matterbridge/vendor/github.com/minio/minio-go/v7/api-select.go

759 lines
21 KiB
Go
Raw Normal View History

2021-10-16 23:11:32 +02:00
/*
* MinIO Go Library for Amazon S3 Compatible Cloud Storage
* (C) 2018-2020 MinIO, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package minio
import (
"bytes"
"context"
"encoding/binary"
"encoding/xml"
"errors"
"fmt"
"hash"
"hash/crc32"
"io"
"net/http"
"net/url"
"strings"
"github.com/minio/minio-go/v7/pkg/encrypt"
"github.com/minio/minio-go/v7/pkg/s3utils"
)
// CSVFileHeaderInfo - is the parameter for whether to utilize headers.
type CSVFileHeaderInfo string
// Constants for file header info.
const (
CSVFileHeaderInfoNone CSVFileHeaderInfo = "NONE"
CSVFileHeaderInfoIgnore = "IGNORE"
CSVFileHeaderInfoUse = "USE"
)
// SelectCompressionType - is the parameter for what type of compression is
// present
type SelectCompressionType string
// Constants for compression types under select API.
const (
SelectCompressionNONE SelectCompressionType = "NONE"
SelectCompressionGZIP = "GZIP"
SelectCompressionBZIP = "BZIP2"
2021-12-12 00:05:15 +01:00
// Non-standard compression schemes, supported by MinIO hosts:
SelectCompressionZSTD = "ZSTD" // Zstandard compression.
SelectCompressionLZ4 = "LZ4" // LZ4 Stream
SelectCompressionS2 = "S2" // S2 Stream
SelectCompressionSNAPPY = "SNAPPY" // Snappy stream
2021-10-16 23:11:32 +02:00
)
// CSVQuoteFields - is the parameter for how CSV fields are quoted.
type CSVQuoteFields string
// Constants for csv quote styles.
const (
CSVQuoteFieldsAlways CSVQuoteFields = "Always"
CSVQuoteFieldsAsNeeded = "AsNeeded"
)
// QueryExpressionType - is of what syntax the expression is, this should only
// be SQL
type QueryExpressionType string
// Constants for expression type.
const (
QueryExpressionTypeSQL QueryExpressionType = "SQL"
)
// JSONType determines json input serialization type.
type JSONType string
// Constants for JSONTypes.
const (
JSONDocumentType JSONType = "DOCUMENT"
JSONLinesType = "LINES"
)
// ParquetInputOptions parquet input specific options
type ParquetInputOptions struct{}
// CSVInputOptions csv input specific options
type CSVInputOptions struct {
FileHeaderInfo CSVFileHeaderInfo
fileHeaderInfoSet bool
RecordDelimiter string
recordDelimiterSet bool
FieldDelimiter string
fieldDelimiterSet bool
QuoteCharacter string
quoteCharacterSet bool
QuoteEscapeCharacter string
quoteEscapeCharacterSet bool
Comments string
commentsSet bool
}
// SetFileHeaderInfo sets the file header info in the CSV input options
func (c *CSVInputOptions) SetFileHeaderInfo(val CSVFileHeaderInfo) {
c.FileHeaderInfo = val
c.fileHeaderInfoSet = true
}
// SetRecordDelimiter sets the record delimiter in the CSV input options
func (c *CSVInputOptions) SetRecordDelimiter(val string) {
c.RecordDelimiter = val
c.recordDelimiterSet = true
}
// SetFieldDelimiter sets the field delimiter in the CSV input options
func (c *CSVInputOptions) SetFieldDelimiter(val string) {
c.FieldDelimiter = val
c.fieldDelimiterSet = true
}
// SetQuoteCharacter sets the quote character in the CSV input options
func (c *CSVInputOptions) SetQuoteCharacter(val string) {
c.QuoteCharacter = val
c.quoteCharacterSet = true
}
// SetQuoteEscapeCharacter sets the quote escape character in the CSV input options
func (c *CSVInputOptions) SetQuoteEscapeCharacter(val string) {
c.QuoteEscapeCharacter = val
c.quoteEscapeCharacterSet = true
}
// SetComments sets the comments character in the CSV input options
func (c *CSVInputOptions) SetComments(val string) {
c.Comments = val
c.commentsSet = true
}
// MarshalXML - produces the xml representation of the CSV input options struct
func (c CSVInputOptions) MarshalXML(e *xml.Encoder, start xml.StartElement) error {
if err := e.EncodeToken(start); err != nil {
return err
}
if c.FileHeaderInfo != "" || c.fileHeaderInfoSet {
