Implement stream-based input record decoder's instead of batch-based

CHANGELOG.md

@@ -0,0 +1,34 @@
+# v0.1.x
+
+## Release v0.1.0 - 2020/08/xx
+
+### Breaking changes
+
+- `Columnifier` interface stops supporting `io.WriterCloser`
+
+### Enhancement
+
+- #52 Implement stream-based input record decoder's to reduce memory consumption.
+
+
+# v0.0.x
+
+## Release v0.0.4 - 2020/07/29
+
+### Bug fix
+
+- Fix #49 Capture the first error caused by columnifier
+
+... And some small improvements.
+
+## Release v0.0.3 - 2020/06/02
+
+Ready to publish columnify as an OSS.
+
+## Release v0.0.2 - 2020/06/02
+
+Ready for production.
+
+## Release v0.0.1 - 2020/04/19
+
+Initial release.

columnifier/columnifier.go

@@ -4,9 +4,9 @@ import "io"
 
 // Columnifier is the interface that converts input file to columnar format file.
 type Columnifier interface {
-	io.WriteCloser
-
+	WriteFromReader(reader io.Reader) (int, error)
 	WriteFromFiles(paths []string) (int, error)
+	Close() error
 }
 
 // NewColumnifier creates a new Columnifier.

columnifier/parquet.go

@@ -1,13 +1,16 @@
 package columnifier
 
 import (
+	"io"
 	"io/ioutil"
+	"os"
 
 	"github.com/reproio/columnify/record"
 
 	"github.com/reproio/columnify/parquet"
 	"github.com/reproio/columnify/schema"
 	"github.com/xitongsys/parquet-go-source/local"
+	"github.com/xitongsys/parquet-go/marshal"
 	parquetSource "github.com/xitongsys/parquet-go/source"
 	"github.com/xitongsys/parquet-go/writer"
 )
@@ -65,36 +68,32 @@ func NewParquetColumnifier(st string, sf string, rt string, output string, confi
 }
 
 // Write reads, converts input binary data and write it to buffer.
-func (c *parquetColumnifier) Write(data []byte) (int, error) {
+func (c *parquetColumnifier) WriteFromReader(reader io.Reader) (int, error) {
 	// Intermediate record type is map[string]interface{}
-	c.w.MarshalFunc = parquet.MarshalMap
-	records, err := record.FormatToMap(data, c.schema, c.rt)
+	c.w.MarshalFunc = marshal.MarshalJSON
+	decoder, err := record.NewJsonDecoder(reader, c.schema, c.rt)
 	if err != nil {
 		return -1, err
 	}
 
 	beforeSize := c.w.Size
-	for _, r := range records {
-		if err := c.w.Write(r); err != nil {
+	for {
+		var v string
+		err = decoder.Decode(&v)
+		if err != nil {
+			if err == io.EOF {
+				break
+			} else {
+				return -1, err
+			}
+		}
+
+		if err := c.w.Write(v); err != nil {
 			return -1, err
 		}
 	}
 	afterSize := c.w.Size
 
-	// Intermediate record type is wrapped Apache Arrow record
-	// It requires Arrow Golang implementation more logical type supports
-	// ref. https://github.com/apache/arrow/blob/9c9dc2012266442d0848e4af0cf52874bc4db151/go/arrow/array/builder.go#L211
-	/*
-		c.w.MarshalFunc = parquet.MarshalArrow
-		records, err := record.FormatToArrow(data, c.schema, c.rt)
-		if err != nil {
-			return err
-		}
-		if err := c.w.Write(&records); err != nil {
-			return err
-		}
-	*/
-
 	return int(afterSize - beforeSize), nil
 }
 
@@ -103,11 +102,12 @@ func (c *parquetColumnifier) WriteFromFiles(paths []string) (int, error) {
 	var n int
 
 	for _, p := range paths {
-		data, err := ioutil.ReadFile(p)
+		f, err := os.Open(p)
 		if err != nil {
 			return -1, err
 		}
-		if n, err = c.Write(data); err != nil {
+
+		if n, err = c.WriteFromReader(f); err != nil {
 			return -1, err
 		}
 	}