A small preparation for better handling of primary key in memory

src/Processors/QueryPlan/PartsSplitter.cpp

@@ -54,7 +54,7 @@ public:
 
     Values getValue(size_t part_idx, size_t mark) const
     {
-        const auto & index = parts[part_idx].data_part->index;
+        const auto & index = parts[part_idx].data_part->getIndex();
         Values values(index.size());
         for (size_t i = 0; i < values.size(); ++i)
         {

src/Storages/MergeTree/IMergeTreeDataPart.cpp

@@ -313,13 +313,13 @@ IMergeTreeDataPart::IMergeTreeDataPart(
     const IMergeTreeDataPart * parent_part_)
     : DataPartStorageHolder(data_part_storage_)
     , storage(storage_)
-    , mutable_name(name_)
     , name(mutable_name)
     , info(info_)
     , index_granularity_info(storage_, part_type_)
     , part_type(part_type_)
     , parent_part(parent_part_)
     , parent_part_name(parent_part ? parent_part->name : "")
+    , mutable_name(name_)
 {
     if (parent_part)
     {

src/Storages/MergeTree/IMergeTreeDataPart.h

@@ -75,6 +75,7 @@ public:
     using ColumnSizeByName = std::unordered_map<std::string, ColumnSize>;
     using NameToNumber = std::unordered_map<std::string, size_t>;
 
+    using Index = Columns;
     using IndexSizeByName = std::unordered_map<std::string, ColumnSize>;
 
     using Type = MergeTreeDataPartType;
@@ -212,10 +213,6 @@ public:
 
     const MergeTreeData & storage;
 
-private:
-    String mutable_name;
-    mutable MergeTreeDataPartState state{MergeTreeDataPartState::Temporary};
-
 public:
     const String & name;    // const ref to private mutable_name
     MergeTreePartInfo info;
@@ -309,12 +306,6 @@ public:
     /// Throws an exception if state of the part is not in affordable_states
     void assertState(const std::initializer_list<MergeTreeDataPartState> & affordable_states) const;
 
-    /// Primary key (correspond to primary.idx file).
-    /// Always loaded in RAM. Contains each index_granularity-th value of primary key tuple.
-    /// Note that marks (also correspond to primary key) is not always in RAM, but cached. See MarkCache.h.
-    using Index = Columns;
-    Index index;
-
     MergeTreePartition partition;
 
     /// Amount of rows between marks
@@ -369,6 +360,9 @@ public:
     /// Version of part metadata (columns, pk and so on). Managed properly only for replicated merge tree.
     int32_t metadata_version;
 
+    const Index & getIndex() const { return index; }
+    void setIndex(Columns index_) { index = std::move(index_); }
+
     /// For data in RAM ('index')
     UInt64 getIndexSizeInBytes() const;
     UInt64 getIndexSizeInAllocatedBytes() const;
@@ -567,6 +561,10 @@ public:
     mutable std::atomic<time_t> last_removal_attempt_time = 0;
 
 protected:
+    /// Primary key (correspond to primary.idx file).
+    /// Always loaded in RAM. Contains each index_granularity-th value of primary key tuple.
+    /// Note that marks (also correspond to primary key) is not always in RAM, but cached. See MarkCache.h.
+    Index index;
 
     /// Total size of all columns, calculated once in calcuateColumnSizesOnDisk
     ColumnSize total_columns_size;
@@ -623,6 +621,9 @@ protected:
     void initializeIndexGranularityInfo();
 
 private:
+    String mutable_name;
+    mutable MergeTreeDataPartState state{MergeTreeDataPartState::Temporary};
+
     /// In compact parts order of columns is necessary
     NameToNumber column_name_to_position;
 

src/Storages/MergeTree/MergeTreeData.cpp

@@ -6824,7 +6824,7 @@ Block MergeTreeData::getMinMaxCountProjectionBlock(
         {
             for (const auto & part : real_parts)
             {
-                const auto & primary_key_column = *part->index[0];
+                const auto & primary_key_column = *part->getIndex()[0];
                 auto & min_column = assert_cast<ColumnAggregateFunction &>(*partition_minmax_count_columns[pos]);
                 insert(min_column, primary_key_column[0]);
             }
@@ -6835,7 +6835,7 @@ Block MergeTreeData::getMinMaxCountProjectionBlock(
         {
             for (const auto & part : real_parts)
             {
-                const auto & primary_key_column = *part->index[0];
+                const auto & primary_key_column = *part->getIndex()[0];
                 auto & max_column = assert_cast<ColumnAggregateFunction &>(*partition_minmax_count_columns[pos]);
                 insert(max_column, primary_key_column[primary_key_column.size() - 1]);
             }

src/Storages/MergeTree/MergeTreeDataSelectExecutor.cpp

@@ -1087,7 +1087,7 @@ MarkRanges MergeTreeDataSelectExecutor::markRangesFromPKRange(
     MarkRanges res;
 
     size_t marks_count = part->index_granularity.getMarksCount();
-    const auto & index = part->index;
+    const auto & index = part->getIndex();
     if (marks_count == 0)
         return res;
 

src/Storages/MergeTree/MergedBlockOutputStream.cpp

@@ -181,7 +181,7 @@ MergedBlockOutputStream::Finalizer MergedBlockOutputStream::finalizePartAsync(
 
     new_part->rows_count = rows_count;
     new_part->modification_time = time(nullptr);
-    new_part->index = writer->releaseIndexColumns();
+    new_part->setIndex(writer->releaseIndexColumns());
     new_part->checksums = checksums;
     new_part->setBytesOnDisk(checksums.getTotalSizeOnDisk());
     new_part->setBytesUncompressedOnDisk(checksums.getTotalSizeUncompressedOnDisk());

src/Storages/MergeTree/MutateTask.cpp

@@ -899,7 +899,7 @@ void finalizeMutatedPart(
 
     new_data_part->rows_count = source_part->rows_count;
     new_data_part->index_granularity = source_part->index_granularity;
-    new_data_part->index = source_part->index;
+    new_data_part->setIndex(source_part->getIndex());
     new_data_part->minmax_idx = source_part->minmax_idx;
     new_data_part->modification_time = time(nullptr);