if err := e.EncodeElement(c.FileHeaderInfo, xml.StartElement{Name: xml.Name{Local: "FileHeaderInfo"}}); err != nil {
return err
}
}
if c.RecordDelimiter != "" || c.recordDelimiterSet {
if err := e.EncodeElement(c.RecordDelimiter, xml.StartElement{Name: xml.Name{Local: "RecordDelimiter"}}); err != nil {
return err
}
}
if c.FieldDelimiter != "" || c.fieldDelimiterSet {
if err := e.EncodeElement(c.FieldDelimiter, xml.StartElement{Name: xml.Name{Local: "FieldDelimiter"}}); err != nil {
return err
}
}
if c.QuoteCharacter != "" || c.quoteCharacterSet {
if err := e.EncodeElement(c.QuoteCharacter, xml.StartElement{Name: xml.Name{Local: "QuoteCharacter"}}); err != nil {
return err
}
}
if c.QuoteEscapeCharacter != "" || c.quoteEscapeCharacterSet {
if err := e.EncodeElement(c.QuoteEscapeCharacter, xml.StartElement{Name: xml.Name{Local: "QuoteEscapeCharacter"}}); err != nil {
return err
}
}
if c.Comments != "" || c.commentsSet {
if err := e.EncodeElement(c.Comments, xml.StartElement{Name: xml.Name{Local: "Comments"}}); err != nil {
return err
}
}
return e.EncodeToken(xml.EndElement{Name: start.Name})
}
// CSVOutputOptions csv output specific options
type CSVOutputOptions struct {
QuoteFields CSVQuoteFields
quoteFieldsSet bool
RecordDelimiter string
recordDelimiterSet bool
FieldDelimiter string
fieldDelimiterSet bool
QuoteCharacter string
quoteCharacterSet bool
QuoteEscapeCharacter string
quoteEscapeCharacterSet bool
}
// SetQuoteFields sets the quote field parameter in the CSV output options
func (c *CSVOutputOptions) SetQuoteFields(val CSVQuoteFields) {
c.QuoteFields = val
c.quoteFieldsSet = true
}
// SetRecordDelimiter sets the record delimiter character in the CSV output options
func (c *CSVOutputOptions) SetRecordDelimiter(val string) {
c.RecordDelimiter = val
c.recordDelimiterSet = true
}
// SetFieldDelimiter sets the field delimiter character in the CSV output options
func (c *CSVOutputOptions) SetFieldDelimiter(val string) {
c.FieldDelimiter = val
c.fieldDelimiterSet = true
}
// SetQuoteCharacter sets the quote character in the CSV output options
func (c *CSVOutputOptions) SetQuoteCharacter(val string) {
c.QuoteCharacter = val
c.quoteCharacterSet = true
}
// SetQuoteEscapeCharacter sets the quote escape character in the CSV output options
func (c *CSVOutputOptions) SetQuoteEscapeCharacter(val string) {
c.QuoteEscapeCharacter = val
c.quoteEscapeCharacterSet = true
}
// MarshalXML - produces the xml representation of the CSVOutputOptions struct
func (c CSVOutputOptions) MarshalXML(e *xml.Encoder, start xml.StartElement) error {
if err := e.EncodeToken(start); err != nil {
return err
}
if c.QuoteFields != "" || c.quoteFieldsSet {
if err := e.EncodeElement(c.QuoteFields, xml.StartElement{Name: xml.Name{Local: "QuoteFields"}}); err != nil {
return err
}
}
if c.RecordDelimiter != "" || c.recordDelimiterSet {
if err := e.EncodeElement(c.RecordDelimiter, xml.StartElement{Name: xml.Name{Local: "RecordDelimiter"}}); err != nil {
return err
}
}
if c.FieldDelimiter != "" || c.fieldDelimiterSet {
if err := e.EncodeElement(c.FieldDelimiter, xml.StartElement{Name: xml.Name{Local: "FieldDelimiter"}}); err != nil {
return err
}
}
if c.QuoteCharacter != "" || c.quoteCharacterSet {
if err := e.EncodeElement(c.QuoteCharacter, xml.StartElement{Name: xml.Name{Local: "QuoteCharacter"}}); err != nil {
return err
}
}
if c.QuoteEscapeCharacter != "" || c.quoteEscapeCharacterSet {
if err := e.EncodeElement(c.QuoteEscapeCharacter, xml.StartElement{Name: xml.Name{Local: "QuoteEscapeCharacter"}}); err != nil {
return err
}
}
return e.EncodeToken(xml.EndElement{Name: start.Name})
}
// JSONInputOptions json input specific options
type JSONInputOptions struct {
Type JSONType
typeSet bool
}
// SetType sets the JSON type in the JSON input options
func (j *JSONInputOptions) SetType(typ JSONType) {
j.Type = typ
j.typeSet = true
}
