ocf_writer.go

// Copyright 2015 LinkedIn Corp. 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.Copyright [201X] LinkedIn Corp. 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.

package goavro

import (
	"bufio"
	"bytes"
	"compress/flate"
	"encoding/binary"
	"fmt"
	"hash/crc32"
	"io"
	"log"
	"math/rand"
	"time"

	"github.com/golang/snappy"
)

// DefaultWriterBlockSizeo specifies the default number of datum items
// in a block when writing.
const DefaultWriterBlockSize = 10

// ErrWriterInit is returned when an error is created during Writer
// initialization.
type ErrWriterInit struct {
	Message string
	Err     error
}

// Error converts the error instance to a string.
func (e *ErrWriterInit) Error() string {
	if e.Err == nil {
		return "cannot create Writer: " + e.Message
	} else if e.Message == "" {
		return "cannot create Writer: " + e.Err.Error()
	} else {
		return "cannot create Writer: " + e.Message + "; " + e.Err.Error()
	}
}

// WriterSetter functions are those those which are used to
// instantiate a new Writer.
type WriterSetter func(*Writer) error

// BlockSize specifies the default number of data items to be grouped
// in a block, compressed, and written to the stream.
//
// It is a valid use case to set both BlockTick and BlockSize. For
// example, if BlockTick is set to time.Minute and BlockSize is set to
// 20, but only 13 items are written to the Writer in a minute, those
// 13 items will be grouped in a block, compressed, and written to the
// stream without waiting for the addition 7 items to complete the
// BlockSize.
//
// By default, BlockSize is set to DefaultWriterBlockSize.
func BlockSize(blockSize int64) WriterSetter {
	return func(fw *Writer) error {
		if blockSize <= 0 {
			return fmt.Errorf("BlockSize must be larger than 0: %d", blockSize)
		}
		fw.blockSize = blockSize
		return nil
	}
}

// BlockTick specifies the duration of time between when the Writer
// will flush the blocks to the stream.
//
// It is a valid use case to set both BlockTick and BlockSize. For
// example, if BlockTick is set to time.Minute and BlockSize is set to
// 20, but only 13 items are written to the Writer in a minute, those
// 13 items will be grouped in a block, compressed, and written to the
// stream without waiting for the addition 7 items to complete the
// BlockSize.
//
// By default, BlockTick is set to 0 and is ignored. This causes the
// blocker to fill up its internal queue of data to BlockSize items
// before flushing them to the stream.
func BlockTick(blockTick time.Duration) WriterSetter {
	return func(fw *Writer) error {
		if blockTick < 0 {
			return fmt.Errorf("BlockTick must be non-negative time duration: %v", blockTick)
		}
		fw.blockTick = blockTick
		return nil
	}
}

// BufferToWriter specifies which io.Writer is the target of the
// Writer stream, and creates a bufio.Writer around that io.Writer. It
// is invalid to specify both BufferToWriter and ToWriter. Exactly one
// of these must be called for a given Writer initialization.
func BufferToWriter(w io.Writer) WriterSetter {
	return func(fw *Writer) error {
		fw.w = bufio.NewWriter(w)
		fw.buffered = true
		return nil
	}
}

// Compression is used to set the compression codec of
// a new Writer instance.
func Compression(someCompressionCodec string) WriterSetter {
	return func(fw *Writer) error {
		fw.CompressionCodec = someCompressionCodec
		return nil
	}
}

// Sync is used to set the sync marker bytes of a new
// instance. It checks to ensure the byte slice is 16 bytes long, but
// does not check that it has been set to something other than the
// zero value. Usually you can elide the `Sync` call and allow it
// to create a random byte sequence.
func Sync(someSync []byte) WriterSetter {
	return func(fw *Writer) error {
		if syncLength != len(someSync) {
			return fmt.Errorf("sync marker ought to be %d bytes long: %d", syncLength, len(someSync))
		}
		fw.Sync = make([]byte, syncLength)
		copy(fw.Sync, someSync)
		return nil
	}
}

// ToWriter specifies which io.Writer is the target of the Writer
// stream. It is invalid to specify both BufferToWriter and
// ToWriter. Exactly one of these must be called for a given Writer
// initialization.
func ToWriter(w io.Writer) WriterSetter {
	return func(fw *Writer) error {
		fw.w = w
		return nil
	}
}

// UseCodec specifies that a Writer should reuse an existing Codec
// rather than creating a new one, needlessly recompling the same
// schema.
func UseCodec(codec Codec) WriterSetter {
	return func(fw *Writer) error {
		if codec != nil {
			fw.dataCodec = codec
			return nil
		}
		return fmt.Errorf("invalid Codec")
	}
}

