sql: initial support for virtual columns

This change adds very basic support for virtual columns to the
non-test descriptors and catalog. This is gated behind an experimental
session setting.

Release note: None

docs/generated/settings/settings.html

@@ -99,6 +99,6 @@
 <tr><td><code>trace.debug.enable</code></td><td>boolean</td><td><code>false</code></td><td>if set, traces for recent requests can be seen in the /debug page</td></tr>
 <tr><td><code>trace.lightstep.token</code></td><td>string</td><td><code></code></td><td>if set, traces go to Lightstep using this token</td></tr>
 <tr><td><code>trace.zipkin.collector</code></td><td>string</td><td><code></code></td><td>if set, traces go to the given Zipkin instance (example: '127.0.0.1:9411'); ignored if trace.lightstep.token is set</td></tr>
-<tr><td><code>version</code></td><td>version</td><td><code>20.2-6</code></td><td>set the active cluster version in the format '<major>.<minor>'</td></tr>
+<tr><td><code>version</code></td><td>version</td><td><code>20.2-8</code></td><td>set the active cluster version in the format '<major>.<minor>'</td></tr>
 </tbody>
 </table>

pkg/clusterversion/cockroach_versions.go

@@ -202,6 +202,8 @@ const (
 	// UniqueWithoutIndexConstraints is when adding UNIQUE WITHOUT INDEX
 	// constraints is supported.
 	UniqueWithoutIndexConstraints
+	// VirtualComputedColumns is when virtual computed columns are supported.
+	VirtualComputedColumns
 
 	// Step (1): Add new versions here.
 )
@@ -328,6 +330,10 @@ var versionsSingleton = keyedVersions([]keyedVersion{
 		Key:     UniqueWithoutIndexConstraints,
 		Version: roachpb.Version{Major: 20, Minor: 2, Internal: 6},
 	},
+	{
+		Key:     VirtualComputedColumns,
+		Version: roachpb.Version{Major: 20, Minor: 2, Internal: 8},
+	},
 
 	// Step (2): Add new versions here.
 })

pkg/sql/catalog/descpb/structured.proto

@@ -119,7 +119,8 @@ message ColumnDescriptor {
   option (gogoproto.equal) = true;
   optional string name = 1 [(gogoproto.nullable) = false];
   optional uint32 id = 2 [(gogoproto.nullable) = false,
-      (gogoproto.customname) = "ID", (gogoproto.casttype) = "ColumnID"];
+                          (gogoproto.customname) = "ID",
+                          (gogoproto.casttype) = "ColumnID"];
   optional sql.sem.types.T type = 3;
   optional bool nullable = 4 [(gogoproto.nullable) = false];
   reserved 8;
@@ -142,6 +143,11 @@ message ColumnDescriptor {
   // have been serialized in a internal format. Instead, use one of the
   // schemaexpr.FormatExpr* functions.
   optional string compute_expr = 11;
+
+  // A computed column can be stored or virtual.
+  // Virtual can only be true if there is a compute expression.
+  optional bool virtual = 16 [(gogoproto.nullable) = false];
+
   // PGAttributeNum must be accessed through the accessor, since it is set
   // lazily, it is incorrect to access it directly.
   // PGAttributeNum represents a column's number in catalog tables.

pkg/sql/catalog/schemaexpr/column.go

@@ -107,7 +107,11 @@ func FormatColumnForDisplay(
 			return "", err
 		}
 		f.WriteString(compExpr)
-		f.WriteString(") STORED")
+		if desc.Virtual {
+			f.WriteString(") VIRTUAL")
+		} else {
+			f.WriteString(") STORED")
+		}
 	}
 	return f.CloseAndGetString(), nil
 }

pkg/sql/catalog/tabledesc/structured.go

@@ -1316,6 +1316,10 @@ func (desc *Mutable) allocateColumnFamilyIDs(columnNames map[string]descpb.Colum
 	}
 
 	ensureColumnInFamily := func(col *descpb.ColumnDescriptor) {
+		if col.Virtual {
+			// Virtual columns don't need to be part of families.
+			return
+		}
 		if columnsInFamilies.Contains(col.ID) {
 			return
 		}
@@ -1944,7 +1948,7 @@ func (desc *Immutable) ValidateTable(ctx context.Context) error {
 	}
 
