// golangFillQueryArguments populates the placeholder map with
// types and values from an array of Go values.
// TODO: This does not support arguments of the SQL 'Date' type, as there is not
// an equivalent type in Go's standard library. It's not currently needed by any
// of our internal tables.
func golangFillQueryArguments(pinfo *parser.PlaceholderInfo, args []interface{}) {
pinfo.Clear()
for i, arg := range args {
k := fmt.Sprint(i + 1)
if arg == nil {
pinfo.SetValue(k, parser.DNull)
continue
}
// A type switch to handle a few explicit types with special semantics:
// - Datums are passed along as is.
// - Time datatypes get special representation in the database.
var d parser.Datum
switch t := arg.(type) {
case parser.Datum:
d = t
case time.Time:
d = parser.MakeDTimestamp(t, time.Microsecond)
case time.Duration:
d = &parser.DInterval{Duration: duration.Duration{Nanos: t.Nanoseconds()}}
case *inf.Dec:
dd := &parser.DDecimal{}
dd.Set(t)
d = dd
}
if d == nil {
// Handle all types which have an underlying type that can be stored in the
// database.
// Note: if this reflection becomes a performance concern in the future,
// commonly used types could be added explicitly into the type switch above
// for a performance gain.
val := reflect.ValueOf(arg)
switch val.Kind() {
case reflect.Bool:
d = parser.MakeDBool(parser.DBool(val.Bool()))
case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64:
d = parser.NewDInt(parser.DInt(val.Int()))
case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64:
d = parser.NewDInt(parser.DInt(val.Uint()))
case reflect.Float32, reflect.Float64:
d = parser.NewDFloat(parser.DFloat(val.Float()))
case reflect.String:
d = parser.NewDString(val.String())
case reflect.Slice:
// Handle byte slices.
if val.Type().Elem().Kind() == reflect.Uint8 {
d = parser.NewDBytes(parser.DBytes(val.Bytes()))
}
}
if d == nil {
panic(fmt.Sprintf("unexpected type %T", arg))
}
}
pinfo.SetValue(k, d)
}
}
// DecodeTableValue decodes a value encoded by EncodeTableValue.
func DecodeTableValue(a *DatumAlloc, valType parser.Type, b []byte) (parser.Datum, []byte, error) {
_, dataOffset, _, typ, err := encoding.DecodeValueTag(b)
if err != nil {
return nil, b, err
}
if typ == encoding.Null {
return parser.DNull, b[dataOffset:], nil
}
switch valType {
case parser.TypeBool:
var x bool
b, x, err = encoding.DecodeBoolValue(b)
// No need to chunk allocate DBool as MakeDBool returns either
// parser.DBoolTrue or parser.DBoolFalse.
return parser.MakeDBool(parser.DBool(x)), b, err
case parser.TypeInt:
var i int64
b, i, err = encoding.DecodeIntValue(b)
return a.NewDInt(parser.DInt(i)), b, err
case parser.TypeFloat:
var f float64
b, f, err = encoding.DecodeFloatValue(b)
return a.NewDFloat(parser.DFloat(f)), b, err
case parser.TypeDecimal:
var d *inf.Dec
b, d, err = encoding.DecodeDecimalValue(b)
dd := a.NewDDecimal(parser.DDecimal{})
dd.Set(d)
return dd, b, err
case parser.TypeString:
var data []byte
b, data, err = encoding.DecodeBytesValue(b)
return a.NewDString(parser.DString(data)), b, err
case parser.TypeBytes:
var data []byte
b, data, err = encoding.DecodeBytesValue(b)
return a.NewDBytes(parser.DBytes(data)), b, err
case parser.TypeDate:
var i int64
b, i, err = encoding.DecodeIntValue(b)
return a.NewDDate(parser.DDate(i)), b, err
case parser.TypeTimestamp:
var t time.Time
b, t, err = encoding.DecodeTimeValue(b)
return a.NewDTimestamp(parser.DTimestamp{Time: t}), b, err
case parser.TypeTimestampTZ:
var t time.Time
b, t, err = encoding.DecodeTimeValue(b)
return a.NewDTimestampTZ(parser.DTimestampTZ{Time: t}), b, err
case parser.TypeInterval:
var d duration.Duration
b, d, err = encoding.DecodeDurationValue(b)
return a.NewDInterval(parser.DInterval{Duration: d}), b, err
default:
return nil, nil, errors.Errorf("TODO(pmattis): decoded index value: %s", valType)
}
}
// ShowColumns of a table.
// Privileges: Any privilege on table.
// Notes: postgres does not have a SHOW COLUMNS statement.
// mysql only returns columns you have privileges on.
func (p *planner) ShowColumns(n *parser.ShowColumns) (planNode, error) {
tn, err := n.Table.NormalizeWithDatabaseName(p.session.Database)
if err != nil {
return nil, err
}
desc, err := p.mustGetTableDesc(tn)
if err != nil {
return nil, err
}
if err := p.anyPrivilege(desc); err != nil {
return nil, err
}
columns := ResultColumns{
{Name: "Field", Typ: parser.TypeString},
{Name: "Type", Typ: parser.TypeString},
{Name: "Null", Typ: parser.TypeBool},
{Name: "Default", Typ: parser.TypeString},
}
return &delayedNode{
p: p,
name: "SHOW COLUMNS FROM " + tn.String(),
columns: columns,
constructor: func(p *planner) (planNode, error) {
v := p.newContainerValuesNode(columns, 0)
for i, col := range desc.Columns {
defaultExpr := parser.DNull
if e := desc.Columns[i].DefaultExpr; e != nil {
defaultExpr = parser.NewDString(*e)
}
newRow := parser.DTuple{
parser.NewDString(desc.Columns[i].Name),
parser.NewDString(col.Type.SQLString()),
parser.MakeDBool(parser.DBool(desc.Columns[i].Nullable)),
defaultExpr,
}
if err := v.rows.AddRow(newRow); err != nil {
v.rows.Close()
return nil, err
}
}
return v, nil
},
}, nil
}
// MakePrimaryIndexKey creates a key prefix that corresponds to a table row
// (in the primary index); it is intended for tests.
