minio/internal/s3select/select.go
Harshavardhana 754f7a8a39
replace io.Discard usage to fix some NUMA copy() latencies (#18394)
replace io.Discard usage to fix NUMA copy() latencies

On NUMA systems copying from 8K buffer allocated via
io.Discard leads to large latency build-up for every

```
copy(new8kbuf, largebuf)
```

can in-cur upto 1ms worth of latencies on NUMA systems
due to memory sharding across NUMA nodes.
2023-11-06 14:26:08 -08:00

773 lines
21 KiB
Go

// Copyright (c) 2015-2021 MinIO, Inc.
//
// This file is part of MinIO Object Storage stack
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU Affero General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU Affero General Public License for more details.
//
// You should have received a copy of the GNU Affero General Public License
// along with this program. If not, see <http://www.gnu.org/licenses/>.
package s3select
import (
"bufio"
"bytes"
"compress/bzip2"
"encoding/xml"
"errors"
"fmt"
"io"
"net/http"
"strings"
"sync"
"github.com/klauspost/compress/s2"
"github.com/klauspost/compress/zstd"
gzip "github.com/klauspost/pgzip"
"github.com/minio/minio/internal/config"
xioutil "github.com/minio/minio/internal/ioutil"
"github.com/minio/minio/internal/s3select/csv"
"github.com/minio/minio/internal/s3select/json"
"github.com/minio/minio/internal/s3select/parquet"
"github.com/minio/minio/internal/s3select/simdj"
"github.com/minio/minio/internal/s3select/sql"
"github.com/minio/pkg/v2/env"
"github.com/minio/simdjson-go"
"github.com/pierrec/lz4"
)
type recordReader interface {
// Read a record.
// dst is optional but will be used if valid.
Read(dst sql.Record) (sql.Record, error)
Close() error
}
const (
csvFormat = "csv"
jsonFormat = "json"
parquetFormat = "parquet"
)
// CompressionType - represents value inside <CompressionType/> in request XML.
type CompressionType string
const (
noneType CompressionType = "none"
gzipType CompressionType = "GZIP"
bzip2Type CompressionType = "BZIP2"
zstdType CompressionType = "ZSTD"
lz4Type CompressionType = "LZ4"
s2Type CompressionType = "S2"
snappyType CompressionType = "SNAPPY"
)
const (
maxRecordSize = 1 << 20 // 1 MiB
)
var parquetSupport bool
func init() {
parquetSupport = env.Get("MINIO_API_SELECT_PARQUET", config.EnableOff) == config.EnableOn
}
var bufPool = sync.Pool{
New: func() interface{} {
// make a buffer with a reasonable capacity.
return bytes.NewBuffer(make([]byte, 0, maxRecordSize))
},
}
var bufioWriterPool = sync.Pool{
New: func() interface{} {
// io.Discard is just used to create the writer. Actual destination
// writer is set later by Reset() before using it.
return bufio.NewWriter(xioutil.Discard)
},
}
// UnmarshalXML - decodes XML data.
func (c *CompressionType) UnmarshalXML(d *xml.Decoder, start xml.StartElement) error {
var s string
if err := d.DecodeElement(&s, &start); err != nil {
return errMalformedXML(err)
}
parsedType := CompressionType(strings.ToUpper(s))
if s == "" || parsedType == "NONE" {
parsedType = noneType
}
switch parsedType {
case noneType, gzipType, bzip2Type, snappyType, s2Type, zstdType, lz4Type:
default:
return errInvalidCompressionFormat(fmt.Errorf("invalid compression format '%v'", s))
}
*c = parsedType
return nil
}
// InputSerialization - represents elements inside <InputSerialization/> in request XML.
type InputSerialization struct {
CompressionType CompressionType `xml:"CompressionType"`
CSVArgs csv.ReaderArgs `xml:"CSV"`
JSONArgs json.ReaderArgs `xml:"JSON"`
ParquetArgs parquet.ReaderArgs `xml:"Parquet"`
unmarshaled bool
format string
}
// IsEmpty - returns whether input serialization is empty or not.
