Amazon S3 Utilities

Transfer Managers

The Amazon Simple Storage Service upload and download managers can break up large objects, so they can be transferred in multiple parts, in parallel. This makes it easy to resume interrupted transfers.

Upload Manager

The Amazon Simple Storage Service upload manager determines if a file can be split into smaller parts and uploaded in parallel. You can customize the number of parallel uploads and the size of the uploaded parts.

The following example uses the Amazon S3 Uploader to upload a file. Using Uploader is similar to the s3.PutObject() operation.

import "context"
import "github.com/aws/aws-sdk-go-v2/config"
import "github.com/aws/aws-sdk-go-v2/service/s3"
import "github.com/aws/aws-sdk-go-v2/feature/s3/manager"

// ...

cfg, err := config.LoadDefaultConfig(context.TODO())
if err != nil {
	log.Printf("error: %v", err)
	return
}

client := s3.NewFromConfig(cfg)

uploader := manager.NewUploader(client)
result, err := uploader.Upload(context.TODO(), &s3.PutObjectInput{
	Bucket: aws.String("my-bucket"),
	Key:    aws.String("my-object-key"),
	Body:   uploadFile,
})

Configuration Options

When you instantiate an Uploader instance using NewUploader, you can specify several configuration options to customize how objects are uploaded. Options are overridden by providing one or more arguments to NewUploader. These options include:

  • PartSize – Specifies the buffer size, in bytes, of each part to upload. The minimum size per part is 5 MiB.
  • Concurrency – Specifies the number of parts to upload in parallel.
  • LeavePartsOnError – Indicates whether to leave successfully uploaded parts in Amazon S3.

The Concurrency value limits the concurrent number of part uploads that can occur for a given Upload call. This is not a global client concurrency limit. Tweak the PartSize and Concurrency configuration values to find the optimal configuration. For example, systems with high-bandwidth connections can send bigger parts and more uploads in parallel.

For example, your application configures Uploader with a Concurrency of setting of 5. If your application then calls Upload from two different goroutines, the result is 10 concurrent part uploads (2 goroutines * 5 Concurrency).

Your application is expected to limit the concurrent calls to Upload to prevent application resource exhaustion.

Below is an example to set the default part size during Uploader creation:

uploader := manager.NewUploader(client, func(u *Uploader) {
	u.PartSize = 10 * 1024 * 1024, // 10 MiB
})

For more information about Uploader and its configurations, see Uploader in the AWS SDK for Go V2 API Reference.

PutObjectInput Body Field (io.ReadSeeker vs. io.Reader)

The Body field of the s3.PutObjectInput struct is an io.Reader type. However, this field can be populated with a type that satisfies both the io.ReadSeeker and io.ReaderAt interface to improve application resource utilization of the host environment. The following example creates the type ReadSeekerAt that satisfies both interfaces:

type ReadSeekerAt interface {
	io.ReadSeeker
	io.ReaderAt
}

For io.Reader types, the bytes of the reader must be buffered in memory before the part can be uploaded. When you increase the PartSize or Concurrency value, the required memory (RAM) for the Uploader increases significantly. The required memory is approximately PartSize * Concurrency. For example, specifying 100 MB for PartSize and 10 for Concurrency, requires at least 1 GB.

Because an io.Reader type cannot determine its size before reading its bytes, Uploader cannot calculate how many parts will be uploaded. Consequently, Uploader can reach the Amazon S3 upload limit of 10,000 parts for large files if you set the PartSize too low. If you try to upload more than 10,000 parts, the upload stops and returns an error.

For body values that implement the ReadSeekerAt type, the Uploader doesn’t buffer the body contents in memory before sending it to Amazon S3. Uploader calculates the expected number of parts before uploading the file to Amazon S3. If the current value of PartSize requires more than 10,000 parts to upload the file, Uploader increases the part size value so that fewer parts are required.

Handling Failed Uploads

If an upload to Amazon S3 fails, by default, Uploader uses the Amazon S3 AbortMultipartUpload operation to remove the uploaded parts. This functionality ensures that failed uploads do not consume Amazon S3 storage.

You can set LeavePartsOnError to true so that the Uploader doesn’t delete successfully uploaded parts. This is useful for resuming partially completed uploads. To operate on uploaded parts, you must get the UploadID of the failed upload. The following example demonstrates how to use the manager.MultiUploadFailure error interface type to get the UploadID.

result, err := uploader.Upload(context.TODO(), &s3.PutObjectInput{
	Bucket: aws.String("my-bucket"),
	Key:    aws.String("my-object-key"),
	Body:   uploadFile,
})
output, err := u.upload(input)
if err != nil {
	var mu manager.MultiUploadFailure
	if errors.As(err, &mu) {
		// Process error and its associated uploadID
		fmt.Println("Error:", mu)
		_ = mu.UploadID() // retrieve the associated UploadID
	} else {
		// Process error generically
		fmt.Println("Error:", err.Error())
	}
	return
}

Overriding Uploader Options Per Upload

You can override the Uploader options when calling Upload by providing one or more arguments to the method. These overrides are concurrency-safe modifications and do not affect ongoing uploads, or subsequent Upload calls to the manager. For example, to override the PartSize configuration for a specific upload request:

params := &s3.PutObjectInput{
	Bucket: aws.String("my-bucket"),
	Key:    aws.String("my-key"),
	Body:   myBody,
}
resp, err := uploader.Upload(context.TODO(), params, func(u *manager.Uploader) {
	u.PartSize = 10 * 1024 * 1024, // 10 MiB
})

