pgx/pgtype/hstore.go
Evan Jones d48d36dc02 pgtype/hstore: Make text parsing about 6X faster
I am working on an application that uses hstore types, and we found
that returning the values is slow, particularly when using the text
protocol, such as when using database/sql. This improves parsing to
be about 6X faster (currently faster than binary). The changes are:

* referencing the original string instead of copying into new strings
  (very large win)
* using string.IndexByte to scan double quoted strings: it has
  architecture-specific assembly implementations, and most of the
  time is spent in key/value strings.
* estimating the number of key/value pairs to allocate the correct
  size of the slice and map up front. This reduces the number of
  allocations and bytes allocated by a factor of 2, and was a small
  CPU win.
* parsing directly into the Hstore, rather than copying into it.

This parser is stricter than the old one. It only accepts hstore
strings serialized by Postgres. The old one was already stricter
than Postgres's own parser, but previously accepted any whitespace
character after a comma. This one only accepts space. Example:

  "k1"=>"v1",\t"k2"=>"v2"

Postgres only ever uses ", " as the separator. See hstore_out:
https://github.com/postgres/postgres/blob/master/contrib/hstore/hstore_io.c

The result of using benchstat to compare the benchmark on my M1 Pro
with the following command line in below. The new text parser is now
faster than the binary parser. I will improve the binary parser in a
separate change.

for i in $(seq 10); do go test ./pgtype -run=none -bench=BenchmarkHstoreScan -benchtime=1s >> new.txt; done

goos: darwin
goarch: arm64
pkg: github.com/jackc/pgx/v5/pgtype
                               │  orig.txt   │               new.txt               │
                               │   sec/op    │   sec/op     vs base                │
HstoreScan/databasesql.Scan-10   82.11µ ± 1%   10.51µ ± 0%  -87.20% (p=0.000 n=10)
HstoreScan/text-10               83.30µ ± 1%   11.49µ ± 1%  -86.20% (p=0.000 n=10)
HstoreScan/binary-10             15.99µ ± 2%   15.77µ ± 1%   -1.35% (p=0.007 n=10)
geomean                          47.82µ        12.40µ       -74.08%

                               │   orig.txt   │               new.txt                │
                               │     B/op     │     B/op      vs base                │
HstoreScan/databasesql.Scan-10   56.23Ki ± 0%   11.68Ki ± 0%  -79.23% (p=0.000 n=10)
HstoreScan/text-10               65.12Ki ± 0%   20.58Ki ± 0%  -68.40% (p=0.000 n=10)
HstoreScan/binary-10             21.09Ki ± 0%   21.09Ki ± 0%        ~ (p=0.378 n=10)
geomean                          42.58Ki        17.18Ki       -59.66%

                               │  orig.txt   │               new.txt                │
                               │  allocs/op  │ allocs/op   vs base                  │
HstoreScan/databasesql.Scan-10   744.00 ± 0%   44.00 ± 0%  -94.09% (p=0.000 n=10)
HstoreScan/text-10               743.00 ± 0%   44.00 ± 0%  -94.08% (p=0.000 n=10)
HstoreScan/binary-10              464.0 ± 0%   464.0 ± 0%        ~ (p=1.000 n=10) ¹
geomean                           635.4        96.49       -84.81%
¹ all samples are equal
2023-06-16 15:30:54 -05:00