func (input *InputSerialization) IsEmpty() bool {
return !input.unmarshaled
}
// UnmarshalXML - decodes XML data.
func (input *InputSerialization) UnmarshalXML(d *xml.Decoder, start xml.StartElement) error {
// Make subtype to avoid recursive UnmarshalXML().
type subInputSerialization InputSerialization
parsedInput := subInputSerialization{}
if err := d.DecodeElement(&parsedInput, &start); err != nil {
return errMalformedXML(err)
}
// If no compression is specified, set to noneType
if parsedInput.CompressionType == "" {
parsedInput.CompressionType = noneType
}
found := 0
if !parsedInput.CSVArgs.IsEmpty() {
parsedInput.format = csvFormat
found++
}
if !parsedInput.JSONArgs.IsEmpty() {
parsedInput.format = jsonFormat
found++
}
if !parsedInput.ParquetArgs.IsEmpty() {
if parsedInput.CompressionType != "" && parsedInput.CompressionType != noneType {
return errInvalidRequestParameter(fmt.Errorf("CompressionType must be NONE for Parquet format"))
}
parsedInput.format = parquetFormat
found++
}
if found != 1 {
return errInvalidDataSource(nil)
}
*input = InputSerialization(parsedInput)
input.unmarshaled = true
return nil
}
// OutputSerialization - represents elements inside <OutputSerialization/> in request XML.
type OutputSerialization struct {
CSVArgs csv.WriterArgs `xml:"CSV"`
JSONArgs json.WriterArgs `xml:"JSON"`
unmarshaled bool
format string
}
// IsEmpty - returns whether output serialization is empty or not.
func (output *OutputSerialization) IsEmpty() bool {
return !output.unmarshaled
}
// UnmarshalXML - decodes XML data.
func (output *OutputSerialization) UnmarshalXML(d *xml.Decoder, start xml.StartElement) error {
// Make subtype to avoid recursive UnmarshalXML().
type subOutputSerialization OutputSerialization
parsedOutput := subOutputSerialization{}
if err := d.DecodeElement(&parsedOutput, &start); err != nil {
return errMalformedXML(err)
}
found := 0
if !parsedOutput.CSVArgs.IsEmpty() {
parsedOutput.format = csvFormat
found++
}
if !parsedOutput.JSONArgs.IsEmpty() {
parsedOutput.format = jsonFormat
found++
}
if found != 1 {
return errObjectSerializationConflict(fmt.Errorf("either CSV or JSON should be present in OutputSerialization"))
}
*output = OutputSerialization(parsedOutput)
output.unmarshaled = true
return nil
}
// RequestProgress - represents elements inside <RequestProgress/> in request XML.
type RequestProgress struct {
Enabled bool `xml:"Enabled"`
}
// ScanRange represents the ScanRange parameter.
type ScanRange struct {
// Start is the byte offset to read from (from the start of the file).
Start *uint64 `xml:"Start"`
// End is the offset of the last byte that should be returned when Start
// is set, otherwise it is the offset from EOF to start reading.
End *uint64 `xml:"End"`
}
// Validate if the scan range is valid.
func (s *ScanRange) Validate() error {
if s == nil {
return nil
}
if s.Start == nil && s.End == nil {
// This parameter is optional, but when specified, it must not be empty.
