Merge pull request ClickHouse#92739 from ClickHouse/backport/25.8/92726

Backport ClickHouse#92726 to 25.8: Fix incorrect granules pruning in`KeyCondition` for `match` function

Author: clickhouse-gh[bot]

src/Storages/MergeTree/KeyCondition.cpp

@@ -3300,6 +3300,16 @@ BoolMask KeyCondition::checkInHyperrectangle(
                 rpn_stack.back().can_be_true = mayExistOnBloomFilter(*element.bloom_filter_data, column_index_to_column_bf);
             }
 
+            /// If the condition is relaxed, the `can_be_false` branch is no longer reliable; it may have false negatives.
+            /// If `element.range` is relaxed (and thus wider) and contains `key_range`, then `can_be_false` becomes false.
+            /// However, in reality `can_be_false` may be true, because the actual range of element may be stricter than `element.range`.
+            /// For example, for `match(...)`, a false negative here (i.e. `can_be_false` is false) would make
+            /// `not match(...)` set `can_be_true = false`, causing us to skip the granule, which would be incorrect.
+            /// Therefore, we must set `can_be_false = true` to be safe.
+            /// Additionally, when `KeyCondition::isRelaxed()` is true, the caller should ignore `can_be_false` anyway.
+            if (element.relaxed)
+                rpn_stack.back().can_be_false = true;
+
             if (element.function == RPNElement::FUNCTION_NOT_IN_RANGE)
                 rpn_stack.back() = !rpn_stack.back();
         }

tests/queries/0_stateless/03772_key_condition_match_function_relaxed.reference

@@ -0,0 +1,86 @@
+-- { echo }
+SELECT count(*)
+FROM 03772_table_match
+WHERE NOT match(url, '^https?://clickhouse[.]com/');
+1
+EXPLAIN indexes = 1
+SELECT count(*)
+FROM 03772_table_match
+WHERE NOT match(url, '^https?://clickhouse[.]com/');
+Expression ((Project names + Projection))
+  Aggregating
+    Expression (Before GROUP BY)
+      Filter ((WHERE + Change column names to column identifiers))
+        ReadFromMergeTree (default.03772_table_match)
+        Indexes:
+          PrimaryKey
+            Keys:
+              url
+            Condition: not((url in [\'http\', \'httq\')))
+            Parts: 1/1
+            Granules: 1/1
+            Search Algorithm: generic exclusion search
+            Ranges: 1
+SELECT count(*)
+FROM 03772_table_match
+WHERE NOT match(url, '^abcd');
+1
+EXPLAIN indexes = 1
+SELECT count(*)
+FROM 03772_table_match
+WHERE NOT match(url, '^abcd');
+Expression ((Project names + Projection))
+  Aggregating
+    Expression (Before GROUP BY)
+      Filter ((WHERE + Change column names to column identifiers))
+        ReadFromMergeTree (default.03772_table_match)
+        Indexes:
+          PrimaryKey
+            Keys:
+              url
+            Condition: not((url in [\'abcd\', \'abce\')))
+            Parts: 1/1
+            Granules: 1/1
+            Search Algorithm: generic exclusion search
+            Ranges: 1
+SELECT count(*)
+FROM 03772_table_match
+WHERE match(url, '^abcd');
+0
+EXPLAIN indexes = 1
+SELECT count(*)
+FROM 03772_table_match
+WHERE match(url, '^abcd');
+Expression ((Project names + Projection))
+  Aggregating
+    Expression (Before GROUP BY)
+      Filter ((WHERE + Change column names to column identifiers))
+        ReadFromMergeTree (default.03772_table_match)
+        Indexes:
+          PrimaryKey
+            Keys:
+              url
+            Condition: (url in [\'abcd\', \'abce\'))
+            Parts: 0/1
+            Granules: 0/1
+            Search Algorithm: binary search
+            Ranges: 0
+SELECT count(*)
+FROM 03772_table_match
+WHERE match(url, '^https?://clickhouse[.]com/') = false;
+1
+EXPLAIN indexes = 1
+SELECT count(*)
+FROM 03772_table_match
+WHERE match(url, '^https?://clickhouse[.]com/') = false;
+Expression ((Project names + Projection))
+  Aggregating
+    Expression (Before GROUP BY)
+      Filter ((WHERE + Change column names to column identifiers))
+        ReadFromMergeTree (default.03772_table_match)
+        Indexes:
+          PrimaryKey
+            Condition: true
+            Parts: 1/1
+            Granules: 1/1
+            Ranges: 1

tests/queries/0_stateless/03772_key_condition_match_function_relaxed.sql

@@ -0,0 +1,48 @@
+-- Tags: no-replicated-database, no-parallel-replicas
+-- no-replicated-database: EXPLAIN output differs for replicated database.
+-- no-parallel-replicas: EXPLAIN output differs for parallel replicas.
+
+
+DROP TABLE IF EXISTS 03772_table_match;
+
+CREATE TABLE 03772_table_match
+ENGINE = MergeTree()
+ORDER BY url AS
+SELECT 'http://example1.com/' AS url;
+
+-- { echo }
+SELECT count(*)
+FROM 03772_table_match
+WHERE NOT match(url, '^https?://clickhouse[.]com/');
+
+EXPLAIN indexes = 1
+SELECT count(*)
+FROM 03772_table_match
+WHERE NOT match(url, '^https?://clickhouse[.]com/');
+
+SELECT count(*)
+FROM 03772_table_match
+WHERE NOT match(url, '^abcd');
+
+EXPLAIN indexes = 1
+SELECT count(*)
+FROM 03772_table_match
+WHERE NOT match(url, '^abcd');
+
+SELECT count(*)
+FROM 03772_table_match
+WHERE match(url, '^abcd');
+
+EXPLAIN indexes = 1
+SELECT count(*)
+FROM 03772_table_match
+WHERE match(url, '^abcd');
+
+SELECT count(*)
+FROM 03772_table_match
+WHERE match(url, '^https?://clickhouse[.]com/') = false;
+
+EXPLAIN indexes = 1
+SELECT count(*)
+FROM 03772_table_match
+WHERE match(url, '^https?://clickhouse[.]com/') = false;