Disallow distributing by numeric with negative scale

PG15 allows numeric scale to be negative or greater than precision. This
causes issues and we may end up routing queries to a wrong shard due to
differing hash results after rounding.

Formerly, when specifying NUMERIC(precision, scale), the scale had to be
in the range [0, precision], which was per SQL spec. PG15 extends the
range of allowed scales to [-1000, 1000].

A negative scale implies rounding before the decimal point. For
example, a column might be declared with a scale of -3 to round values
to the nearest thousand. Note that the display scale remains
non-negative, so in this case the display scale will be zero, and all
digits before the decimal point will be displayed.

Relevant PG commit: 085f931f52494e1f304e35571924efa6fcdc2b44

src/backend/distributed/commands/create_distributed_table.c

@@ -139,6 +139,14 @@ static Oid DropFKeysAndUndistributeTable(Oid relationId);
 static void DropFKeysRelationInvolvedWithTableType(Oid relationId, int tableTypeFlag);
 static void CopyLocalDataIntoShards(Oid relationId);
 static List * TupleDescColumnNameList(TupleDesc tupleDescriptor);
+
+#if (PG_VERSION_NUM >= PG_VERSION_15)
+static bool DistributionColumnUsesNumericColumnNegativeScale(TupleDesc relationDesc,
+															 Var *distributionColumn);
+static int numeric_typmod_scale(int32 typmod);
+static bool is_valid_numeric_typmod(int32 typmod);
+#endif
+
 static bool DistributionColumnUsesGeneratedStoredColumn(TupleDesc relationDesc,
 														Var *distributionColumn);
 static bool CanUseExclusiveConnections(Oid relationId, bool localTableEmpty);
@@ -1681,6 +1689,20 @@ EnsureRelationCanBeDistributed(Oid relationId, Var *distributionColumn,
 								  "AS (...) STORED.")));
 	}
 
+#if (PG_VERSION_NUM >= PG_VERSION_15)
+
+	/* verify target relation is not distributed by a column of type numeric with negative scale */
+	if (distributionMethod != DISTRIBUTE_BY_NONE &&
+		DistributionColumnUsesNumericColumnNegativeScale(relationDesc,
+														 distributionColumn))
+	{
+		ereport(ERROR, (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+						errmsg("cannot distribute relation: %s", relationName),
+						errdetail("Distribution column must not use numeric type "
+								  "with negative scale")));
+	}
+#endif
+
 	/* check for support function needed by specified partition method */
 	if (distributionMethod == DISTRIBUTE_BY_HASH)
 	{
@@ -2401,6 +2423,59 @@ RelationUsesIdentityColumns(TupleDesc relationDesc)
 }
 
 
+#if (PG_VERSION_NUM >= PG_VERSION_15)
+
+/*
+ * is_valid_numeric_typmod checks if the typmod value is valid
+ *
+ * Because of the offset, valid numeric typmods are at least VARHDRSZ
+ *
+ * Copied from PG. See numeric.c for understanding how this works.
+ */
+static bool
+is_valid_numeric_typmod(int32 typmod)
+{
+	return typmod >= (int32) VARHDRSZ;
+}
+
+
+/*
+ * numeric_typmod_scale extracts the scale from a numeric typmod.
+ *
+ * Copied from PG. See numeric.c for understanding how this works.
+ *
+ */
+static int
+numeric_typmod_scale(int32 typmod)
+{
+	return (((typmod - VARHDRSZ) & 0x7ff) ^ 1024) - 1024;
+}
+
+
+/*
+ * DistributionColumnUsesNumericColumnNegativeScale returns whether a given relation uses
+ * numeric data type with negative scale on distribution column
+ */
+static bool
+DistributionColumnUsesNumericColumnNegativeScale(TupleDesc relationDesc,
+												 Var *distributionColumn)
+{
+	Form_pg_attribute attributeForm = TupleDescAttr(relationDesc,
+													distributionColumn->varattno - 1);
+
+	if (attributeForm->atttypid == NUMERICOID &&
+		is_valid_numeric_typmod(attributeForm->atttypmod) &&
+		numeric_typmod_scale(attributeForm->atttypmod) < 0)
+	{
+		return true;
+	}
+
+	return false;
+}
+
+
+#endif
+
 /*
  * DistributionColumnUsesGeneratedStoredColumn returns whether a given relation uses
  * GENERATED ALWAYS AS (...) STORED on distribution column

src/test/regress/expected/pg15.out

@@ -392,9 +392,37 @@ ON (true)
 WHEN MATCHED THEN
     UPDATE SET x = (SELECT count(*) FROM tbl2);
 ERROR:  MERGE command is not supported on Citus tables yet
+-- test numeric types with negative scale
+CREATE TABLE numeric_negative_scale(numeric_column numeric(3,-1), orig_value int);
+INSERT into numeric_negative_scale SELECT x,x FROM generate_series(111, 115) x;
+-- verify that we can not distribute by a column that has numeric type with negative scale
+SELECT create_distributed_table('numeric_negative_scale','numeric_column');
+ERROR:  cannot distribute relation: numeric_negative_scale
+DETAIL:  Distribution column must not use numeric type with negative scale
+-- However, we can distribute by other columns
+SELECT create_distributed_table('numeric_negative_scale','orig_value');
+NOTICE:  Copying data from local table...
+NOTICE:  copying the data has completed
+DETAIL:  The local data in the table is no longer visible, but is still on disk.
+HINT:  To remove the local data, run: SELECT truncate_local_data_after_distributing_table($$pg15.numeric_negative_scale$$)
+ create_distributed_table
+---------------------------------------------------------------------
+
+(1 row)
+
+SELECT * FROM numeric_negative_scale ORDER BY 1,2;
+ numeric_column | orig_value
+---------------------------------------------------------------------
+            110 |        111
+            110 |        112
+            110 |        113
+            110 |        114
+            120 |        115
+(5 rows)
+
 -- Clean up
 DROP SCHEMA pg15 CASCADE;
-NOTICE:  drop cascades to 9 other objects
+NOTICE:  drop cascades to 10 other objects
 DETAIL:  drop cascades to collation german_phonebook_test
 drop cascades to collation default_provider
 drop cascades to table sale
@@ -404,3 +432,4 @@ drop cascades to view sale_triggers
 drop cascades to table generated_stored_ref
 drop cascades to table tbl1
 drop cascades to table tbl2
+drop cascades to table numeric_negative_scale

src/test/regress/sql/pg15.sql

@@ -245,5 +245,15 @@ ON (true)
 WHEN MATCHED THEN
     UPDATE SET x = (SELECT count(*) FROM tbl2);
 
+-- test numeric types with negative scale
+CREATE TABLE numeric_negative_scale(numeric_column numeric(3,-1), orig_value int);
+INSERT into numeric_negative_scale SELECT x,x FROM generate_series(111, 115) x;
+-- verify that we can not distribute by a column that has numeric type with negative scale
+SELECT create_distributed_table('numeric_negative_scale','numeric_column');
+-- However, we can distribute by other columns
+SELECT create_distributed_table('numeric_negative_scale','orig_value');
+
+SELECT * FROM numeric_negative_scale ORDER BY 1,2;
+
 -- Clean up
 DROP SCHEMA pg15 CASCADE;