// Ref: https://docs.aws.amazon.com/AmazonS3/latest/API/API_SelectObjectContent.html#AmazonS3-SelectObjectContent-request-ScanRange
return errors.New("ScanRange: No Start or End specified")
}
if s.Start == nil || s.End == nil {
return nil
}
if *s.Start > *s.End {
return errors.New("ScanRange: Start cannot be after end")
}
return nil
}
// StartLen returns start offset plus length from range.
func (s *ScanRange) StartLen() (start, length int64, err error) {
if s == nil {
return 0, -1, nil
}
err = s.Validate()
if err != nil {
return 0, 0, err
}
if s.End == nil && s.Start == nil {
// Not valid, but should be caught above.
return 0, -1, nil
}
if s.End == nil {
start := int64(*s.Start)
if start < 0 {
return 0, 0, errors.New("ScanRange: Start after EOF")
}
return start, -1, nil
}
if s.Start == nil {
// Suffix length
end := int64(*s.End)
if end < 0 {
return 0, 0, errors.New("ScanRange: End bigger than file")
}
// Suffix length
return -end, -1, nil
}
start = int64(*s.Start)
end := int64(*s.End)
return start, end - start + 1, nil
}
// S3Select - filters the contents on a simple structured query language (SQL) statement. It
// represents elements inside <SelectRequest/> in request XML specified in detail at
// https://docs.aws.amazon.com/AmazonS3/latest/API/RESTObjectSELECTContent.html.
type S3Select struct {
XMLName xml.Name `xml:"SelectRequest"`
Expression string `xml:"Expression"`
ExpressionType string `xml:"ExpressionType"`
Input InputSerialization `xml:"InputSerialization"`
Output OutputSerialization `xml:"OutputSerialization"`
Progress RequestProgress `xml:"RequestProgress"`
ScanRange *ScanRange `xml:"ScanRange"`
statement *sql.SelectStatement
progressReader *progressReader
recordReader recordReader
}
var legacyXMLName = "SelectObjectContentRequest"
// UnmarshalXML - decodes XML data.
func (s3Select *S3Select) UnmarshalXML(d *xml.Decoder, start xml.StartElement) error {
// S3 also supports the older SelectObjectContentRequest tag,
// though it is no longer found in documentation. This is
// checked and renamed below to allow older clients to also
// work.
if start.Name.Local == legacyXMLName {
start.Name = xml.Name{Space: "", Local: "SelectRequest"}
}
// Make subtype to avoid recursive UnmarshalXML().
type subS3Select S3Select
parsedS3Select := subS3Select{}
if err := d.DecodeElement(&parsedS3Select, &start); err != nil {
if _, ok := err.(*s3Error); ok {
return err
}
return errMalformedXML(err)
}
if err := parsedS3Select.ScanRange.Validate(); err != nil {
return errInvalidScanRangeParameter(err)
}
parsedS3Select.ExpressionType = strings.ToLower(parsedS3Select.ExpressionType)
if parsedS3Select.ExpressionType != "sql" {
return errInvalidExpressionType(fmt.Errorf("invalid expression type '%v'", parsedS3Select.ExpressionType))
}
if parsedS3Select.Input.IsEmpty() {
return errMissingRequiredParameter(fmt.Errorf("InputSerialization must be provided"))
}
if parsedS3Select.Output.IsEmpty() {
return errMissingRequiredParameter(fmt.Errorf("OutputSerialization must be provided"))
}
statement, err := sql.ParseSelectStatement(parsedS3Select.Expression)
if err != nil {
return err
}
parsedS3Select.statement = &statement
*s3Select = S3Select(parsedS3Select)
return nil
}
func (s3Select *S3Select) outputRecord() sql.Record {
switch s3Select.Output.format {
case csvFormat:
return csv.NewRecord()
case jsonFormat:
return json.NewRecord(sql.SelectFmtJSON)
}
panic(fmt.Errorf("unknown output format '%v'", s3Select.Output.format))
}
func (s3Select *S3Select) getProgress() (bytesScanned, bytesProcessed int64) {
if s3Select.progressReader != nil {
return s3Select.progressReader.Stats()
}
return -1, -1
}
// Open - opens S3 object by using callback for SQL selection query.
