f2109e99 100644
--- a/docs/en/operations/table_engines/mergetree.md
+++ b/docs/en/operations/table_engines/mergetree.md
@@ -250,7 +250,7 @@ CREATE TABLE table_name
     s String,
     ...
     INDEX a (u64 * i32, s) TYPE minmax GRANULARITY 3,
-    INDEX b (u64 * length(s)) TYPE minmax GRANULARITY 4
+    INDEX b (u64 * length(s)) TYPE unique GRANULARITY 4
 ) ENGINE = MergeTree()
 ...
 ```
@@ -263,10 +263,17 @@ SELECT count() FROM table WHERE u64 * i32 == 10 AND u64 * length(s) >= 1234
 
 #### Available Types of Indices
 
-* `minmax` Stores extremes of specified expression (if the expression is `tuple`, then it stores extremes for each element of `tuple`), uses stored info for skipping blocks of data like primary key.
+* `minmax`
+Stores extremes of the specified expression (if the expression is `tuple`, then it stores extremes for each element of `tuple`), uses stored info for skipping blocks of the data like the primary key.
+
+* `unique(max_rows)`
+Stores unique values of the specified expression (no more than `max_rows` rows), use them to check if the `WHERE` expression is not satisfiable on a block of the data.
+If `max_rows=0`, then there are no limits for storing values. `unique` without parameters is equal to `unique(0)`.  
 
 ```sql
 INDEX sample_index (u64 * length(s)) TYPE minmax GRANULARITY 4
+INDEX b (u64 * length(str), i32 + f64 * 100, date, str) TYPE unique GRANULARITY 4
+INDEX b (u64 * length(str), i32 + f64 * 100, date, str) TYPE unique(100) GRANULARITY 4
 ```
 
 
diff --git a/docs/ru/operations/table_engines/mergetree.md b/docs/ru/operations/table_engines/mergetree.md
index 0f949319de1..7318e4d8145 100644
--- a/docs/ru/operations/table_engines/mergetree.md
+++ b/docs/ru/operations/table_engines/mergetree.md
@@ -241,7 +241,7 @@ CREATE TABLE table_name
     s String,
     ...
     INDEX a (u64 * i32, s) TYPE minmax GRANULARITY 3,
-    INDEX b (u64 * length(s)) TYPE minmax GRANULARITY 4
+    INDEX b (u64 * length(s), i32) TYPE unique GRANULARITY 4
 ) ENGINE = MergeTree()
 ...
 ```
@@ -254,13 +254,22 @@ SELECT count() FROM table WHERE u64 * i32 == 10 AND u64 * length(s) >= 1234
 
 #### Доступные индексы
 
-* `minmax` Хранит минимум и максимум выражения (если выражение - `tuple`, то для каждого элемента `tuple`), используя их для пропуска кусков аналогично первичному ключу.
+* `minmax`
+Хранит минимум и максимум выражения (если выражение - `tuple`, то для каждого элемента `tuple`), используя их для пропуска блоков аналогично первичному ключу.
 
-Пример
+* `unique(max_rows)`
+Хранит уникальные значения выражения на блоке в количестве не более `max_rows`, используя их для пропуска блоков, оценивая выполнимость `WHERE` выражения на хранимых данных.
+Если `max_rows=0`, то хранит значения выражения без ограничений. Если параметров не передано, то полагается `max_rows=0`.
+
+
+Примеры
 ```sql
 INDEX b (u64 * length(str), i32 + f64 * 100, date, str) TYPE minmax GRANULARITY 4
+INDEX b (u64 * length(str), i32 + f64 * 100, date, str) TYPE unique GRANULARITY 4
+INDEX b (u64 * length(str), i32 + f64 * 100, date, str) TYPE unique(100) GRANULARITY 4
 ```
 
+
 ## Конкурентный доступ к данным
 
 Для конкурентного доступа к таблице используется мультиверсионность. То есть, при одновременном чтении и обновлении таблицы, данные будут читаться из набора кусочков, актуального на момент запроса. Длинных блокировок нет. Вставки никак не мешают чтениям.