// MarshalXML - produces the xml representation of the JSONInputOptions struct
func (j JSONInputOptions) MarshalXML(e *xml.Encoder, start xml.StartElement) error {
if err := e.EncodeToken(start); err != nil {
return err
}
if j.Type != "" || j.typeSet {
if err := e.EncodeElement(j.Type, xml.StartElement{Name: xml.Name{Local: "Type"}}); err != nil {
return err
}
}
return e.EncodeToken(xml.EndElement{Name: start.Name})
}
// JSONOutputOptions - json output specific options
type JSONOutputOptions struct {
RecordDelimiter string
recordDelimiterSet bool
}
// SetRecordDelimiter sets the record delimiter in the JSON output options
func (j *JSONOutputOptions) SetRecordDelimiter(val string) {
j.RecordDelimiter = val
j.recordDelimiterSet = true
}
// MarshalXML - produces the xml representation of the JSONOutputOptions struct
func (j JSONOutputOptions) MarshalXML(e *xml.Encoder, start xml.StartElement) error {
if err := e.EncodeToken(start); err != nil {
return err
}
if j.RecordDelimiter != "" || j.recordDelimiterSet {
if err := e.EncodeElement(j.RecordDelimiter, xml.StartElement{Name: xml.Name{Local: "RecordDelimiter"}}); err != nil {
return err
}
}
return e.EncodeToken(xml.EndElement{Name: start.Name})
}
// SelectObjectInputSerialization - input serialization parameters
type SelectObjectInputSerialization struct {
2021-12-12 00:05:15 +01:00
CompressionType SelectCompressionType `xml:"CompressionType,omitempty"`
Parquet *ParquetInputOptions `xml:"Parquet,omitempty"`
CSV *CSVInputOptions `xml:"CSV,omitempty"`
JSON *JSONInputOptions `xml:"JSON,omitempty"`
2021-10-16 23:11:32 +02:00
}
// SelectObjectOutputSerialization - output serialization parameters.
type SelectObjectOutputSerialization struct {
CSV *CSVOutputOptions `xml:"CSV,omitempty"`
JSON *JSONOutputOptions `xml:"JSON,omitempty"`
}
// SelectObjectOptions - represents the input select body
type SelectObjectOptions struct {
XMLName xml.Name `xml:"SelectObjectContentRequest" json:"-"`
ServerSideEncryption encrypt.ServerSide `xml:"-"`
Expression string
ExpressionType QueryExpressionType
InputSerialization SelectObjectInputSerialization
OutputSerialization SelectObjectOutputSerialization
RequestProgress struct {
Enabled bool
}
}
// Header returns the http.Header representation of the SelectObject options.
func (o SelectObjectOptions) Header() http.Header {
headers := make(http.Header)
if o.ServerSideEncryption != nil && o.ServerSideEncryption.Type() == encrypt.SSEC {
o.ServerSideEncryption.Marshal(headers)
}
return headers
}
// SelectObjectType - is the parameter which defines what type of object the
// operation is being performed on.
type SelectObjectType string
// Constants for input data types.
const (
SelectObjectTypeCSV SelectObjectType = "CSV"
SelectObjectTypeJSON = "JSON"
SelectObjectTypeParquet = "Parquet"
)
// preludeInfo is used for keeping track of necessary information from the
// prelude.
type preludeInfo struct {
totalLen uint32
headerLen uint32
}
// SelectResults is used for the streaming responses from the server.
type SelectResults struct {
pipeReader *io.PipeReader
resp *http.Response
stats *StatsMessage
progress *ProgressMessage
}
// ProgressMessage is a struct for progress xml message.
type ProgressMessage struct {
XMLName xml.Name `xml:"Progress" json:"-"`
StatsMessage
}
// StatsMessage is a struct for stat xml message.
type StatsMessage struct {
XMLName xml.Name `xml:"Stats" json:"-"`
BytesScanned int64
BytesProcessed int64
BytesReturned int64
}
// messageType represents the type of message.
type messageType string
const (
errorMsg messageType = "error"
commonMsg = "event"
)
// eventType represents the type of event.
type eventType string
// list of event-types returned by Select API.
const (
endEvent eventType = "End"
recordsEvent = "Records"
progressEvent = "Progress"
statsEvent = "Stats"
)
// contentType represents content type of event.