// Currently CSV, JSON and Apache Parquet formats are supported.
func (s3Select *S3Select) Open(rsc io.ReadSeekCloser) error {
offset, length, err := s3Select.ScanRange.StartLen()
if err != nil {
return err
}
seekDirection := io.SeekStart
if offset < 0 {
seekDirection = io.SeekEnd
}
switch s3Select.Input.format {
case csvFormat:
_, err = rsc.Seek(offset, seekDirection)
if err != nil {
return err
}
var rc io.ReadCloser = rsc
if length != -1 {
rc = newLimitedReadCloser(rsc, length)
}
s3Select.progressReader, err = newProgressReader(rc, s3Select.Input.CompressionType)
if err != nil {
rsc.Close()
return err
}
s3Select.recordReader, err = csv.NewReader(s3Select.progressReader, &s3Select.Input.CSVArgs)
if err != nil {
// Close all reader resources opened so far.
s3Select.progressReader.Close()
var stErr bzip2.StructuralError
if errors.As(err, &stErr) {
return errInvalidCompression(err, s3Select.Input.CompressionType)
}
// Test these compressor errors
errs := []error{
gzip.ErrHeader, gzip.ErrChecksum,
s2.ErrCorrupt, s2.ErrUnsupported, s2.ErrCRC,
zstd.ErrBlockTooSmall, zstd.ErrMagicMismatch, zstd.ErrWindowSizeExceeded, zstd.ErrUnknownDictionary, zstd.ErrWindowSizeTooSmall,
lz4.ErrInvalid, lz4.ErrBlockDependency,
}
for _, e := range errs {
if errors.Is(err, e) {
return errInvalidCompression(err, s3Select.Input.CompressionType)
}
}
return err
}
return nil
case jsonFormat:
_, err = rsc.Seek(offset, seekDirection)
if err != nil {
return err
}
var rc io.ReadCloser = rsc
if length != -1 {
rc = newLimitedReadCloser(rsc, length)
}
s3Select.progressReader, err = newProgressReader(rc, s3Select.Input.CompressionType)
if err != nil {
rsc.Close()
return err
}
if strings.EqualFold(s3Select.Input.JSONArgs.ContentType, "lines") {
if simdjson.SupportedCPU() {
s3Select.recordReader = simdj.NewReader(s3Select.progressReader, &s3Select.Input.JSONArgs)
} else {
s3Select.recordReader = json.NewPReader(s3Select.progressReader, &s3Select.Input.JSONArgs)
}
} else {
s3Select.recordReader = json.NewReader(s3Select.progressReader, &s3Select.Input.JSONArgs)
}
return nil
case parquetFormat:
if !parquetSupport {
return errors.New("parquet format parsing not enabled on server")
}
if offset != 0 || length != -1 {
// Offsets do not make sense in parquet files.
return errors.New("parquet format does not support offsets")
}
var err error
s3Select.recordReader, err = parquet.NewParquetReader(rsc, &s3Select.Input.ParquetArgs)
return err
}
return fmt.Errorf("unknown input format '%v'", s3Select.Input.format)
}
func (s3Select *S3Select) marshal(buf *bytes.Buffer, record sql.Record) error {
switch s3Select.Output.format {
case csvFormat:
// Use bufio Writer to prevent csv.Writer from allocating a new buffer.