484 lines
11 KiB
Go

package pgtype
import (
"database/sql/driver"
"encoding/binary"
"errors"
"fmt"
"strings"
"github.com/jackc/pgx/v5/internal/pgio"
)
type HstoreScanner interface {
ScanHstore(v Hstore) error
}
type HstoreValuer interface {
HstoreValue() (Hstore, error)
}
// Hstore represents an hstore column that can be null or have null values
// associated with its keys.
type Hstore map[string]*string
func (h *Hstore) ScanHstore(v Hstore) error {
*h = v
return nil
}
func (h Hstore) HstoreValue() (Hstore, error) {
return h, nil
}
// Scan implements the database/sql Scanner interface.
func (h *Hstore) Scan(src any) error {
if src == nil {
*h = nil
return nil
}
switch src := src.(type) {
case string:
return scanPlanTextAnyToHstoreScanner{}.scanString(src, h)
}
return fmt.Errorf("cannot scan %T", src)
}
// Value implements the database/sql/driver Valuer interface.
func (h Hstore) Value() (driver.Value, error) {
if h == nil {
return nil, nil
}
buf, err := HstoreCodec{}.PlanEncode(nil, 0, TextFormatCode, h).Encode(h, nil)
if err != nil {
return nil, err
}
return string(buf), err
}
type HstoreCodec struct{}
func (HstoreCodec) FormatSupported(format int16) bool {
return format == TextFormatCode || format == BinaryFormatCode
}
func (HstoreCodec) PreferredFormat() int16 {
return BinaryFormatCode
}
func (HstoreCodec) PlanEncode(m *Map, oid uint32, format int16, value any) EncodePlan {
if _, ok := value.(HstoreValuer); !ok {
return nil
}
switch format {
case BinaryFormatCode:
return encodePlanHstoreCodecBinary{}
case TextFormatCode:
return encodePlanHstoreCodecText{}
}
return nil
}
type encodePlanHstoreCodecBinary struct{}
func (encodePlanHstoreCodecBinary) Encode(value any, buf []byte) (newBuf []byte, err error) {
hstore, err := value.(HstoreValuer).HstoreValue()
if err != nil {
return nil, err
}
if hstore == nil {
return nil, nil
}
buf = pgio.AppendInt32(buf, int32(len(hstore)))
for k, v := range hstore {
buf = pgio.AppendInt32(buf, int32(len(k)))
buf = append(buf, k...)
if v == nil {
buf = pgio.AppendInt32(buf, -1)
} else {
buf = pgio.AppendInt32(buf, int32(len(*v)))
buf = append(buf, (*v)...)
}
}
return buf, nil
}
type encodePlanHstoreCodecText struct{}
func (encodePlanHstoreCodecText) Encode(value any, buf []byte) (newBuf []byte, err error) {
hstore, err := value.(HstoreValuer).HstoreValue()
if err != nil {
return nil, err
}
if hstore == nil {
return nil, nil
}
firstPair := true
for k, v := range hstore {
if firstPair {
firstPair = false
} else {
buf = append(buf, ',')
}
// unconditionally quote hstore keys/values like Postgres does
// this avoids a Mac OS X Postgres hstore parsing bug:
// https://www.postgresql.org/message-id/CA%2BHWA9awUW0%2BRV_gO9r1ABZwGoZxPztcJxPy8vMFSTbTfi4jig%40mail.gmail.com
buf = append(buf, '"')
buf = append(buf, quoteArrayReplacer.Replace(k)...)
buf = append(buf, '"')
buf = append(buf, "=>"...)
if v == nil {
buf = append(buf, "NULL"...)
} else {
buf = append(buf, '"')
buf = append(buf, quoteArrayReplacer.Replace(*v)...)
buf = append(buf, '"')
}
}
return buf, nil
}
func (HstoreCodec) PlanScan(m *Map, oid uint32, format int16, target any) ScanPlan {
switch format {
case BinaryFormatCode:
switch target.(type) {
case HstoreScanner:
return scanPlanBinaryHstoreToHstoreScanner{}
}
case TextFormatCode:
switch target.(type) {
case HstoreScanner:
return scanPlanTextAnyToHstoreScanner{}
}
}
return nil
}
type scanPlanBinaryHstoreToHstoreScanner struct{}
func (scanPlanBinaryHstoreToHstoreScanner) Scan(src []byte, dst any) error {
scanner := (dst).(HstoreScanner)
if src == nil {
return scanner.ScanHstore(Hstore(nil))
}
rp := 0
if len(src[rp:]) < 4 {