//
// The value types must match the primary key columns (or a prefix of them);
// supported types are: - Datum
// - bool (converts to DBool)
// - int (converts to DInt)
// - string (converts to DString)
func MakePrimaryIndexKey(desc *TableDescriptor, vals ...interface{}) (roachpb.Key, error) {
index := &desc.PrimaryIndex
if len(vals) > len(index.ColumnIDs) {
return nil, errors.Errorf("got %d values, PK has %d columns", len(vals), len(index.ColumnIDs))
}
datums := make([]parser.Datum, len(vals))
for i, v := range vals {
switch v := v.(type) {
case bool:
datums[i] = parser.MakeDBool(parser.DBool(v))
case int:
datums[i] = parser.NewDInt(parser.DInt(v))
case string:
datums[i] = parser.NewDString(v)
case parser.Datum:
datums[i] = v
default:
return nil, errors.Errorf("unexpected value type %T", v)
}
// Check that the value type matches.
colID := index.ColumnIDs[i]
for _, c := range desc.Columns {
if c.ID == colID {
if t := DatumTypeToColumnKind(datums[i].ResolvedType()); t != c.Type.Kind {
return nil, errors.Errorf("column %d of type %s, got value of type %s", i, c.Type.Kind, t)
}
break
}
}
}
// Create the ColumnID to index in datums slice map needed by
// MakeIndexKeyPrefix.
colIDToRowIndex := make(map[ColumnID]int)
for i := range vals {
colIDToRowIndex[index.ColumnIDs[i]] = i
}
keyPrefix := MakeIndexKeyPrefix(desc, index.ID)
key, _, err := EncodeIndexKey(desc, index, colIDToRowIndex, datums, keyPrefix)
if err != nil {
return nil, err
}
return roachpb.Key(key), nil
}
// ShowIndex returns all the indexes for a table.
// Privileges: Any privilege on table.
// Notes: postgres does not have a SHOW INDEXES statement.
// mysql requires some privilege for any column.
func (p *planner) ShowIndex(n *parser.ShowIndex) (planNode, error) {
tn, err := n.Table.NormalizeWithDatabaseName(p.session.Database)
if err != nil {
return nil, err
}
desc, err := p.mustGetTableDesc(tn)
if err != nil {
return nil, err
}
if err := p.anyPrivilege(desc); err != nil {
return nil, err
}
columns := ResultColumns{
{Name: "Table", Typ: parser.TypeString},
{Name: "Name", Typ: parser.TypeString},
{Name: "Unique", Typ: parser.TypeBool},
{Name: "Seq", Typ: parser.TypeInt},
{Name: "Column", Typ: parser.TypeString},
{Name: "Direction", Typ: parser.TypeString},
{Name: "Storing", Typ: parser.TypeBool},
}
return &delayedNode{
p: p,
name: "SHOW INDEX FROM " + tn.String(),
columns: columns,
constructor: func(p *planner) (planNode, error) {
v := p.newContainerValuesNode(columns, 0)
appendRow := func(index sqlbase.IndexDescriptor, colName string, sequence int,
direction string, isStored bool) error {
newRow := parser.DTuple{
parser.NewDString(tn.Table()),
parser.NewDString(index.Name),
parser.MakeDBool(parser.DBool(index.Unique)),
parser.NewDInt(parser.DInt(sequence)),
parser.NewDString(colName),
parser.NewDString(direction),
parser.MakeDBool(parser.DBool(isStored)),
}
_, err := v.rows.AddRow(newRow)
return err
}
for _, index := range append([]sqlbase.IndexDescriptor{desc.PrimaryIndex}, desc.Indexes...) {
sequence := 1
for i, col := range index.ColumnNames {
if err := appendRow(index, col, sequence, index.ColumnDirections[i].String(), false); err != nil {
v.rows.Close()
return nil, err
}
sequence++
}
for _, col := range index.StoreColumnNames {
if err := appendRow(index, col, sequence, "N/A", true); err != nil {
v.rows.Close()
return nil, err
}
sequence++
}
}
return v, nil
},
}, nil
}
func dumpTable(w io.Writer, conn *sqlConn, origDBName, origTableName string) error {
const limit = 100
// Escape names since they can't be used in placeholders.
dbname := parser.Name(origDBName).String()
tablename := parser.Name(origTableName).String()
if err := conn.Exec(fmt.Sprintf("SET DATABASE = %s", dbname), nil); err != nil {
return err
}
// Fetch all table metadata in a transaction and its time to guarantee it
// doesn't change between the various SHOW statements.
if err := conn.Exec("BEGIN", nil); err != nil {
return err
}
vals, err := conn.QueryRow("SELECT cluster_logical_timestamp()", nil)
if err != nil {
return err
}
clusterTS := string(vals[0].([]byte))
// A previous version of the code did a SELECT on system.descriptor. This
// required the SELECT privilege to the descriptor table, which only root
// has. Allowing non-root to do this would let users see other users' table
// descriptors which is a problem in multi-tenancy.