// WriterSchema is used to set the Avro schema of a new instance. If a
// codec has already been compiled for the schema, it is faster to use
// the UseCodec method instead of WriterSchema.
func WriterSchema(someSchema string) WriterSetter {
	return func(fw *Writer) (err error) {
		if fw.dataCodec, err = NewCodec(someSchema); err != nil {
			return
		}
		return
	}
}

// Writer structure contains data necessary to write Avro files.
type Writer struct {
	CompressionCodec string
	Sync             []byte
	blockSize        int64
	buffered         bool
	dataCodec        Codec
	err              error
	toBlock          chan interface{}
	w                io.Writer
	writerDone       chan struct{}
	blockTick        time.Duration
}

// NewWriter returns a object to write data to an io.Writer using the
// Avro Object Container Files format.
//
//     func serveClient(conn net.Conn, codec goavro.Codec) {
//         fw, err := goavro.NewWriter(
//             goavro.BlockSize(100),                 // flush data every 100 items
//             goavro.BlockTick(10 * time.Second),    // but at least every 10 seconds
//             goavro.Compression(goavro.CompressionSnappy),
//             goavro.ToWriter(conn),
//             goavro.UseCodec(codec))
//         if err != nil {
//             log.Fatal("cannot create Writer: ", err)
//         }
//         defer fw.Close()
//
//         // create a record that matches the schema we want to encode
//         someRecord, err := goavro.NewRecord(goavro.RecordSchema(recordSchema))
//         if err != nil {
//             log.Fatal(err)
//         }
//         // identify field name to set datum for
//         someRecord.Set("username", "Aquaman")
//         someRecord.Set("comment", "The Atlantic is oddly cold this morning!")
//         // you can fully qualify the field name
//         someRecord.Set("com.example.timestamp", int64(1082196484))
//         fw.Write(someRecord)
//
//         // create another record
//         someRecord, err = goavro.NewRecord(goavro.RecordSchema(recordSchema))
//         if err != nil {
//             log.Fatal(err)
//         }
//         someRecord.Set("username", "Batman")
//         someRecord.Set("comment", "Who are all of these crazies?")
//         someRecord.Set("com.example.timestamp", int64(1427383430))
//         fw.Write(someRecord)
//     }
func NewWriter(setters ...WriterSetter) (*Writer, error) {
	var err error
	fw := &Writer{CompressionCodec: CompressionNull, blockSize: DefaultWriterBlockSize}
	for _, setter := range setters {
		err = setter(fw)
		if err != nil {
			return nil, &ErrWriterInit{Err: err}
		}
	}
	if fw.w == nil {
		return nil, &ErrWriterInit{Message: "must specify io.Writer"}
	}
	// writer: stuff should already be initialized
	if !IsCompressionCodecSupported(fw.CompressionCodec) {
		return nil, &ErrWriterInit{Message: fmt.Sprintf("unsupported codec: %s", fw.CompressionCodec)}
	}
	if fw.dataCodec == nil {
		return nil, &ErrWriterInit{Message: "missing schema"}
	}
	if fw.Sync == nil {
		r := rand.New(rand.NewSource(time.Now().Unix()))

		// create random sequence of bytes for file sync marker
		fw.Sync = make([]byte, syncLength)
		for i := range fw.Sync {
			fw.Sync[i] = byte(r.Intn(256))
		}
	}
	if err = fw.writeHeader(); err != nil {
		return nil, &ErrWriterInit{Err: err}
	}
	// setup writing pipeline
	fw.toBlock = make(chan interface{})
	toEncode := make(chan *writerBlock)
	toCompress := make(chan *writerBlock)
	toWrite := make(chan *writerBlock)
	fw.writerDone = make(chan struct{})
	go blocker(fw, fw.toBlock, toEncode)
	go encoder(fw, toEncode, toCompress)
	go compressor(fw, toCompress, toWrite)
	go writer(fw, toWrite)
	return fw, nil
}

// Close is called when the open file is no longer needed. It flushes
// the bytes to the io.Writer if the file is being writtern.
func (fw *Writer) Close() error {
	close(fw.toBlock)
	<-fw.writerDone
	if fw.buffered {
		// NOTE: error that happened before Close has
		// precedence of buffer flush error
		err := fw.w.(*bufio.Writer).Flush()
		if fw.err == nil {
			return err
		}
	}
	return fw.err
}