type contentType string
const (
xmlContent contentType = "text/xml"
)
// SelectObjectContent is a implementation of http://docs.aws.amazon.com/AmazonS3/latest/API/RESTObjectSELECTContent.html AWS S3 API.
func (c *Client) SelectObjectContent(ctx context.Context, bucketName, objectName string, opts SelectObjectOptions) (*SelectResults, error) {
2021-10-16 23:11:32 +02:00
// Input validation.
if err := s3utils.CheckValidBucketName(bucketName); err != nil {
return nil, err
}
if err := s3utils.CheckValidObjectName(objectName); err != nil {
return nil, err
}
selectReqBytes, err := xml.Marshal(opts)
if err != nil {
return nil, err
}
urlValues := make(url.Values)
urlValues.Set("select", "")
urlValues.Set("select-type", "2")
// Execute POST on bucket/object.
resp, err := c.executeMethod(ctx, http.MethodPost, requestMetadata{
bucketName: bucketName,
objectName: objectName,
queryValues: urlValues,
customHeader: opts.Header(),
contentMD5Base64: sumMD5Base64(selectReqBytes),
contentSHA256Hex: sum256Hex(selectReqBytes),
contentBody: bytes.NewReader(selectReqBytes),
contentLength: int64(len(selectReqBytes)),
})
if err != nil {
return nil, err
}
return NewSelectResults(resp, bucketName)
}
// NewSelectResults creates a Select Result parser that parses the response
// and returns a Reader that will return parsed and assembled select output.
func NewSelectResults(resp *http.Response, bucketName string) (*SelectResults, error) {
if resp.StatusCode != http.StatusOK {
return nil, httpRespToErrorResponse(resp, bucketName, "")
}
pipeReader, pipeWriter := io.Pipe()
streamer := &SelectResults{
resp: resp,
stats: &StatsMessage{},
progress: &ProgressMessage{},
pipeReader: pipeReader,
}
streamer.start(pipeWriter)
return streamer, nil
}
// Close - closes the underlying response body and the stream reader.
func (s *SelectResults) Close() error {
defer closeResponse(s.resp)
return s.pipeReader.Close()
}
// Read - is a reader compatible implementation for SelectObjectContent records.
func (s *SelectResults) Read(b []byte) (n int, err error) {
return s.pipeReader.Read(b)
}
// Stats - information about a request's stats when processing is complete.
func (s *SelectResults) Stats() *StatsMessage {
return s.stats
}
// Progress - information about the progress of a request.
func (s *SelectResults) Progress() *ProgressMessage {
return s.progress
}
// start is the main function that decodes the large byte array into
// several events that are sent through the eventstream.
func (s *SelectResults) start(pipeWriter *io.PipeWriter) {
go func() {
for {
var prelude preludeInfo
var headers = make(http.Header)
var err error
// Create CRC code
crc := crc32.New(crc32.IEEETable)
crcReader := io.TeeReader(s.resp.Body, crc)
// Extract the prelude(12 bytes) into a struct to extract relevant information.
prelude, err = processPrelude(crcReader, crc)
if err != nil {
pipeWriter.CloseWithError(err)
closeResponse(s.resp)
return
}
// Extract the headers(variable bytes) into a struct to extract relevant information
if prelude.headerLen > 0 {
if err = extractHeader(io.LimitReader(crcReader, int64(prelude.headerLen)), headers); err != nil {
pipeWriter.CloseWithError(err)
closeResponse(s.resp)
return
}
}
// Get the actual payload length so that the appropriate amount of
// bytes can be read or parsed.
payloadLen := prelude.PayloadLen()
m := messageType(headers.Get("message-type"))
switch m {
case errorMsg:
pipeWriter.CloseWithError(errors.New(headers.Get("error-code") + ":\"" + headers.Get("error-message") + "\""))
closeResponse(s.resp)
return
case commonMsg:
// Get content-type of the payload.
c := contentType(headers.Get("content-type"))
// Get event type of the payload.
e := eventType(headers.Get("event-type"))
// Handle all supported events.