// Fetch column types.
rows, err := conn.Query(fmt.Sprintf("SHOW COLUMNS FROM %s", tablename), nil)
if err != nil {
return err
}
vals = make([]driver.Value, 2)
coltypes := make(map[string]string)
for {
if err := rows.Next(vals); err == io.EOF {
break
} else if err != nil {
return err
}
nameI, typI := vals[0], vals[1]
name, ok := nameI.(string)
if !ok {
return fmt.Errorf("unexpected value: %T", nameI)
}
typ, ok := typI.(string)
if !ok {
return fmt.Errorf("unexpected value: %T", typI)
}
coltypes[name] = typ
}
if err := rows.Close(); err != nil {
return err
}
// index holds the names, in order, of the primary key columns.
var index []string
// Primary index is always the first index returned by SHOW INDEX.
rows, err = conn.Query(fmt.Sprintf("SHOW INDEX FROM %s", tablename), nil)
if err != nil {
return err
}
vals = make([]driver.Value, 5)
var primaryIndex string
// Find the primary index columns.
for {
if err := rows.Next(vals); err == io.EOF {
break
} else if err != nil {
return err
}
b, ok := vals[1].(string)
if !ok {
return fmt.Errorf("unexpected value: %T", vals[1])
}
if primaryIndex == "" {
primaryIndex = b
} else if primaryIndex != b {
break
}
b, ok = vals[4].(string)
if !ok {
return fmt.Errorf("unexpected value: %T", vals[4])
}
index = append(index, parser.Name(b).String())
}
if err := rows.Close(); err != nil {
return err
}
if len(index) == 0 {
return fmt.Errorf("no primary key index found")
}
indexes := strings.Join(index, ", ")
// Build the SELECT query.
var sbuf bytes.Buffer
fmt.Fprintf(&sbuf, "SELECT %s, * FROM %s@%s AS OF SYSTEM TIME %s", indexes, tablename, primaryIndex, clusterTS)
var wbuf bytes.Buffer
fmt.Fprintf(&wbuf, " WHERE ROW (%s) > ROW (", indexes)
for i := range index {
if i > 0 {
wbuf.WriteString(", ")
}
fmt.Fprintf(&wbuf, "$%d", i+1)
}
wbuf.WriteString(")")
// No WHERE clause first time, so add a place to inject it.
fmt.Fprintf(&sbuf, "%%s ORDER BY %s LIMIT %d", indexes, limit)
bs := sbuf.String()
vals, err = conn.QueryRow(fmt.Sprintf("SHOW CREATE TABLE %s", tablename), nil)
if err != nil {
return err
}
create := vals[1].(string)
if _, err := w.Write([]byte(create)); err != nil {
return err
}
if _, err := w.Write([]byte(";\n")); err != nil {
return err
}
if err := conn.Exec("COMMIT", nil); err != nil {
return err
}
// pk holds the last values of the fetched primary keys
var pk []driver.Value
q := fmt.Sprintf(bs, "")
for {
rows, err := conn.Query(q, pk)
if err != nil {
return err
}
cols := rows.Columns()
pkcols := cols[:len(index)]
cols = cols[len(index):]
inserts := make([][]string, 0, limit)
i := 0
for i < limit {
vals := make([]driver.Value, len(cols)+len(pkcols))
if err := rows.Next(vals); err == io.EOF {
break
} else if err != nil {
return err
}
if pk == nil {
q = fmt.Sprintf(bs, wbuf.String())
}
pk = vals[:len(index)]
vals = vals[len(index):]
ivals := make([]string, len(vals))
// Values need to be correctly encoded for INSERT statements in a text file.
for si, sv := range vals {
switch t := sv.(type) {
case nil:
ivals[si] = "NULL"
case bool:
ivals[si] = parser.MakeDBool(parser.DBool(t)).String()
case int64:
ivals[si] = parser.NewDInt(parser.DInt(t)).String()
case float64:
ivals[si] = parser.NewDFloat(parser.DFloat(t)).String()
case string:
ivals[si] = parser.NewDString(t).String()
case []byte:
switch ct := coltypes[cols[si]]; ct {
case "INTERVAL":
ivals[si] = fmt.Sprintf("'%s'", t)
case "BYTES":
ivals[si] = parser.NewDBytes(parser.DBytes(t)).String()
default:
// STRING and DECIMAL types can have optional length
// suffixes, so only examine the prefix of the type.
if strings.HasPrefix(coltypes[cols[si]], "STRING") {
ivals[si] = parser.NewDString(string(t)).String()
} else if strings.HasPrefix(coltypes[cols[si]], "DECIMAL") {
ivals[si] = string(t)
} else {
panic(errors.Errorf("unknown []byte type: %s, %v: %s", t, cols[si], coltypes[cols[si]]))
}
}
case time.Time:
var d parser.Datum
ct := coltypes[cols[si]]
switch ct {
case "DATE":
d = parser.NewDDateFromTime(t, time.UTC)
case "TIMESTAMP":
d = parser.MakeDTimestamp(t, time.Nanosecond)
case "TIMESTAMP WITH TIME ZONE":
d = parser.MakeDTimestampTZ(t, time.Nanosecond)
default:
panic(errors.Errorf("unknown timestamp type: %s, %v: %s", t, cols[si], coltypes[cols[si]]))
}
ivals[si] = fmt.Sprintf("'%s'", d)
default:
panic(errors.Errorf("unknown field type: %T (%s)", t, cols[si]))
}
}
inserts = append(inserts, ivals)
i++
}
for si, sv := range pk {
b, ok := sv.([]byte)
if ok && strings.HasPrefix(coltypes[pkcols[si]], "STRING") {
// Primary key strings need to be converted to a go string, but not SQL
// encoded since they aren't being written to a text file.
pk[si] = string(b)
}
}
if err := rows.Close(); err != nil {
return err
}
if i == 0 {
break
}
fmt.Fprintf(w, "\nINSERT INTO %s VALUES", tablename)
for idx, values := range inserts {
if idx > 0 {
fmt.Fprint(w, ",")
}
fmt.Fprint(w, "\n\t(")
for vi, v := range values {
if vi > 0 {
fmt.Fprint(w, ", ")
}
fmt.Fprint(w, v)
}
fmt.Fprint(w, ")")
}
fmt.Fprintln(w, ";")
if i < limit {
break
}
}
return nil
}
// DecodeTableKey decodes a table key/value.