// Write places a datum into the pipeline to be written to the Writer.
func (fw *Writer) Write(datum interface{}) {
	fw.toBlock <- datum
}

func (fw *Writer) writeHeader() (err error) {
	if _, err = fw.w.Write([]byte(magicBytes)); err != nil {
		return
	}
	// header metadata
	hm := make(map[string]interface{})
	hm["avro.schema"] = []byte(fw.dataCodec.Schema())
	hm["avro.codec"] = []byte(fw.CompressionCodec)
	if err = metadataCodec.Encode(fw.w, hm); err != nil {
		return
	}
	_, err = fw.w.Write(fw.Sync)
	return
}

type writerBlock struct {
	items      []interface{}
	encoded    *bytes.Buffer
	compressed []byte
	err        error
}

// NOTE: this is bad because it waits for enough items to show up
// before it starts encoding, rather than encode items as they arrive.

func blocker(fw *Writer, toBlock <-chan interface{}, toEncode chan<- *writerBlock) {
	items := make([]interface{}, 0, fw.blockSize)

	if fw.blockTick > 0 {
	blockerLoop:
		for {
			select {
			case item, more := <-toBlock:
				if !more {
					break blockerLoop
				}
				items = append(items, item)
				if int64(len(items)) >= fw.blockSize {
					toEncode <- &writerBlock{items: items}
					items = make([]interface{}, 0, fw.blockSize)
				}
			case <-time.After(fw.blockTick):
				if len(items) > 0 {
					toEncode <- &writerBlock{items: items}
					items = make([]interface{}, 0, fw.blockSize)
				}
			}
		}
	} else {
		for item := range toBlock {
			items = append(items, item)
			if int64(len(items)) >= fw.blockSize {
				toEncode <- &writerBlock{items: items}
				items = make([]interface{}, 0, fw.blockSize)
			}
		}
	}
	if len(items) > 0 {
		toEncode <- &writerBlock{items: items}
	}
	close(toEncode)
}

func encoder(fw *Writer, toEncode <-chan *writerBlock, toCompress chan<- *writerBlock) {
	for block := range toEncode {
		if block.err == nil {
			block.encoded = new(bytes.Buffer)
			for _, item := range block.items {
				block.err = fw.dataCodec.Encode(block.encoded, item)
				if block.err != nil {
					break // ??? drops remainder of items on the floor
				}
			}
		}
		toCompress <- block
	}
	close(toCompress)
}

func compressor(fw *Writer, toCompress <-chan *writerBlock, toWrite chan<- *writerBlock) {
	switch fw.CompressionCodec {
	case CompressionNull:
		for block := range toCompress {
			block.compressed = block.encoded.Bytes()
			toWrite <- block
		}

	case CompressionDeflate:
		bb := new(bytes.Buffer)
		cw, _ := flate.NewWriter(bb, flate.DefaultCompression)

		for block := range toCompress {
			bb = new(bytes.Buffer)
			cw.Reset(bb)

			if _, block.err = cw.Write(block.encoded.Bytes()); block.err != nil {
				continue
			}

			if block.err = cw.Close(); block.err != nil {
				continue
			}

			block.compressed = bb.Bytes()
			toWrite <- block
		}

	case CompressionSnappy:
		var bb *bytes.Buffer

		var dst []byte

		for block := range toCompress {
			checksum := crc32.ChecksumIEEE(block.encoded.Bytes())

			dst = snappy.Encode(nil, block.encoded.Bytes())
			bb = bytes.NewBuffer(dst)
			if block.err = binary.Write(bb, binary.BigEndian, checksum); block.err != nil {
				continue
			}

			block.compressed = bb.Bytes()
			toWrite <- block
		}

	}
	close(toWrite)
}

func writer(fw *Writer, toWrite <-chan *writerBlock) {
	for block := range toWrite {
		if block.err == nil {
			block.err = longCodec.Encode(fw.w, int64(len(block.items)))
		}
		if block.err == nil {
			block.err = longCodec.Encode(fw.w, int64(len(block.compressed)))
		}
		if block.err == nil {
			_, block.err = fw.w.Write(block.compressed)
		}
		if block.err == nil {
			_, block.err = fw.w.Write(fw.Sync)
		}
		if block.err != nil {
			log.Printf("[WARNING] cannot write block: %v", block.err)
			fw.err = block.err // ???
			break
			// } else {
			// 	log.Printf("[DEBUG] block written: %d, %d, %v", len(block.items), len(block.compressed), block.compressed)
		}
	}
	fw.writerDone <- struct{}{}
}