Merge pull request #61257 from CurtizJ/fix-uniq-exact

Fix possible incorrect result of aggregate function `uniqExact`

src/AggregateFunctions/AggregateFunctionUniq.h

@@ -483,6 +483,7 @@ public:
     }
 
     bool isAbleToParallelizeMerge() const override { return is_able_to_parallelize_merge; }
+    bool canOptimizeEqualKeysRanges() const override { return !is_able_to_parallelize_merge; }
 
     void merge(AggregateDataPtr __restrict place, ConstAggregateDataPtr rhs, ThreadPool & thread_pool, Arena *) const override
     {
@@ -576,6 +577,7 @@ public:
     }
 
     bool isAbleToParallelizeMerge() const override { return is_able_to_parallelize_merge; }
+    bool canOptimizeEqualKeysRanges() const override { return !is_able_to_parallelize_merge; }
 
     void merge(AggregateDataPtr __restrict place, ConstAggregateDataPtr rhs, ThreadPool & thread_pool, Arena *) const override
     {

src/AggregateFunctions/Combinators/AggregateFunctionArray.h

@@ -142,6 +142,7 @@ public:
     }
 
     bool isAbleToParallelizeMerge() const override { return nested_func->isAbleToParallelizeMerge(); }
+    bool canOptimizeEqualKeysRanges() const override { return nested_func->canOptimizeEqualKeysRanges(); }
 
     void merge(AggregateDataPtr __restrict place, ConstAggregateDataPtr rhs, ThreadPool & thread_pool, Arena * arena) const override
     {

src/AggregateFunctions/Combinators/AggregateFunctionIf.h

@@ -165,6 +165,7 @@ public:
     }
 
     bool isAbleToParallelizeMerge() const override { return nested_func->isAbleToParallelizeMerge(); }
+    bool canOptimizeEqualKeysRanges() const override { return nested_func->canOptimizeEqualKeysRanges(); }
 
     void merge(AggregateDataPtr __restrict place, ConstAggregateDataPtr rhs, ThreadPool & thread_pool, Arena * arena) const override
     {

src/AggregateFunctions/Combinators/AggregateFunctionMerge.h

@@ -111,6 +111,7 @@ public:
     }
 
     bool isAbleToParallelizeMerge() const override { return nested_func->isAbleToParallelizeMerge(); }
+    bool canOptimizeEqualKeysRanges() const override { return nested_func->canOptimizeEqualKeysRanges(); }
 
     void merge(AggregateDataPtr __restrict place, ConstAggregateDataPtr rhs, ThreadPool & thread_pool, Arena * arena) const override
     {

src/AggregateFunctions/Combinators/AggregateFunctionNull.h

@@ -152,6 +152,7 @@ public:
     }
 
     bool isAbleToParallelizeMerge() const override { return nested_function->isAbleToParallelizeMerge(); }
+    bool canOptimizeEqualKeysRanges() const override { return nested_function->canOptimizeEqualKeysRanges(); }
 
     void merge(AggregateDataPtr __restrict place, ConstAggregateDataPtr rhs, ThreadPool & thread_pool, Arena * arena) const override
     {

src/AggregateFunctions/Combinators/AggregateFunctionState.h

@@ -92,6 +92,7 @@ public:
     }
 
     bool isAbleToParallelizeMerge() const override { return nested_func->isAbleToParallelizeMerge(); }
+    bool canOptimizeEqualKeysRanges() const override { return nested_func->canOptimizeEqualKeysRanges(); }
 
     void merge(AggregateDataPtr __restrict place, ConstAggregateDataPtr rhs, ThreadPool & thread_pool, Arena * arena) const override
     {

src/AggregateFunctions/IAggregateFunction.h

@@ -162,6 +162,10 @@ public:
     /// Tells if merge() with thread pool parameter could be used.
     virtual bool isAbleToParallelizeMerge() const { return false; }
 
+    /// Return true if it is allowed to replace call of `addBatch`
+    /// to `addBatchSinglePlace` for ranges of consecutive equal keys.
+    virtual bool canOptimizeEqualKeysRanges() const { return true; }
+
     /// Should be used only if isAbleToParallelizeMerge() returned true.
     virtual void
     merge(AggregateDataPtr __restrict /*place*/, ConstAggregateDataPtr /*rhs*/, ThreadPool & /*thread_pool*/, Arena * /*arena*/) const

src/Common/ColumnsHashingImpl.h

@@ -62,7 +62,6 @@ struct LastElementCache
     bool check(const Key & key) const { return value.first == key; }
 
     bool hasOnlyOneValue() const { return found && misses == 1; }
-    UInt64 getMisses() const { return misses; }
 };
 