func DecodeTableKey(
a *DatumAlloc, valType parser.Type, key []byte, dir encoding.Direction,
) (parser.Datum, []byte, error) {
if (dir != encoding.Ascending) && (dir != encoding.Descending) {
return nil, nil, errors.Errorf("invalid direction: %d", dir)
}
var isNull bool
if key, isNull = encoding.DecodeIfNull(key); isNull {
return parser.DNull, key, nil
}
var rkey []byte
var err error
switch valType {
case parser.TypeBool:
var i int64
if dir == encoding.Ascending {
rkey, i, err = encoding.DecodeVarintAscending(key)
} else {
rkey, i, err = encoding.DecodeVarintDescending(key)
}
// No need to chunk allocate DBool as MakeDBool returns either
// parser.DBoolTrue or parser.DBoolFalse.
return parser.MakeDBool(parser.DBool(i != 0)), rkey, err
case parser.TypeInt:
var i int64
if dir == encoding.Ascending {
rkey, i, err = encoding.DecodeVarintAscending(key)
} else {
rkey, i, err = encoding.DecodeVarintDescending(key)
}
return a.NewDInt(parser.DInt(i)), rkey, err
case parser.TypeFloat:
var f float64
if dir == encoding.Ascending {
rkey, f, err = encoding.DecodeFloatAscending(key)
} else {
rkey, f, err = encoding.DecodeFloatDescending(key)
}
return a.NewDFloat(parser.DFloat(f)), rkey, err
case parser.TypeDecimal:
var d *inf.Dec
if dir == encoding.Ascending {
rkey, d, err = encoding.DecodeDecimalAscending(key, nil)
} else {
rkey, d, err = encoding.DecodeDecimalDescending(key, nil)
}
dd := a.NewDDecimal(parser.DDecimal{})
dd.Set(d)
return dd, rkey, err
case parser.TypeString:
var r string
if dir == encoding.Ascending {
rkey, r, err = encoding.DecodeUnsafeStringAscending(key, nil)
} else {
rkey, r, err = encoding.DecodeUnsafeStringDescending(key, nil)
}
return a.NewDString(parser.DString(r)), rkey, err
case parser.TypeBytes:
var r []byte
if dir == encoding.Ascending {
rkey, r, err = encoding.DecodeBytesAscending(key, nil)
} else {
rkey, r, err = encoding.DecodeBytesDescending(key, nil)
}
return a.NewDBytes(parser.DBytes(r)), rkey, err
case parser.TypeDate:
var t int64
if dir == encoding.Ascending {
rkey, t, err = encoding.DecodeVarintAscending(key)
} else {
rkey, t, err = encoding.DecodeVarintDescending(key)
}
return a.NewDDate(parser.DDate(t)), rkey, err
case parser.TypeTimestamp:
var t time.Time
if dir == encoding.Ascending {
rkey, t, err = encoding.DecodeTimeAscending(key)
} else {
rkey, t, err = encoding.DecodeTimeDescending(key)
}
return a.NewDTimestamp(parser.DTimestamp{Time: t}), rkey, err
case parser.TypeTimestampTZ:
var t time.Time
if dir == encoding.Ascending {
rkey, t, err = encoding.DecodeTimeAscending(key)
} else {
rkey, t, err = encoding.DecodeTimeDescending(key)
}
return a.NewDTimestampTZ(parser.DTimestampTZ{Time: t}), rkey, err
case parser.TypeInterval:
var d duration.Duration
if dir == encoding.Ascending {
rkey, d, err = encoding.DecodeDurationAscending(key)
} else {
rkey, d, err = encoding.DecodeDurationDescending(key)
}
return a.NewDInterval(parser.DInterval{Duration: d}), rkey, err
default:
return nil, nil, errors.Errorf("TODO(pmattis): decoded index key: %s", valType)
}
}
// UnmarshalColumnValue decodes the value from a key-value pair using the type
// expected by the column. An error is returned if the value's type does not
// match the column's type.
func UnmarshalColumnValue(
a *DatumAlloc, kind ColumnType_Kind, value *roachpb.Value,
) (parser.Datum, error) {
if value == nil {
return parser.DNull, nil
}
switch kind {
case ColumnType_BOOL:
v, err := value.GetBool()
if err != nil {
return nil, err
}
return parser.MakeDBool(parser.DBool(v)), nil
case ColumnType_INT:
v, err := value.GetInt()
if err != nil {
return nil, err
}
return a.NewDInt(parser.DInt(v)), nil
case ColumnType_FLOAT:
v, err := value.GetFloat()
if err != nil {
return nil, err
}
return a.NewDFloat(parser.DFloat(v)), nil
case ColumnType_DECIMAL:
v, err := value.GetDecimal()
if err != nil {
return nil, err
}
dd := a.NewDDecimal(parser.DDecimal{})
dd.Set(v)
return dd, nil
case ColumnType_STRING:
v, err := value.GetBytes()
if err != nil {
return nil, err
}
return a.NewDString(parser.DString(v)), nil
case ColumnType_BYTES:
v, err := value.GetBytes()
if err != nil {
return nil, err
}
return a.NewDBytes(parser.DBytes(v)), nil
case ColumnType_DATE:
v, err := value.GetInt()
if err != nil {
return nil, err
}
return a.NewDDate(parser.DDate(v)), nil
case ColumnType_TIMESTAMP:
v, err := value.GetTime()
if err != nil {
return nil, err
}
return a.NewDTimestamp(parser.DTimestamp{Time: v}), nil
case ColumnType_TIMESTAMPTZ:
v, err := value.GetTime()
if err != nil {
return nil, err
}
return a.NewDTimestampTZ(parser.DTimestampTZ{Time: v}), nil
case ColumnType_INTERVAL:
d, err := value.GetDuration()
if err != nil {
return nil, err
}
return a.NewDInterval(parser.DInterval{Duration: d}), nil
default:
return nil, errors.Errorf("unsupported column type: %s", kind)
}
}
// decodeOidDatum decodes bytes with specified Oid and format code into
// a datum.