bufioWriter := bufioWriterPool.Get().(*bufio.Writer)
defer func() {
bufioWriter.Reset(xioutil.Discard)
bufioWriterPool.Put(bufioWriter)
}()
bufioWriter.Reset(buf)
opts := sql.WriteCSVOpts{
FieldDelimiter: []rune(s3Select.Output.CSVArgs.FieldDelimiter)[0],
Quote: []rune(s3Select.Output.CSVArgs.QuoteCharacter)[0],
QuoteEscape: []rune(s3Select.Output.CSVArgs.QuoteEscapeCharacter)[0],
AlwaysQuote: strings.EqualFold(s3Select.Output.CSVArgs.QuoteFields, "always"),
}
err := record.WriteCSV(bufioWriter, opts)
if err != nil {
return err
}
err = bufioWriter.Flush()
if err != nil {
return err
}
if buf.Bytes()[buf.Len()-1] == '\n' {
buf.Truncate(buf.Len() - 1)
}
buf.WriteString(s3Select.Output.CSVArgs.RecordDelimiter)
return nil
case jsonFormat:
err := record.WriteJSON(buf)
if err != nil {
return err
}
// Trim trailing newline from non-simd output
if buf.Bytes()[buf.Len()-1] == '\n' {
buf.Truncate(buf.Len() - 1)
}
buf.WriteString(s3Select.Output.JSONArgs.RecordDelimiter)
return nil
}
panic(fmt.Errorf("unknown output format '%v'", s3Select.Output.format))
}
// Evaluate - filters and sends records read from opened reader as per select statement to http response writer.
func (s3Select *S3Select) Evaluate(w http.ResponseWriter) {
getProgressFunc := s3Select.getProgress
if !s3Select.Progress.Enabled {
getProgressFunc = nil
}
writer := newMessageWriter(w, getProgressFunc)
var outputQueue []sql.Record
// Create queue based on the type.
if s3Select.statement.IsAggregated() {
outputQueue = make([]sql.Record, 0, 1)
} else {
outputQueue = make([]sql.Record, 0, 100)
}
var err error
sendRecord := func() bool {
buf := bufPool.Get().(*bytes.Buffer)
buf.Reset()
for _, outputRecord := range outputQueue {
if outputRecord == nil {
continue
}
before := buf.Len()
if err = s3Select.marshal(buf, outputRecord); err != nil {
bufPool.Put(buf)
return false
}
if buf.Len()-before > maxRecordSize {
writer.FinishWithError("OverMaxRecordSize", "The length of a record in the input or result is greater than maxCharsPerRecord of 1 MB.")
bufPool.Put(buf)
return false
}
}
if err = writer.SendRecord(buf); err != nil {
// FIXME: log this error.
err = nil
bufPool.Put(buf)
return false
}
outputQueue = outputQueue[:0]
return true
}
var rec sql.Record
OuterLoop:
for {
if s3Select.statement.LimitReached() {
if !sendRecord() {
break
}
if err = writer.Finish(s3Select.getProgress()); err != nil {
// FIXME: log this error.
err = nil
}
break
}
if rec, err = s3Select.recordReader.Read(rec); err != nil {
if err != io.EOF {
break
}
if s3Select.statement.IsAggregated() {
outputRecord := s3Select.outputRecord()
if err = s3Select.statement.AggregateResult(outputRecord); err != nil {
break
}
outputQueue = append(outputQueue, outputRecord)
}
if !sendRecord() {
break
}
if err = writer.Finish(s3Select.getProgress()); err != nil {
// FIXME: log this error.
err = nil
}
break
}
var inputRecords []*sql.Record
if inputRecords, err = s3Select.statement.EvalFrom(s3Select.Input.format, rec); err != nil {
break
}
for _, inputRecord := range inputRecords {
if s3Select.statement.IsAggregated() {
if err = s3Select.statement.AggregateRow(*inputRecord); err != nil {
break OuterLoop
}
} else {
var outputRecord sql.Record
// We will attempt to reuse the records in the table.
// The type of these should not change.
// The queue should always have at least one entry left for this to work.
outputQueue = outputQueue[:len(outputQueue)+1]
if t := outputQueue[len(outputQueue)-1]; t != nil {
// If the output record is already set, we reuse it.
outputRecord = t
outputRecord.Reset()
} else {
// Create new one
outputRecord = s3Select.outputRecord()
outputQueue[len(outputQueue)-1] = outputRecord
}
outputRecord, err = s3Select.statement.Eval(*inputRecord, outputRecord)
if outputRecord == nil || err != nil {
// This should not be written.
