Merge pull request #55683 from amosbird/issue-55653

Reuse granule during skip index reading

src/Interpreters/BloomFilter.cpp

@@ -41,8 +41,15 @@ BloomFilter::BloomFilter(const BloomFilterParameters & params)
 BloomFilter::BloomFilter(size_t size_, size_t hashes_, size_t seed_)
     : size(size_), hashes(hashes_), seed(seed_), words((size + sizeof(UnderType) - 1) / sizeof(UnderType)), filter(words, 0)
 {
-    assert(size != 0);
-    assert(hashes != 0);
+    chassert(size != 0);
+    chassert(hashes != 0);
+}
+
+void BloomFilter::resize(size_t size_)
+{
+    size = size_;
+    words = ((size + sizeof(UnderType) - 1) / sizeof(UnderType));
+    filter.resize(words);
 }
 
 bool BloomFilter::find(const char * data, size_t len)

src/Interpreters/BloomFilter.h

@@ -37,6 +37,7 @@ class BloomFilter
     /// seed -- random seed for hash functions generation.
     BloomFilter(size_t size_, size_t hashes_, size_t seed_);
 
+    void resize(size_t size_);
     bool find(const char * data, size_t len);
     void add(const char * data, size_t len);
     void clear();

src/Storages/MergeTree/MergeTreeDataSelectExecutor.cpp

@@ -1676,7 +1676,7 @@ MarkRanges MergeTreeDataSelectExecutor::filterMarksUsingIndex(
         for (size_t index_mark = index_range.begin; index_mark < index_range.end; ++index_mark)
         {
             if (index_mark != index_range.begin || !granule || last_index_mark != index_range.begin)
-                granule = reader.read();
+                reader.read(granule);
 
             auto ann_condition = std::dynamic_pointer_cast<IMergeTreeIndexConditionApproximateNearestNeighbor>(condition);
             if (ann_condition != nullptr)
@@ -1794,7 +1794,7 @@ MarkRanges MergeTreeDataSelectExecutor::filterMarksUsingMergedIndex(
             {
                 for (size_t i = 0; i < readers.size(); ++i)
                 {
-                    granules[i] = readers[i]->read();
+                    readers[i]->read(granules[i]);
                     granules_filled = true;
                 }
             }

src/Storages/MergeTree/MergeTreeIndexFullText.h

@@ -25,8 +25,8 @@ struct MergeTreeIndexGranuleFullText final : public IMergeTreeIndexGranule
 
     bool empty() const override { return !has_elems; }
 
-    String index_name;
-    BloomFilterParameters params;
+    const String index_name;
+    const BloomFilterParameters params;
 
     std::vector<BloomFilter> bloom_filters;
     bool has_elems;

src/Storages/MergeTree/MergeTreeIndexGranuleBloomFilter.cpp

@@ -16,10 +16,11 @@ namespace ErrorCodes
 }
 
 MergeTreeIndexGranuleBloomFilter::MergeTreeIndexGranuleBloomFilter(size_t bits_per_row_, size_t hash_functions_, size_t index_columns_)
-    : bits_per_row(bits_per_row_), hash_functions(hash_functions_)
+    : bits_per_row(bits_per_row_), hash_functions(hash_functions_), bloom_filters(index_columns_)
 {
     total_rows = 0;
-    bloom_filters.resize(index_columns_);
+    for (size_t column = 0; column < index_columns_; ++column)
+        bloom_filters[column] = std::make_shared<BloomFilter>(bits_per_row, hash_functions, 0);
 }
 
 MergeTreeIndexGranuleBloomFilter::MergeTreeIndexGranuleBloomFilter(
@@ -55,8 +56,6 @@ bool MergeTreeIndexGranuleBloomFilter::empty() const
 
 void MergeTreeIndexGranuleBloomFilter::deserializeBinary(ReadBuffer & istr, MergeTreeIndexVersion version)
 {
-    if (!empty())
-        throw Exception(ErrorCodes::LOGICAL_ERROR, "Cannot read data to a non-empty bloom filter index.");
     if (version != 1)
         throw Exception(ErrorCodes::LOGICAL_ERROR, "Unknown index version {}.", version);
 