func decodeOidDatum(id oid.Oid, code formatCode, b []byte) (parser.Datum, error) {
var d parser.Datum
switch id {
case oid.T_bool:
switch code {
case formatText:
v, err := strconv.ParseBool(string(b))
if err != nil {
return d, err
}
d = parser.MakeDBool(parser.DBool(v))
case formatBinary:
switch b[0] {
case 0:
d = parser.MakeDBool(false)
case 1:
d = parser.MakeDBool(true)
default:
return d, errors.Errorf("unsupported binary bool: %q", b)
}
default:
return d, errors.Errorf("unsupported bool format code: %d", code)
}
case oid.T_int2:
switch code {
case formatText:
i, err := strconv.ParseInt(string(b), 10, 64)
if err != nil {
return d, err
}
d = parser.NewDInt(parser.DInt(i))
case formatBinary:
if len(b) < 2 {
return d, errors.Errorf("int2 requires 2 bytes for binary format")
}
i := int16(binary.BigEndian.Uint16(b))
d = parser.NewDInt(parser.DInt(i))
default:
return d, errors.Errorf("unsupported int2 format code: %d", code)
}
case oid.T_int4:
switch code {
case formatText:
i, err := strconv.ParseInt(string(b), 10, 64)
if err != nil {
return d, err
}
d = parser.NewDInt(parser.DInt(i))
case formatBinary:
if len(b) < 4 {
return d, errors.Errorf("int4 requires 4 bytes for binary format")
}
i := int32(binary.BigEndian.Uint32(b))
d = parser.NewDInt(parser.DInt(i))
default:
return d, errors.Errorf("unsupported int4 format code: %d", code)
}
case oid.T_int8:
switch code {
case formatText:
i, err := strconv.ParseInt(string(b), 10, 64)
if err != nil {
return d, err
}
d = parser.NewDInt(parser.DInt(i))
case formatBinary:
if len(b) < 8 {
return d, errors.Errorf("int8 requires 8 bytes for binary format")
}
i := int64(binary.BigEndian.Uint64(b))
d = parser.NewDInt(parser.DInt(i))
default:
return d, errors.Errorf("unsupported int8 format code: %d", code)
}
case oid.T_float4:
switch code {
case formatText:
f, err := strconv.ParseFloat(string(b), 64)
if err != nil {
return d, err
}
d = parser.NewDFloat(parser.DFloat(f))
case formatBinary:
if len(b) < 4 {
return d, errors.Errorf("float4 requires 4 bytes for binary format")
}
f := math.Float32frombits(binary.BigEndian.Uint32(b))
d = parser.NewDFloat(parser.DFloat(f))
default:
return d, errors.Errorf("unsupported float4 format code: %d", code)
}
case oid.T_float8:
switch code {
case formatText:
f, err := strconv.ParseFloat(string(b), 64)
if err != nil {
return d, err
}
d = parser.NewDFloat(parser.DFloat(f))
case formatBinary:
if len(b) < 8 {
return d, errors.Errorf("float8 requires 8 bytes for binary format")
}
f := math.Float64frombits(binary.BigEndian.Uint64(b))
d = parser.NewDFloat(parser.DFloat(f))
default:
return d, errors.Errorf("unsupported float8 format code: %d", code)
}
case oid.T_numeric:
switch code {
case formatText:
dd := &parser.DDecimal{}
if _, ok := dd.SetString(string(b)); !ok {
return nil, errors.Errorf("could not parse string %q as decimal", b)
}
d = dd
case formatBinary:
r := bytes.NewReader(b)
alloc := struct {
pgNum pgNumeric
i16 int16
dd parser.DDecimal
}{}
for _, ptr := range []interface{}{
&alloc.pgNum.ndigits,
&alloc.pgNum.weight,
&alloc.pgNum.sign,
&alloc.pgNum.dscale,
} {
if err := binary.Read(r, binary.BigEndian, ptr); err != nil {
return d, err
}
}
if alloc.pgNum.ndigits > 0 {
decDigits := make([]byte, 0, alloc.pgNum.ndigits*pgDecDigits)
nextDigit := func() error {
if err := binary.Read(r, binary.BigEndian, &alloc.i16); err != nil {
return err
}
numZeroes := pgDecDigits
for i16 := alloc.i16; i16 > 0; i16 /= 10 {
numZeroes--
}
for ; numZeroes > 0; numZeroes-- {
decDigits = append(decDigits, '0')
}
return nil
}
for i := int16(0); i < alloc.pgNum.ndigits-1; i++ {
if err := nextDigit(); err != nil {
return d, err
}
if alloc.i16 > 0 {
decDigits = strconv.AppendUint(decDigits, uint64(alloc.i16), 10)
}
}
// The last digit may contain padding, which we need to deal with.
if err := nextDigit(); err != nil {
return d, err
}
dscale := (alloc.pgNum.ndigits - (alloc.pgNum.weight + 1)) * pgDecDigits
if overScale := dscale - alloc.pgNum.dscale; overScale > 0 {
dscale -= overScale
for i := int16(0); i < overScale; i++ {
alloc.i16 /= 10
}
}
decDigits = strconv.AppendUint(decDigits, uint64(alloc.i16), 10)
decString := string(decDigits)
if _, ok := alloc.dd.UnscaledBig().SetString(decString, 10); !ok {
return nil, errors.Errorf("could not parse string %q as decimal", decString)
}
alloc.dd.SetScale(inf.Scale(dscale))
}
switch alloc.pgNum.sign {
case pgNumericPos:
case pgNumericNeg:
alloc.dd.Neg(&alloc.dd.Dec)
default:
return d, errors.Errorf("unsupported numeric sign: %d", alloc.pgNum.sign)
}
d = &alloc.dd
default:
return d, errors.Errorf("unsupported numeric format code: %d", code)
}
case oid.T_text, oid.T_varchar:
switch code {
case formatText, formatBinary:
d = parser.NewDString(string(b))
default:
return d, errors.Errorf("unsupported text format code: %d", code)
}
case oid.T_bytea:
switch code {
case formatText:
// http://www.postgresql.org/docs/current/static/datatype-binary.html#AEN5667
// Code cribbed from github.com/lib/pq.