return fmt.Errorf("hstore incomplete %v", src)
}
pairCount := int(int32(binary.BigEndian.Uint32(src[rp:])))
rp += 4
hstore := make(Hstore, pairCount)
for i := 0; i < pairCount; i++ {
if len(src[rp:]) < 4 {
return fmt.Errorf("hstore incomplete %v", src)
}
keyLen := int(int32(binary.BigEndian.Uint32(src[rp:])))
rp += 4
if len(src[rp:]) < keyLen {
return fmt.Errorf("hstore incomplete %v", src)
}
key := string(src[rp : rp+keyLen])
rp += keyLen
if len(src[rp:]) < 4 {
return fmt.Errorf("hstore incomplete %v", src)
}
valueLen := int(int32(binary.BigEndian.Uint32(src[rp:])))
rp += 4
var valueBuf []byte
if valueLen >= 0 {
valueBuf = src[rp : rp+valueLen]
rp += valueLen
}
var value Text
err := scanPlanTextAnyToTextScanner{}.Scan(valueBuf, &value)
if err != nil {
return err
}
if value.Valid {
hstore[key] = &value.String
} else {
hstore[key] = nil
}
}
return scanner.ScanHstore(hstore)
}
type scanPlanTextAnyToHstoreScanner struct{}
func (s scanPlanTextAnyToHstoreScanner) Scan(src []byte, dst any) error {
scanner := (dst).(HstoreScanner)
if src == nil {
return scanner.ScanHstore(Hstore(nil))
}
return s.scanString(string(src), scanner)
}
// scanString does not return nil hstore values because string cannot be nil.
func (scanPlanTextAnyToHstoreScanner) scanString(src string, scanner HstoreScanner) error {
hstore, err := parseHstore(src)
if err != nil {
return err
}
return scanner.ScanHstore(hstore)
}
func (c HstoreCodec) DecodeDatabaseSQLValue(m *Map, oid uint32, format int16, src []byte) (driver.Value, error) {
return codecDecodeToTextFormat(c, m, oid, format, src)
}
func (c HstoreCodec) DecodeValue(m *Map, oid uint32, format int16, src []byte) (any, error) {
if src == nil {
return nil, nil
}
var hstore Hstore
err := codecScan(c, m, oid, format, src, &hstore)
if err != nil {
return nil, err
}
return hstore, nil
}
type hstoreParser struct {
str string
pos int
nextBackslash int
}
func newHSP(in string) *hstoreParser {
return &hstoreParser{
pos: 0,
str: in,
nextBackslash: strings.IndexByte(in, '\\'),
}
}
func (p *hstoreParser) atEnd() bool {
return p.pos >= len(p.str)
}
// consume returns the next byte of the string, or end if the string is done.
func (p *hstoreParser) consume() (b byte, end bool) {
if p.pos >= len(p.str) {
return 0, true
}
b = p.str[p.pos]
p.pos++
return b, false
}
func unexpectedByteErr(actualB byte, expectedB byte) error {
return fmt.Errorf("expected '%c' ('%#v'); found '%c' ('%#v')", expectedB, expectedB, actualB, actualB)
}
// consumeExpectedByte consumes expectedB from the string, or returns an error.
func (p *hstoreParser) consumeExpectedByte(expectedB byte) error {
nextB, end := p.consume()
if end {
return fmt.Errorf("expected '%c' ('%#v'); found end", expectedB, expectedB)
}
if nextB != expectedB {
return unexpectedByteErr(nextB, expectedB)
}
return nil
}
// consumeExpected2 consumes two expected bytes or returns an error.
// This was a bit faster than using a string argument (better inlining? Not sure).
func (p *hstoreParser) consumeExpected2(one byte, two byte) error {
if p.pos+2 > len(p.str) {
return errors.New("unexpected end of string")
}
if p.str[p.pos] != one {
return unexpectedByteErr(p.str[p.pos], one)
}
if p.str[p.pos+1] != two {
return unexpectedByteErr(p.str[p.pos+1], two)
}
p.pos += 2
return nil
}
var errEOSInQuoted = errors.New(`found end before closing double-quote ('"')`)
// consumeDoubleQuoted consumes a double-quoted string from p. The double quote must have been
// parsed already.
func (p *hstoreParser) consumeDoubleQuoted() (string, error) {
// fast path: assume most keys/values do not contain escapes