 	columnNames := make(map[string]descpb.ColumnID, len(desc.Columns))
-	columnIDs := make(map[descpb.ColumnID]string, len(desc.Columns))
+	columnIDs := make(map[descpb.ColumnID]*descpb.ColumnDescriptor, len(desc.Columns))
 	if err := desc.validateColumns(columnNames, columnIDs); err != nil {
 		return err
 	}
@@ -1959,7 +1963,7 @@ func (desc *Immutable) ValidateTable(ctx context.Context) error {
 					"mutation in state %s, direction %s, col %q, id %v",
 					errors.Safe(m.State), errors.Safe(m.Direction), col.Name, errors.Safe(col.ID))
 			}
-			columnIDs[col.ID] = col.Name
+			columnIDs[col.ID] = col
 		case *descpb.DescriptorMutation_Index:
 			if unSetEnums {
 				idx := desc.Index
@@ -2080,9 +2084,12 @@ func (desc *Immutable) ValidateTable(ctx context.Context) error {
 }
 
 func (desc *Immutable) validateColumns(
-	columnNames map[string]descpb.ColumnID, columnIDs map[descpb.ColumnID]string,
+	columnNames map[string]descpb.ColumnID, columnIDs map[descpb.ColumnID]*descpb.ColumnDescriptor,
 ) error {
-	for _, column := range desc.AllNonDropColumns() {
+	colDescs := desc.AllNonDropColumns()
+	for colIdx := range colDescs {
+		column := &colDescs[colIdx]
+
 		if err := catalog.ValidateName(column.Name, "column"); err != nil {
 			return err
 		}
@@ -2108,9 +2115,9 @@ func (desc *Immutable) validateColumns(
 
 		if other, ok := columnIDs[column.ID]; ok {
 			return fmt.Errorf("column %q duplicate ID of column %q: %d",
-				column.Name, other, column.ID)
+				column.Name, other.ID, column.ID)
 		}
-		columnIDs[column.ID] = column.Name
+		columnIDs[column.ID] = column
 
 		if column.ID >= desc.NextColumnID {
 			return errors.AssertionFailedf("column %q invalid ID (%d) >= next column ID (%d)",
@@ -2131,12 +2138,16 @@ func (desc *Immutable) validateColumns(
 				return fmt.Errorf("computed column %q refers to unknown columns in expression: %s",
 					column.Name, *column.ComputeExpr)
 			}
+		} else if column.Virtual {
+			return fmt.Errorf("virtual column %q is not computed", column.Name)
 		}
 	}
 	return nil
 }
 
-func (desc *Immutable) validateColumnFamilies(columnIDs map[descpb.ColumnID]string) error {
+func (desc *Immutable) validateColumnFamilies(
+	columnIDs map[descpb.ColumnID]*descpb.ColumnDescriptor,
+) error {
 	if len(desc.Families) < 1 {
 		return fmt.Errorf("at least 1 column family must be specified")
 	}
@@ -2184,13 +2195,16 @@ func (desc *Immutable) validateColumnFamilies(columnIDs map[descpb.ColumnID]stri
 		}
 
 		for i, colID := range family.ColumnIDs {
-			name, ok := columnIDs[colID]
+			col, ok := columnIDs[colID]
 			if !ok {
 				return fmt.Errorf("family %q contains unknown column \"%d\"", family.Name, colID)
 			}
-			if name != family.ColumnNames[i] {
+			if col.Name != family.ColumnNames[i] {
 				return fmt.Errorf("family %q column %d should have name %q, but found name %q",
-					family.Name, colID, name, family.ColumnNames[i])
+					family.Name, colID, col.Name, family.ColumnNames[i])
+			}
+			if col.Virtual {
+				return fmt.Errorf("virtual computed column %q cannot be part of a family", col.Name)
 			}
 		}
 
@@ -2201,9 +2215,11 @@ func (desc *Immutable) validateColumnFamilies(columnIDs map[descpb.ColumnID]stri
 			colIDToFamilyID[colID] = family.ID
 		}
 	}
-	for colID := range columnIDs {
-		if _, ok := colIDToFamilyID[colID]; !ok {
-			return fmt.Errorf("column %d is not in any column family", colID)
+	for colID, colDesc := range columnIDs {
+		if !colDesc.Virtual {
+			if _, ok := colIDToFamilyID[colID]; !ok {
+				return fmt.Errorf("column %q is not in any column family", colDesc.Name)
+			}
 		}
 	}
 	return nil
@@ -2212,7 +2228,9 @@ func (desc *Immutable) validateColumnFamilies(columnIDs map[descpb.ColumnID]stri
 // validateCheckConstraints validates that check constraints are well formed.
 // Checks include validating the column IDs and verifying that check expressions
 // do not reference non-existent columns.
-func (desc *Immutable) validateCheckConstraints(columnIDs map[descpb.ColumnID]string) error {
+func (desc *Immutable) validateCheckConstraints(
+	columnIDs map[descpb.ColumnID]*descpb.ColumnDescriptor,
+) error {
 	for _, chk := range desc.AllActiveAndInactiveChecks() {
 		// Verify that the check's column IDs are valid.
 		for _, colID := range chk.ColumnIDs {
@@ -2243,7 +2261,7 @@ func (desc *Immutable) validateCheckConstraints(columnIDs map[descpb.ColumnID]st
 // constraints are well formed. Checks include validating the column IDs and
 // column names.
 func (desc *Immutable) validateUniqueWithoutIndexConstraints(
-	columnIDs map[descpb.ColumnID]string,
+	columnIDs map[descpb.ColumnID]*descpb.ColumnDescriptor,
 ) error {
 	for _, c := range desc.AllActiveAndInactiveUniqueWithoutIndexConstraints() {
 		if err := catalog.ValidateName(c.Name, "unique without index constraint"); err != nil {