switch e {
case endEvent:
pipeWriter.Close()
closeResponse(s.resp)
return
case recordsEvent:
if _, err = io.Copy(pipeWriter, io.LimitReader(crcReader, payloadLen)); err != nil {
pipeWriter.CloseWithError(err)
closeResponse(s.resp)
return
}
case progressEvent:
switch c {
case xmlContent:
if err = xmlDecoder(io.LimitReader(crcReader, payloadLen), s.progress); err != nil {
pipeWriter.CloseWithError(err)
closeResponse(s.resp)
return
}
default:
pipeWriter.CloseWithError(fmt.Errorf("Unexpected content-type %s sent for event-type %s", c, progressEvent))
closeResponse(s.resp)
return
}
case statsEvent:
switch c {
case xmlContent:
if err = xmlDecoder(io.LimitReader(crcReader, payloadLen), s.stats); err != nil {
pipeWriter.CloseWithError(err)
closeResponse(s.resp)
return
}
default:
pipeWriter.CloseWithError(fmt.Errorf("Unexpected content-type %s sent for event-type %s", c, statsEvent))
closeResponse(s.resp)
return
}
}
}
// Ensures that the full message's CRC is correct and
// that the message is not corrupted
if err := checkCRC(s.resp.Body, crc.Sum32()); err != nil {
pipeWriter.CloseWithError(err)
closeResponse(s.resp)
return
}
}
}()
}
// PayloadLen is a function that calculates the length of the payload.
func (p preludeInfo) PayloadLen() int64 {
return int64(p.totalLen - p.headerLen - 16)
}
// processPrelude is the function that reads the 12 bytes of the prelude and
// ensures the CRC is correct while also extracting relevant information into
// the struct,
func processPrelude(prelude io.Reader, crc hash.Hash32) (preludeInfo, error) {
var err error
var pInfo = preludeInfo{}
// reads total length of the message (first 4 bytes)
pInfo.totalLen, err = extractUint32(prelude)
if err != nil {
return pInfo, err
}
// reads total header length of the message (2nd 4 bytes)
pInfo.headerLen, err = extractUint32(prelude)
if err != nil {
return pInfo, err
}
// checks that the CRC is correct (3rd 4 bytes)
preCRC := crc.Sum32()
if err := checkCRC(prelude, preCRC); err != nil {
return pInfo, err
}
return pInfo, nil
}
// extracts the relevant information from the Headers.
func extractHeader(body io.Reader, myHeaders http.Header) error {
for {
// extracts the first part of the header,
headerTypeName, err := extractHeaderType(body)
if err != nil {
// Since end of file, we have read all of our headers
if err == io.EOF {
break
}
return err
}
// reads the 7 present in the header and ignores it.
extractUint8(body)
headerValueName, err := extractHeaderValue(body)
if err != nil {
return err
}
myHeaders.Set(headerTypeName, headerValueName)
}
return nil
}
// extractHeaderType extracts the first half of the header message, the header type.
func extractHeaderType(body io.Reader) (string, error) {
// extracts 2 bit integer
headerNameLen, err := extractUint8(body)
if err != nil {
return "", err
}
// extracts the string with the appropriate number of bytes
headerName, err := extractString(body, int(headerNameLen))
if err != nil {
return "", err
}
return strings.TrimPrefix(headerName, ":"), nil
}
// extractsHeaderValue extracts the second half of the header message, the
// header value
func extractHeaderValue(body io.Reader) (string, error) {
bodyLen, err := extractUint16(body)
if err != nil {
return "", err
}
bodyName, err := extractString(body, int(bodyLen))
if err != nil {
return "", err
}
return bodyName, nil
}
// extracts a string from byte array of a particular number of bytes.
func extractString(source io.Reader, lenBytes int) (string, error) {
myVal := make([]byte, lenBytes)
_, err := source.Read(myVal)
if err != nil {
return "", err
}
return string(myVal), nil
}
// extractUint32 extracts a 4 byte integer from the byte array.
func extractUint32(r io.Reader) (uint32, error) {
buf := make([]byte, 4)
_, err := readFull(r, buf)
if err != nil {
return 0, err
}
return binary.BigEndian.Uint32(buf), nil
}
// extractUint16 extracts a 2 byte integer from the byte array.
func extractUint16(r io.Reader) (uint16, error) {
buf := make([]byte, 2)
_, err := readFull(r, buf)
if err != nil {
return 0, err
}
return binary.BigEndian.Uint16(buf), nil
}
// extractUint8 extracts a 1 byte integer from the byte array.
func extractUint8(r io.Reader) (uint8, error) {
buf := make([]byte, 1)
_, err := readFull(r, buf)
if err != nil {
return 0, err
}
return buf[0], nil
}
// checkCRC ensures that the CRC matches with the one from the reader.
func checkCRC(r io.Reader, expect uint32) error {
msgCRC, err := extractUint32(r)
if err != nil {
return err
}
if msgCRC != expect {
return fmt.Errorf("Checksum Mismatch, MessageCRC of 0x%X does not equal expected CRC of 0x%X", msgCRC, expect)
}
return nil
}