nextDoubleQuote := strings.IndexByte(p.str[p.pos:], '"')
if nextDoubleQuote == -1 {
return "", errEOSInQuoted
}
nextDoubleQuote += p.pos
if p.nextBackslash == -1 || p.nextBackslash > nextDoubleQuote {
// no escapes in this string
s := p.str[p.pos:nextDoubleQuote]
p.pos = nextDoubleQuote + 1
return s, nil
}
// slow path: string contains escapes
s, err := p.consumeDoubleQuotedWithEscapes(p.nextBackslash)
p.nextBackslash = strings.IndexByte(p.str[p.pos:], '\\')
if p.nextBackslash != -1 {
p.nextBackslash += p.pos
}
return s, err
}
// consumeDoubleQuotedWithEscapes consumes a double-quoted string containing escapes, starting
// at p.pos, and with the first backslash at firstBackslash.
func (p *hstoreParser) consumeDoubleQuotedWithEscapes(firstBackslash int) (string, error) {
// copy the prefix that does not contain backslashes
var builder strings.Builder
builder.WriteString(p.str[p.pos:firstBackslash])
// skip to the backslash
p.pos = firstBackslash
// copy bytes until the end, unescaping backslashes
for {
nextB, end := p.consume()
if end {
return "", errEOSInQuoted
} else if nextB == '"' {
break
} else if nextB == '\\' {
// escape: skip the backslash and copy the char
nextB, end = p.consume()
if end {
return "", errEOSInQuoted
}
if !(nextB == '\\' || nextB == '"') {
return "", fmt.Errorf("unexpected escape in quoted string: found '%#v'", nextB)
}
builder.WriteByte(nextB)
} else {
// normal byte: copy it
builder.WriteByte(nextB)
}
}
return builder.String(), nil
}
// consumePairSeparator consumes the Hstore pair separator ", " or returns an error.
func (p *hstoreParser) consumePairSeparator() error {
return p.consumeExpected2(',', ' ')
}
// consumeKVSeparator consumes the Hstore key/value separator "=>" or returns an error.
func (p *hstoreParser) consumeKVSeparator() error {
return p.consumeExpected2('=', '>')
}
// consumeDoubleQuotedOrNull consumes the Hstore key/value separator "=>" or returns an error.
func (p *hstoreParser) consumeDoubleQuotedOrNull() (Text, error) {
// peek at the next byte
if p.atEnd() {
return Text{}, errors.New("found end instead of value")
}
next := p.str[p.pos]
if next == 'N' {
// must be the exact string NULL: use consumeExpected2 twice
err := p.consumeExpected2('N', 'U')
if err != nil {
return Text{}, err
}
err = p.consumeExpected2('L', 'L')
if err != nil {
return Text{}, err
}
return Text{String: "", Valid: false}, nil
} else if next != '"' {
return Text{}, unexpectedByteErr(next, '"')
}
// skip the double quote
p.pos += 1
s, err := p.consumeDoubleQuoted()
if err != nil {
return Text{}, err
}
return Text{String: s, Valid: true}, nil
}
func parseHstore(s string) (Hstore, error) {
p := newHSP(s)
// This is an over-estimate of the number of key/value pairs. Use '>' because I am guessing it
// is less likely to occur in keys/values than '=' or ','.
numPairsEstimate := strings.Count(s, ">")
// makes one allocation of strings for the entire Hstore, rather than one allocation per value.
valueStrings := make([]string, 0, numPairsEstimate)
result := make(Hstore, numPairsEstimate)
first := true
for !p.atEnd() {
if !first {
err := p.consumePairSeparator()
if err != nil {
return nil, err
}
} else {
first = false
}
err := p.consumeExpectedByte('"')
if err != nil {
return nil, err
}
key, err := p.consumeDoubleQuoted()
if err != nil {
return nil, err
}
err = p.consumeKVSeparator()
if err != nil {
return nil, err
}
value, err := p.consumeDoubleQuotedOrNull()
if err != nil {
return nil, err
}
if value.Valid {
valueStrings = append(valueStrings, value.String)
result[key] = &valueStrings[len(valueStrings)-1]
} else {
result[key] = nil
}
}
return result, nil
}