pkg/sql/catalog/tabledesc/table.go

@@ -68,6 +68,7 @@ func MakeColumnDefDescs(
 	col := &descpb.ColumnDescriptor{
 		Name:     string(d.Name),
 		Nullable: d.Nullable.Nullability != tree.NotNull && !d.PrimaryKey.IsPrimaryKey,
+		Virtual:  d.IsVirtual(),
 	}
 
 	// Validate and assign column type.

pkg/sql/catalog/tabledesc/validate_test.go

@@ -169,6 +169,7 @@ var validationMap = []struct {
 				status: todoIAmKnowinglyAddingTechDebt,
 				reason: "initial import: TODO(features): add validation"},
 			"ComputeExpr": {status: iSolemnlySwearThisFieldIsValidated},
+			"Virtual":     {status: iSolemnlySwearThisFieldIsValidated},
 			"PGAttributeNum": {
 				status: todoIAmKnowinglyAddingTechDebt,
 				reason: "initial import: TODO(features): add validation"},

pkg/sql/create_table.go

@@ -1405,7 +1405,17 @@ func NewTableDesc(
 				columnDefaultExprs = append(columnDefaultExprs, nil)
 			}
 			if d.IsVirtual() {
-				return nil, unimplemented.NewWithIssue(57608, "virtual computed columns")
+				if !sessionData.VirtualColumnsEnabled {
+					return nil, unimplemented.NewWithIssue(57608, "virtual computed columns")
+				}
+				if !evalCtx.Settings.Version.IsActive(ctx, clusterversion.VirtualComputedColumns) {
+					return nil, pgerror.Newf(pgcode.FeatureNotSupported,
+						"version %v must be finalized to use virtual columns",
+						clusterversion.VirtualComputedColumns)
+				}
+				if d.HasColumnFamily() {
+					return nil, pgerror.Newf(pgcode.Syntax, "virtual columns cannot have family specifications")
+				}
 			}
 
 			col, idx, expr, err := tabledesc.MakeColumnDefDescs(ctx, d, semaCtx, evalCtx)
@@ -2285,7 +2295,11 @@ func incTelemetryForNewColumn(def *tree.ColumnTableDef, desc *descpb.ColumnDescr
 		telemetry.Inc(sqltelemetry.SchemaNewTypeCounter(desc.Type.TelemetryName()))
 	}
 	if desc.IsComputed() {
-		telemetry.Inc(sqltelemetry.SchemaNewColumnTypeQualificationCounter("computed"))
+		if desc.Virtual {
+			telemetry.Inc(sqltelemetry.SchemaNewColumnTypeQualificationCounter("virtual"))
+		} else {
+			telemetry.Inc(sqltelemetry.SchemaNewColumnTypeQualificationCounter("computed"))
+		}
 	}
 	if desc.HasDefault() {
 		telemetry.Inc(sqltelemetry.SchemaNewColumnTypeQualificationCounter("default_expr"))

pkg/sql/exec_util.go

@@ -2188,6 +2188,11 @@ func (m *sessionDataMutator) SetMutliColumnInvertedIndexes(val bool) {
 	m.data.EnableMultiColumnInvertedIndexes = val
 }
 