 template <typename Data>
@@ -232,7 +231,7 @@ public:
     ALWAYS_INLINE UInt64 getCacheMissesSinceLastReset() const
     {
         if constexpr (consecutive_keys_optimization)
-            return cache.getMisses();
+            return cache.misses;
         return 0;
     }
 

src/Common/ProfileEvents.cpp

@@ -534,6 +534,7 @@ The server successfully detected this situation and will download merged part fr
     \
     M(AggregationPreallocatedElementsInHashTables, "How many elements were preallocated in hash tables for aggregation.") \
     M(AggregationHashTablesInitializedAsTwoLevel, "How many hash tables were inited as two-level for aggregation.") \
+    M(AggregationOptimizedEqualRangesOfKeys, "For how many blocks optimization of equal ranges of keys was applied") \
     \
     M(MetadataFromKeeperCacheHit, "Number of times an object storage metadata request was answered from cache without making request to Keeper") \
     M(MetadataFromKeeperCacheMiss, "Number of times an object storage metadata request had to be answered from Keeper") \

src/Interpreters/Aggregator.cpp

@@ -53,6 +53,7 @@ namespace ProfileEvents
     extern const Event OverflowThrow;
     extern const Event OverflowBreak;
     extern const Event OverflowAny;
+    extern const Event AggregationOptimizedEqualRangesOfKeys;
 }
 
 namespace CurrentMetrics
@@ -1344,6 +1345,7 @@ void NO_INLINE Aggregator::executeImplBatch(
     if (use_compiled_functions)
     {
         std::vector<ColumnData> columns_data;
+        bool can_optimize_equal_keys_ranges = true;
 
         for (size_t i = 0; i < aggregate_functions.size(); ++i)
         {
@@ -1352,13 +1354,15 @@ void NO_INLINE Aggregator::executeImplBatch(
 
             AggregateFunctionInstruction * inst = aggregate_instructions + i;
             size_t arguments_size = inst->that->getArgumentTypes().size(); // NOLINT
+            can_optimize_equal_keys_ranges &= inst->can_optimize_equal_keys_ranges;
 
             for (size_t argument_index = 0; argument_index < arguments_size; ++argument_index)
                 columns_data.emplace_back(getColumnData(inst->batch_arguments[argument_index]));
         }
 
-        if (all_keys_are_const || (!no_more_keys && state.hasOnlyOneValueSinceLastReset()))
+        if (all_keys_are_const || (can_optimize_equal_keys_ranges && state.hasOnlyOneValueSinceLastReset()))
         {
+            ProfileEvents::increment(ProfileEvents::AggregationOptimizedEqualRangesOfKeys);
             auto add_into_aggregate_states_function_single_place = compiled_aggregate_functions_holder->compiled_aggregate_functions.add_into_aggregate_states_function_single_place;
             add_into_aggregate_states_function_single_place(row_begin, row_end, columns_data.data(), places[key_start]);
         }
@@ -1380,10 +1384,15 @@ void NO_INLINE Aggregator::executeImplBatch(
 
         AggregateFunctionInstruction * inst = aggregate_instructions + i;
 
-        if (all_keys_are_const || (!no_more_keys && state.hasOnlyOneValueSinceLastReset()))
+        if (all_keys_are_const || (inst->can_optimize_equal_keys_ranges && state.hasOnlyOneValueSinceLastReset()))
+        {
+            ProfileEvents::increment(ProfileEvents::AggregationOptimizedEqualRangesOfKeys);
             addBatchSinglePlace(row_begin, row_end, inst, places[key_start] + inst->state_offset, aggregates_pool);
+        }
         else
+        {
             addBatch(row_begin, row_end, inst, places.get(), aggregates_pool);
+        }
     }
 }
 
@@ -1573,6 +1582,7 @@ void Aggregator::prepareAggregateInstructions(
         }
 
         aggregate_functions_instructions[i].has_sparse_arguments = has_sparse_arguments;
+        aggregate_functions_instructions[i].can_optimize_equal_keys_ranges = aggregate_functions[i]->canOptimizeEqualKeysRanges();
         aggregate_functions_instructions[i].arguments = aggregate_columns[i].data();
         aggregate_functions_instructions[i].state_offset = offsets_of_aggregate_states[i];
 

src/Interpreters/Aggregator.h

@@ -1221,6 +1221,7 @@ public:
         const IColumn ** batch_arguments{};
         const UInt64 * offsets{};
         bool has_sparse_arguments = false;
+        bool can_optimize_equal_keys_ranges = true;
     };
 