// We only support hex encoding.
if len(b) >= 2 && bytes.Equal(b[:2], []byte("\\x")) {
b = b[2:] // trim off leading "\\x"
result := make([]byte, hex.DecodedLen(len(b)))
_, err := hex.Decode(result, b)
if err != nil {
return d, err
}
d = parser.NewDBytes(parser.DBytes(result))
} else {
return d, errors.Errorf("unsupported bytea encoding: %q", b)
}
case formatBinary:
d = parser.NewDBytes(parser.DBytes(b))
default:
return d, errors.Errorf("unsupported bytea format code: %d", code)
}
case oid.T_timestamp:
switch code {
case formatText:
ts, err := parseTs(string(b))
if err != nil {
return d, errors.Errorf("could not parse string %q as timestamp", b)
}
d = parser.MakeDTimestamp(ts, time.Microsecond)
case formatBinary:
if len(b) < 8 {
return d, errors.Errorf("timestamp requires 8 bytes for binary format")
}
i := int64(binary.BigEndian.Uint64(b))
d = parser.MakeDTimestamp(pgBinaryToTime(i), time.Microsecond)
default:
return d, errors.Errorf("unsupported timestamp format code: %d", code)
}
case oid.T_timestamptz:
switch code {
case formatText:
ts, err := parseTs(string(b))
if err != nil {
return d, errors.Errorf("could not parse string %q as timestamp", b)
}
d = parser.MakeDTimestampTZ(ts, time.Microsecond)
case formatBinary:
if len(b) < 8 {
return d, errors.Errorf("timestamptz requires 8 bytes for binary format")
}
i := int64(binary.BigEndian.Uint64(b))
d = parser.MakeDTimestampTZ(pgBinaryToTime(i), time.Microsecond)
default:
return d, errors.Errorf("unsupported timestamptz format code: %d", code)
}
case oid.T_date:
switch code {
case formatText:
ts, err := parseTs(string(b))
if err != nil {
res, err := parser.ParseDDate(string(b), time.UTC)
if err != nil {
return d, errors.Errorf("could not parse string %q as date", b)
}
d = res
} else {
daysSinceEpoch := ts.Unix() / secondsInDay
d = parser.NewDDate(parser.DDate(daysSinceEpoch))
}
case formatBinary:
if len(b) < 4 {
return d, errors.Errorf("date requires 4 bytes for binary format")
}
i := int32(binary.BigEndian.Uint32(b))
d = pgBinaryToDate(i)
default:
return d, errors.Errorf("unsupported date format code: %d", code)
}
case oid.T_interval:
switch code {
case formatText:
d, err := parser.ParseDInterval(string(b))
if err != nil {
return d, errors.Errorf("could not parse string %q as interval", b)
}
return d, nil
default:
return d, errors.Errorf("unsupported interval format code: %d", code)
}
default:
return d, errors.Errorf("unsupported OID: %v", id)
}
return d, nil
}
func typByVal(typ parser.Type) parser.Datum {
_, variable := typ.Size()
return parser.MakeDBool(parser.DBool(!variable))
}
addColumn := func(column *sqlbase.ColumnDescriptor, attRelID parser.Datum, colNum int) error {
colTyp := column.Type.ToDatumType()
return addRow(
attRelID, // attrelid
parser.NewDString(column.Name), // attname
typOid(colTyp), // atttypid
zeroVal, // attstattarget
typLen(colTyp), // attlen
parser.NewDInt(parser.DInt(colNum)), // attnum
zeroVal, // attndims
negOneVal, // attcacheoff
negOneVal, // atttypmod
parser.DNull, // attbyval (see pg_type.typbyval)
parser.DNull, // attstorage
parser.DNull, // attalign
parser.MakeDBool(parser.DBool(!column.Nullable)), // attnotnull
parser.MakeDBool(parser.DBool(column.DefaultExpr != nil)), // atthasdef
parser.MakeDBool(false), // attisdropped
parser.MakeDBool(true), // attislocal
zeroVal, // attinhcount
parser.DNull, // attacl
parser.DNull, // attoptions
parser.DNull, // attfdwoptions
)
}
// Columns for table.
colNum := 0
if err := forEachColumnInTable(table, func(column *sqlbase.ColumnDescriptor) error {
colNum++
tableID := h.TableOid(db, table)
COLUMN_NAME STRING NOT NULL DEFAULT '',
"COLLATION" STRING NOT NULL DEFAULT '',
CARDINALITY INT NOT NULL DEFAULT 0,
DIRECTION STRING NOT NULL DEFAULT '',
STORING BOOL NOT NULL DEFAULT FALSE
);`,
populate: func(p *planner, addRow func(...parser.Datum) error) error {
return forEachTableDesc(p,
func(db *sqlbase.DatabaseDescriptor, table *sqlbase.TableDescriptor) error {
appendRow := func(index *sqlbase.IndexDescriptor, colName string, sequence int,
direction string, isStored bool) error {
return addRow(
defString, // table_catalog
parser.NewDString(db.GetName()), // table_schema
parser.NewDString(table.GetName()), // table_name
parser.MakeDBool(parser.DBool(index.Unique)), // non_unique
parser.NewDString(db.GetName()), // index_schema
parser.NewDString(index.Name), // index_name
parser.NewDInt(parser.DInt(sequence)), // seq_in_index
parser.NewDString(colName), // column_name
parser.DNull, // collation
parser.DNull, // cardinality
parser.NewDString(direction), // direction
parser.MakeDBool(parser.DBool(isStored)), // storing
)
}
return forEachIndexInTable(table, func(index *sqlbase.IndexDescriptor) error {
sequence := 1
for i, col := range index.ColumnNames {
// We add a row for each column of index.