// Remove it from the queue.
outputQueue = outputQueue[:len(outputQueue)-1]
if err != nil {
break OuterLoop
}
continue
}
outputQueue[len(outputQueue)-1] = outputRecord
if s3Select.statement.LimitReached() {
if !sendRecord() {
break
}
if err = writer.Finish(s3Select.getProgress()); err != nil {
// FIXME: log this error.
err = nil
}
return
}
if len(outputQueue) < cap(outputQueue) {
continue
}
if !sendRecord() {
break OuterLoop
}
}
}
}
if err != nil {
_ = writer.FinishWithError("InternalError", err.Error())
}
}
// Close - closes opened S3 object.
func (s3Select *S3Select) Close() error {
if s3Select.recordReader == nil {
return nil
}
return s3Select.recordReader.Close()
}
// NewS3Select - creates new S3Select by given request XML reader.
func NewS3Select(r io.Reader) (*S3Select, error) {
s3Select := &S3Select{}
if err := xml.NewDecoder(r).Decode(s3Select); err != nil {
return nil, err
}
return s3Select, nil
}
//////////////////
// Helpers
/////////////////
// limitedReadCloser is like io.LimitedReader, but also implements io.Closer.
type limitedReadCloser struct {
io.LimitedReader
io.Closer
}
func newLimitedReadCloser(r io.ReadCloser, n int64) *limitedReadCloser {
return &limitedReadCloser{
LimitedReader: io.LimitedReader{R: r, N: n},
Closer: r,
}
}
// ObjectSegmentReaderFn is a function that returns a reader for a contiguous
// suffix segment of an object starting at the given (non-negative) offset.
type ObjectSegmentReaderFn func(offset int64) (io.ReadCloser, error)
// ObjectReadSeekCloser implements ReadSeekCloser interface for reading objects.
// It uses a function that returns a io.ReadCloser for the object.
type ObjectReadSeekCloser struct {
segmentReader ObjectSegmentReaderFn
size int64 // actual object size regardless of compression/encryption
offset int64
reader io.ReadCloser
// reader can be closed idempotently multiple times
closerOnce sync.Once
// Error storing reader.Close()
closerErr error
}
// NewObjectReadSeekCloser creates a new ObjectReadSeekCloser.
func NewObjectReadSeekCloser(segmentReader ObjectSegmentReaderFn, actualSize int64) *ObjectReadSeekCloser {
return &ObjectReadSeekCloser{
segmentReader: segmentReader,
size: actualSize,
offset: 0,
reader: nil,
}
}
// Seek call to implement io.Seeker
func (rsc *ObjectReadSeekCloser) Seek(offset int64, whence int) (int64, error) {
// fmt.Printf("actual: %v offset: %v (%v) whence: %v\n", rsc.size, offset, rsc.offset, whence)
switch whence {
case io.SeekStart:
rsc.offset = offset
case io.SeekCurrent:
rsc.offset += offset
case io.SeekEnd:
rsc.offset = rsc.size + offset
}
if rsc.offset < 0 {
return rsc.offset, errors.New("seek to invalid negative offset")
}
if rsc.offset >= rsc.size {
return rsc.offset, errors.New("seek past end of object")
}
if rsc.reader != nil {
_ = rsc.reader.Close()
rsc.reader = nil
}
return rsc.offset, nil
}
// Read call to implement io.Reader
func (rsc *ObjectReadSeekCloser) Read(p []byte) (n int, err error) {
if rsc.reader == nil {
rsc.reader, err = rsc.segmentReader(rsc.offset)
if err != nil {
return 0, err
}
}
return rsc.reader.Read(p)
}
// Close call to implement io.Closer. Calling Read/Seek after Close reopens the
// object for reading and a subsequent Close call is required to ensure
// resources are freed.
func (rsc *ObjectReadSeekCloser) Close() error {
rsc.closerOnce.Do(func() {
if rsc.reader != nil {
rsc.closerErr = rsc.reader.Close()
rsc.reader = nil
}
})
return rsc.closerErr
}