+// TODO(radu): remove this once the feature is stable.
+func (m *sessionDataMutator) SetVirtualColumnsEnabled(val bool) {
+	m.data.VirtualColumnsEnabled = val
+}
+
 // TODO(rytaft): remove this once unique without index constraints are fully
 // supported.
 func (m *sessionDataMutator) SetUniqueWithoutIndexConstraints(val bool) {

pkg/sql/logictest/testdata/logic_test/pg_catalog

@@ -1850,6 +1850,7 @@ experimental_enable_hash_sharded_indexes              off                 NULL
 experimental_enable_multi_column_inverted_indexes     off                 NULL      NULL        NULL        string
 experimental_enable_temp_tables                       off                 NULL      NULL        NULL        string
 experimental_enable_unique_without_index_constraints  off                 NULL      NULL        NULL        string
+experimental_enable_virtual_columns                   off                 NULL      NULL        NULL        string
 extra_float_digits                                    0                   NULL      NULL        NULL        string
 force_savepoint_restart                               off                 NULL      NULL        NULL        string
 foreign_key_cascades_limit                            10000               NULL      NULL        NULL        string
@@ -1923,6 +1924,7 @@ experimental_enable_hash_sharded_indexes              off                 NULL
 experimental_enable_multi_column_inverted_indexes     off                 NULL  user     NULL      off                 off
 experimental_enable_temp_tables                       off                 NULL  user     NULL      off                 off
 experimental_enable_unique_without_index_constraints  off                 NULL  user     NULL      off                 off
+experimental_enable_virtual_columns                   off                 NULL  user     NULL      off                 off
 extra_float_digits                                    0                   NULL  user     NULL      0                   2
 force_savepoint_restart                               off                 NULL  user     NULL      off                 off
 foreign_key_cascades_limit                            10000               NULL  user     NULL      10000               10000
@@ -1992,6 +1994,7 @@ experimental_enable_hash_sharded_indexes              NULL    NULL     NULL
 experimental_enable_multi_column_inverted_indexes     NULL    NULL     NULL     NULL        NULL
 experimental_enable_temp_tables                       NULL    NULL     NULL     NULL        NULL
 experimental_enable_unique_without_index_constraints  NULL    NULL     NULL     NULL        NULL
+experimental_enable_virtual_columns                   NULL    NULL     NULL     NULL        NULL
 extra_float_digits                                    NULL    NULL     NULL     NULL        NULL
 force_savepoint_restart                               NULL    NULL     NULL     NULL        NULL
 foreign_key_cascades_limit                            NULL    NULL     NULL     NULL        NULL

pkg/sql/logictest/testdata/logic_test/show_source

@@ -50,6 +50,7 @@ experimental_enable_hash_sharded_indexes              off
 experimental_enable_multi_column_inverted_indexes     off
 experimental_enable_temp_tables                       off
 experimental_enable_unique_without_index_constraints  off
+experimental_enable_virtual_columns                   off
 extra_float_digits                                    0
 force_savepoint_restart                               off
 foreign_key_cascades_limit                            10000

pkg/sql/logictest/testdata/logic_test/virtual_columns

@@ -0,0 +1,45 @@
+statement ok
+SET experimental_enable_virtual_columns = true
+
+# Test that we don't allow FAMILY constraints with virtual columns.
+statement error virtual computed column "v" cannot be part of a family
+CREATE TABLE t (
+  a INT PRIMARY KEY,
+  b INT,
+  v INT AS (a+b) VIRTUAL,
+  FAMILY (a, b, v)
+)
+
+statement error virtual computed column "v" cannot be part of a family
+CREATE TABLE t (
+  a INT PRIMARY KEY,
+  b INT,
+  v INT AS (a+b) VIRTUAL,
+  FAMILY (a),
+  FAMILY (b),
+  FAMILY (v)
+)
+
+statement ok
+CREATE TABLE t (
+  a INT PRIMARY KEY,
+  b INT,
+  v INT AS (a+b) VIRTUAL,
+  FAMILY (a,b)
+)
+
+statement ok
+INSERT INTO t VALUES (1, 1)
+
+statement ok
+INSERT INTO t(a,b) VALUES (2, 2)
+
+statement error cannot write directly to computed column
+INSERT INTO t(a,b,v) VALUES (2, 2, 0)
+
+query III colnames,rowsort
+SELECT * FROM t
+----
+a  b  v
+1  1  2
+2  2  4

pkg/sql/opt/cat/column.go

@@ -49,11 +49,11 @@ func (c *Column) Ordinal() int {
 // dropped and then re-added with the same name; the new column will have a
 // different ID. See the comment for StableID for more detail.
 //
-// Virtual columns don't have stable IDs; for these columns ColID() must not be
-// called.
+// Virtual inverted columns don't have stable IDs; for these columns ColID()
+// must not be called.
 func (c *Column) ColID() StableID {
-	if c.kind.IsVirtual() {
-		panic(errors.AssertionFailedf("virtual columns have no StableID"))
+	if c.kind == VirtualInverted {
+		panic(errors.AssertionFailedf("virtual inverted columns have no StableID"))
 	}
 	return c.stableID
 }