addColumn := func(column *sqlbase.ColumnDescriptor, attRelID parser.Datum, colNum int) error {
colTyp := column.Type.ToDatumType()
return addRow(
attRelID, // attrelid
parser.NewDString(column.Name), // attname
typOid(colTyp), // atttypid
zeroVal, // attstattarget
typLen(colTyp), // attlen
parser.NewDInt(parser.DInt(colNum)), // attnum
zeroVal, // attndims
negOneVal, // attcacheoff
negOneVal, // atttypmod
parser.DNull, // attbyval (see pg_type.typbyval)
parser.DNull, // attstorage
parser.DNull, // attalign
parser.MakeDBool(parser.DBool(!column.Nullable)), // attnotnull
parser.MakeDBool(parser.DBool(column.DefaultExpr != nil)), // atthasdef
parser.MakeDBool(false), // attisdropped
parser.MakeDBool(true), // attislocal
zeroVal, // attinhcount
parser.DNull, // attacl
parser.DNull, // attoptions
parser.DNull, // attfdwoptions
)
}
// Columns for table.
colNum := 0
if err := forEachColumnInTable(table, func(column *sqlbase.ColumnDescriptor) error {
colNum++
tableID := h.TableOid(db, table)
func TestValues(t *testing.T) {
defer leaktest.AfterTest(t)()
p := makeTestPlanner()
vInt := int64(5)
vNum := 3.14159
vStr := "two furs one cub"
vBool := true
unsupp := &parser.RangeCond{}
intVal := func(v int64) *parser.NumVal {
return &parser.NumVal{Value: constant.MakeInt64(v)}
}
floatVal := func(f float64) *parser.CastExpr {
return &parser.CastExpr{
Expr: &parser.NumVal{Value: constant.MakeFloat64(f)},
Type: &parser.FloatColType{},
}
}
asRow := func(datums ...parser.Datum) []parser.DTuple {
return []parser.DTuple{datums}
}
makeValues := func(tuples ...*parser.Tuple) *parser.ValuesClause {
return &parser.ValuesClause{Tuples: tuples}
}
makeTuple := func(exprs ...parser.Expr) *parser.Tuple {
return &parser.Tuple{Exprs: exprs}
}
testCases := []struct {
stmt *parser.ValuesClause
rows []parser.DTuple
ok bool
}{
{
makeValues(makeTuple(intVal(vInt))),
asRow(parser.NewDInt(parser.DInt(vInt))),
true,
},
{
makeValues(makeTuple(intVal(vInt), intVal(vInt))),
asRow(parser.NewDInt(parser.DInt(vInt)), parser.NewDInt(parser.DInt(vInt))),
true,
},
{
makeValues(makeTuple(floatVal(vNum))),
asRow(parser.NewDFloat(parser.DFloat(vNum))),
true,
},
{
makeValues(makeTuple(parser.NewDString(vStr))),
asRow(parser.NewDString(vStr)),
true,
},
{
makeValues(makeTuple(parser.NewDBytes(parser.DBytes(vStr)))),
asRow(parser.NewDBytes(parser.DBytes(vStr))),
true,
},
{
makeValues(makeTuple(parser.MakeDBool(parser.DBool(vBool)))),
asRow(parser.MakeDBool(parser.DBool(vBool))),
true,
},
{
makeValues(makeTuple(unsupp)),
nil,
false,
},
}
for i, tc := range testCases {
plan, err := func() (_ planNode, err error) {
defer func() {
if r := recover(); r != nil {
err = errors.Errorf("%v", r)
}
}()
return p.ValuesClause(tc.stmt, nil)
}()
if err == nil != tc.ok {
t.Errorf("%d: error_expected=%t, but got error %v", i, tc.ok, err)
}
if plan != nil {
defer plan.Close()
plan, err = p.optimizePlan(plan, allColumns(plan))
if err != nil {
t.Errorf("%d: unexpected error in optimizePlan: %v", i, err)
continue
}
if err := p.startPlan(plan); err != nil {
t.Errorf("%d: unexpected error in Start: %v", i, err)
continue
}
var rows []parser.DTuple
next, err := plan.Next()
for ; next; next, err = plan.Next() {
rows = append(rows, plan.Values())
}
if err != nil {
t.Error(err)
continue
}
if !reflect.DeepEqual(rows, tc.rows) {
t.Errorf("%d: expected rows:\n%+v\nactual rows:\n%+v", i, tc.rows, rows)
}
}
}
}
// splitBoolExpr splits a boolean expression E into two boolean expressions RES and REM such that:
//
// - RES only has variables known to the conversion function (it is "restricted" to a particular
// set of variables)
//
// - If weaker is true, for any setting of variables x:
// E(x) = (RES(x) AND REM(x))
// This implies RES(x) <= E(x), i.e. RES is "weaker"
//
// - If weaker is false:
// E(x) = (RES(x) OR REM(x))
// This implies RES(x) => E(x), i.e. RES is "stronger"
func splitBoolExpr(
expr parser.TypedExpr, conv varConvertFunc, weaker bool,
) (restricted, remainder parser.TypedExpr) {
// If the expression only contains "restricted" vars, the split is trivial.