@@ -67,7 +66,7 @@ void MergeTreeIndexGranuleBloomFilter::deserializeBinary(ReadBuffer & istr, Merg
     size_t read_size = bytes_size;
     for (auto & filter : bloom_filters)
     {
-        filter = std::make_shared<BloomFilter>(bytes_size, hash_functions, 0);
+        filter->resize(bytes_size);
 #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
         read_size = filter->getFilter().size() * sizeof(BloomFilter::UnderType);
 #endif

src/Storages/MergeTree/MergeTreeIndexGranuleBloomFilter.h

@@ -22,9 +22,10 @@ class MergeTreeIndexGranuleBloomFilter final : public IMergeTreeIndexGranule
     const std::vector<BloomFilterPtr> & getFilters() const { return bloom_filters; }
 
 private:
+    const size_t bits_per_row;
+    const size_t hash_functions;
+
     size_t total_rows = 0;
-    size_t bits_per_row;
-    size_t hash_functions;
     std::vector<BloomFilterPtr> bloom_filters;
 
     void fillingBloomFilter(BloomFilterPtr & bf, const HashSet<UInt64> & hashes) const;

src/Storages/MergeTree/MergeTreeIndexInverted.cpp

@@ -73,11 +73,11 @@ void MergeTreeIndexGranuleInverted::deserializeBinary(ReadBuffer & istr, MergeTr
     {
         size_serialization->deserializeBinary(field_rows, istr, {});
         size_t filter_size = field_rows.get<size_t>();
+        gin_filter.getFilter().resize(filter_size);
 
         if (filter_size == 0)
             continue;
 
-        gin_filter.getFilter().assign(filter_size, {});
         istr.readStrict(reinterpret_cast<char *>(gin_filter.getFilter().data()), filter_size * sizeof(GinSegmentWithRowIdRangeVector::value_type));
     }
     has_elems = true;

src/Storages/MergeTree/MergeTreeIndexInverted.h

@@ -24,8 +24,8 @@ struct MergeTreeIndexGranuleInverted final : public IMergeTreeIndexGranule
 
     bool empty() const override { return !has_elems; }
 
-    String index_name;
-    GinFilterParameters params;
+    const String index_name;
+    const GinFilterParameters params;
     GinFilters gin_filters;
     bool has_elems;
 };

src/Storages/MergeTree/MergeTreeIndexMinMax.h

@@ -25,8 +25,9 @@ struct MergeTreeIndexGranuleMinMax final : public IMergeTreeIndexGranule
 
     bool empty() const override { return hyperrectangle.empty(); }
 
-    String index_name;
-    Block index_sample_block;
+    const String index_name;
+    const Block index_sample_block;
+
     std::vector<Range> hyperrectangle;
 };
 

src/Storages/MergeTree/MergeTreeIndexReader.cpp

@@ -69,11 +69,12 @@ void MergeTreeIndexReader::seek(size_t mark)
     stream->seekToMark(mark);
 }
 
-MergeTreeIndexGranulePtr MergeTreeIndexReader::read()
+void MergeTreeIndexReader::read(MergeTreeIndexGranulePtr & granule)
 {
-    auto granule = index->createIndexGranule();
+    if (granule == nullptr)
+        granule = index->createIndexGranule();
+
     granule->deserializeBinary(*stream->getDataBuffer(), version);
-    return granule;
 }
 
 }

src/Storages/MergeTree/MergeTreeIndexReader.h

@@ -23,7 +23,7 @@ class MergeTreeIndexReader
 
     void seek(size_t mark);
 
-    MergeTreeIndexGranulePtr read();
+    void read(MergeTreeIndexGranulePtr & granule);
 
 private:
     MergeTreeIndexPtr index;

src/Storages/MergeTree/MergeTreeIndexSet.h

@@ -35,9 +35,10 @@ struct MergeTreeIndexGranuleSet final : public IMergeTreeIndexGranule
 
     ~MergeTreeIndexGranuleSet() override = default;
 
-    String index_name;
-    size_t max_rows;
-    Block index_sample_block;
+    const String index_name;
+    const size_t max_rows;
+    const Block index_sample_block;
+
     Block block;
 };
 

tests/performance/min_max_index.xml

@@ -0,0 +1,11 @@
+<test>
+    <create_query>CREATE TABLE index_test (z UInt32, INDEX i_x (mortonDecode(2, z).1) TYPE minmax, INDEX i_y (mortonDecode(2, z).2) TYPE minmax) ENGINE = MergeTree ORDER BY z</create_query>
+
+    <fill_query>INSERT INTO index_test SELECT number FROM numbers(0x100000000) WHERE rand() % 3 = 1</fill_query>
+
+    <query><![CDATA[
+    SELECT count() FROM index_test WHERE mortonDecode(2, z).1 >= 20000 AND mortonDecode(2, z).1 <= 20100 AND mortonDecode(2, z).2 >= 10000 AND mortonDecode(2, z).2 <= 10100
+    ]]></query>
+
+    <drop_query>DROP TABLE IF EXISTS index_test</drop_query>
+</test>