     /// Used for optimize_aggregation_in_order:

tests/queries/0_stateless/03008_optimize_equal_ranges.reference

@@ -0,0 +1,16 @@
+0	30000
+1	30000
+2	30000
+0	30000
+1	30000
+2	30000
+0	449985000
+1	449985000
+2	449985000
+0	449985000
+1	449985000
+2	449985000
+sum	1	1
+sum	16	1
+uniqExact	1	1
+uniqExact	16	0

tests/queries/0_stateless/03008_optimize_equal_ranges.sql

@@ -0,0 +1,29 @@
+DROP TABLE IF EXISTS t_optimize_equal_ranges;
+
+CREATE TABLE t_optimize_equal_ranges (a UInt64, b String, c UInt64) ENGINE = MergeTree ORDER BY a;
+
+SET max_block_size = 1024;
+SET max_bytes_before_external_group_by = 0;
+SET optimize_aggregation_in_order = 0;
+SET optimize_use_projections = 0;
+
+INSERT INTO t_optimize_equal_ranges SELECT 0, toString(number), number FROM numbers(30000);
+INSERT INTO t_optimize_equal_ranges SELECT 1, toString(number), number FROM numbers(30000);
+INSERT INTO t_optimize_equal_ranges SELECT 2, toString(number), number FROM numbers(30000);
+
+SELECT a, uniqExact(b) FROM t_optimize_equal_ranges GROUP BY a ORDER BY a SETTINGS max_threads = 16;
+SELECT a, uniqExact(b) FROM t_optimize_equal_ranges GROUP BY a ORDER BY a SETTINGS max_threads = 1;
+SELECT a, sum(c) FROM t_optimize_equal_ranges GROUP BY a ORDER BY a SETTINGS max_threads = 16;
+SELECT a, sum(c) FROM t_optimize_equal_ranges GROUP BY a ORDER BY a SETTINGS max_threads = 1;
+
+SYSTEM FLUSH LOGS;
+
+SELECT
+    used_aggregate_functions[1] AS func,
+    Settings['max_threads'] AS threads,
+    ProfileEvents['AggregationOptimizedEqualRangesOfKeys'] > 0
+FROM system.query_log
+WHERE type = 'QueryFinish' AND current_database = currentDatabase() AND query LIKE '%SELECT%FROM%t_optimize_equal_ranges%'
+ORDER BY func, threads;
+
+DROP TABLE t_optimize_equal_ranges;

tests/queries/0_stateless/03008_uniq_exact_equal_ranges.sql

@@ -0,0 +1,36 @@
+DROP TABLE IF EXISTS t_uniq_exact;
+
+CREATE TABLE t_uniq_exact (a UInt64, b String, c UInt64) ENGINE = MergeTree ORDER BY a;
+
+SET group_by_two_level_threshold_bytes = 1;
+SET group_by_two_level_threshold = 1;
+SET max_threads = 4;
+SET max_bytes_before_external_group_by = 0;
+SET optimize_aggregation_in_order = 0;
+
+INSERT INTO t_uniq_exact SELECT 0, randomPrintableASCII(5), rand() FROM numbers(300000);
+INSERT INTO t_uniq_exact SELECT 1, randomPrintableASCII(5), rand() FROM numbers(300000);
+INSERT INTO t_uniq_exact SELECT 2, randomPrintableASCII(5), rand() FROM numbers(300000);
+INSERT INTO t_uniq_exact SELECT 3, randomPrintableASCII(5), rand() FROM numbers(300000);
+INSERT INTO t_uniq_exact SELECT 4, randomPrintableASCII(5), rand() FROM numbers(300000);
+INSERT INTO t_uniq_exact SELECT 5, randomPrintableASCII(5), rand() FROM numbers(300000);
+INSERT INTO t_uniq_exact SELECT 6, randomPrintableASCII(5), rand() FROM numbers(300000);
+INSERT INTO t_uniq_exact SELECT 7, randomPrintableASCII(5), rand() FROM numbers(300000);
+INSERT INTO t_uniq_exact SELECT 8, randomPrintableASCII(5), rand() FROM numbers(300000);
+INSERT INTO t_uniq_exact SELECT 9, randomPrintableASCII(5), rand() FROM numbers(300000);
+
+OPTIMIZE TABLE t_uniq_exact FINAL;
+
+SELECT a, uniqExact(b) FROM t_uniq_exact GROUP BY a ORDER BY a
+SETTINGS min_hit_rate_to_use_consecutive_keys_optimization = 1.0
+EXCEPT
+SELECT a, uniqExact(b) FROM t_uniq_exact GROUP BY a ORDER BY a
+SETTINGS min_hit_rate_to_use_consecutive_keys_optimization = 0.5;
+
+SELECT a, sum(c) FROM t_uniq_exact GROUP BY a ORDER BY a
+SETTINGS min_hit_rate_to_use_consecutive_keys_optimization = 1.0
+EXCEPT
+SELECT a, sum(c) FROM t_uniq_exact GROUP BY a ORDER BY a
+SETTINGS min_hit_rate_to_use_consecutive_keys_optimization = 0.5;
+
+DROP TABLE t_uniq_exact;