if exprCheckVars(expr, conv) {
// An "empty" filter is always true in the weaker (normal) case (where the filter is
// equivalent to RES AND REM) and always false in the stronger (inverted) case (where the
// filter is equivalent to RES OR REM).
return exprConvertVars(expr, conv), parser.MakeDBool(parser.DBool(weaker))
}
switch t := expr.(type) {
case *parser.AndExpr:
if weaker {
// In the weaker (normal) case, we have
// E = (leftRes AND leftRem) AND (rightRes AND rightRem)
// We can just rearrange:
// E = (leftRes AND rightRes) AND (leftRem AND rightRem)
leftRes, leftRem := splitBoolExpr(t.TypedLeft(), conv, weaker)
rightRes, rightRem := splitBoolExpr(t.TypedRight(), conv, weaker)
return makeAnd(leftRes, rightRes), makeAnd(leftRem, rightRem)
}
// In the stronger (inverted) case, we have
// E = (leftRes OR leftRem) AND (rightRes OR rightRem)
// We can't do more than:
// E = (leftRes AND rightRes) OR E
leftRes, _ := splitBoolExpr(t.TypedLeft(), conv, weaker)
rightRes, _ := splitBoolExpr(t.TypedRight(), conv, weaker)
return makeAnd(leftRes, rightRes), expr
case *parser.OrExpr:
if !weaker {
// In the stronger (inverted) case, we have
// E = (leftRes OR leftRem) OR (rightRes AND rightRem)
// We can just rearrange:
// E = (leftRes OR rightRes) OR (leftRem AND rightRem)
leftRes, leftRem := splitBoolExpr(t.TypedLeft(), conv, weaker)
rightRes, rightRem := splitBoolExpr(t.TypedRight(), conv, weaker)
return makeOr(leftRes, rightRes), makeOr(leftRem, rightRem)
}
// In the weaker (normal) case, we have
// E = (leftRes AND leftRem) OR (rightRes AND rightRem)
// We can't do more than:
// E = (leftRes OR rightRes) OR E
leftRes, _ := splitBoolExpr(t.TypedLeft(), conv, weaker)
rightRes, _ := splitBoolExpr(t.TypedRight(), conv, weaker)
return makeOr(leftRes, rightRes), expr
case *parser.ParenExpr:
return splitBoolExpr(t.TypedInnerExpr(), conv, weaker)
case *parser.NotExpr:
exprRes, exprRem := splitBoolExpr(t.TypedInnerExpr(), conv, !weaker)
return makeNot(exprRes), makeNot(exprRem)
default:
// We can't split off anything (we already handled the case when expr contains only
// restricted vars above).
// For why we return DBool(weaker), see the comment above on "empty" filters.
return parser.MakeDBool(parser.DBool(weaker)), expr
}
}
// Test that distributing agg functions according to DistAggregationTable
// yields correct results. We're going to run each aggregation as either the
// two-stage process described by the DistAggregationTable or as a single global
// process, and verify that the results are the same.
func TestDistAggregationTable(t *testing.T) {
defer leaktest.AfterTest(t)()
const numRows = 100
tc := serverutils.StartTestCluster(t, 1, base.TestClusterArgs{})
defer tc.Stopper().Stop()
// Create a table with a few columns:
// - random integer values from 0 to numRows
// - random integer values (with some NULLs)
// - random bool value (mostly false)
// - random bool value (mostly true)
// - random decimals
// - random decimals (with some NULLs)
rng, _ := randutil.NewPseudoRand()
sqlutils.CreateTable(
t, tc.ServerConn(0), "t",
"k INT PRIMARY KEY, int1 INT, int2 INT, bool1 BOOL, bool2 BOOL, dec1 DECIMAL, dec2 DECIMAL",
numRows,
func(row int) []parser.Datum {
return []parser.Datum{
parser.NewDInt(parser.DInt(row)),
parser.NewDInt(parser.DInt(rng.Intn(numRows))),
sqlbase.RandDatum(rng, sqlbase.ColumnType{Kind: sqlbase.ColumnType_INT}, true),
parser.MakeDBool(parser.DBool(rng.Intn(10) == 0)),
parser.MakeDBool(parser.DBool(rng.Intn(10) != 0)),
sqlbase.RandDatum(rng, sqlbase.ColumnType{Kind: sqlbase.ColumnType_DECIMAL}, false),
sqlbase.RandDatum(rng, sqlbase.ColumnType{Kind: sqlbase.ColumnType_DECIMAL}, true),
}
},
)
kvDB := tc.Server(0).KVClient().(*client.DB)
desc := sqlbase.GetTableDescriptor(kvDB, "test", "t")
for fn, info := range DistAggregationTable {
if info.LocalStage == distsqlrun.AggregatorSpec_IDENT &&
info.FinalStage == distsqlrun.AggregatorSpec_IDENT {
// IDENT only works as expected if all rows have the same value on the
// relevant column; skip testing this trivial case.
continue
}
// We're going to test each aggregation function on every column that can be
// used as input for it.
foundCol := false
for colIdx := 1; colIdx < len(desc.Columns); colIdx++ {
// See if this column works with this function.
_, _, err := distsqlrun.GetAggregateInfo(fn, desc.Columns[colIdx].Type)
if err != nil {
continue
}
foundCol = true
for _, numRows := range []int{5, numRows / 10, numRows / 2, numRows} {
name := fmt.Sprintf("%s/%s/%d", fn, desc.Columns[colIdx].Name, numRows)
t.Run(name, func(t *testing.T) {
checkDistAggregationInfo(t, tc.Server(0), desc, colIdx, numRows, fn, info)
})
}
}
if !foundCol {
t.Errorf("aggregation function %s was not tested (no suitable column)", fn)
}
}
}