Examples

Upload a Folder to Amazon S3

The following example uses the path/filepath package to recursively gather a list of files and upload them to the specified Amazon S3 bucket. The keys of the Amazon S3 objects are prefixed with the file’s relative path.

package main

import (
	"context"
	"log"
	"os"
	"path/filepath"

	"github.com/aws/aws-sdk-go-v2/aws"
	"github.com/aws/aws-sdk-go-v2/config"
	"github.com/aws/aws-sdk-go-v2/feature/s3/manager"
	"github.com/aws/aws-sdk-go-v2/service/s3"
)

var (
	localPath string
	bucket    string
	prefix    string
)

func init() {
	if len(os.Args) != 4 {
		log.Fatalln("Usage:", os.Args[0], "<local path> <bucket> <prefix>")
	}
	localPath = os.Args[1]
	bucket = os.Args[2]
	prefix = os.Args[3]
}

func main() {
	walker := make(fileWalk)
	go func() {
		// Gather the files to upload by walking the path recursively 
		if err := filepath.Walk(localPath, walker.Walk); err != nil {
			log.Fatalln("Walk failed:", err)
		}
		close(walker)
	}()

	cfg, err := config.LoadDefaultConfig(context.TODO())
	if err != nil {
		log.Fatalln("error:", err)
	}
	
	// For each file found walking, upload it to Amazon S3
	uploader := manager.NewUploader(s3.NewFromConfig(cfg))
	for path := range walker {
		rel, err := filepath.Rel(localPath, path)
		if err != nil {
			log.Fatalln("Unable to get relative path:", path, err)
		}
		file, err := os.Open(path)
		if err != nil {
			log.Println("Failed opening file", path, err)
			continue
		}
		defer file.Close()
		result, err := uploader.Upload(context.TODO(), &s3.PutObjectInput{
			Bucket: &bucket,
			Key:    aws.String(filepath.Join(prefix, rel)),
			Body:   file,
		})
		if err != nil {
			log.Fatalln("Failed to upload", path, err)
		}
		log.Println("Uploaded", path, result.Location)
	}
}

type fileWalk chan string

func (f fileWalk) Walk(path string, info os.FileInfo, err error) error {
	if err != nil {
		return err
	}
	if !info.IsDir() {
		f <- path
	}
	return nil
}

Download Manager

The Amazon S3 Downloader manager determines if a file can be split into smaller parts and downloaded in parallel. You can customize the number of parallel downloads and the size of the downloaded parts.

Example: Download a File

The following example uses the Amazon S3 Downloader to download a file. Using Downloader is similar to the s3.GetObject operation.

import "context"
import "github.com/aws/aws-sdk-go-v2/aws"
import "github.com/aws/aws-sdk-go-v2/config"
import "github.com/aws/aws-sdk-go-v2/service/s3"
import "github.com/aws/aws-sdk-go-v2/feature/s3/manager"

// ...

cfg, err := config.LoadDefaultConfig(context.TODO())
if err != nil {
	log.Println("error:", err)
	return
}

client := s3.NewFromConfig(cfg)

downloader := manager.NewDownloader(client)
numBytes, err := downloader.Download(context.TODO(), downloadFile, &s3.GetObjectInput{
	Bucket: aws.String("my-bucket"), 
	Key:    aws.String("my-key"),
})

The downloadFile parameter is an io.WriterAt type. The WriterAt interface enables the Downloader to write multiple parts of the file in parallel.

Configuration Options

When you instantiate a Downloader instance, you can specify configuration options to customize how objects are downloaded:

  • PartSize – Specifies the buffer size, in bytes, of each part to download. The minimum size per part is 5 MB.
  • Concurrency – Specifies the number of parts to download in parallel.

The Concurrency value limits the concurrent number of part download that can occur for a given Download call. This is not a global client concurrency limit. Tweak the PartSize and Concurrency configuration values to find the optimal configuration. For example, systems with high-bandwidth connections can receive bigger parts and more downloads in parallel.

For example, your application configures Downloader with a Concurrency of 5. Your application then calls Download from two different goroutines, the result will be 10 concurrent part downloads (2 goroutines * 5 Concurrency).

Your application is expected to limit the concurrent calls to Download to prevent application resource exhaustion.

For more information about Downloader and its other configuration options, see manager.Downloader in the AWS SDK for Go V2 API Reference.

Overriding Downloader Options Per Download

You can override the Downloader options when calling Download by providing one or more functional arguments to the method. These overrides are concurrency safe modifications and do not affect ongoing uploads, or subsequent Download calls to the manager. For example, to override the PartSize configuration for a specific upload request:

params := &s3.GetObjectInput{
	Bucket: aws.String("my-bucket"),
	Key:    aws.String("my-key"),
}
resp, err := downloader.Download(context.TODO(), targetWriter, params, func(u *manager.Downloader) {
	u.PartSize = 10 * 1024 * 1024, // 10 MiB
})

Examples

Download All Objects in a Bucket

The following example uses pagination to gather a list of objects from an Amazon S3 bucket. Then it downloads each object to a local file.

package main

import (
	"context"
	"fmt"
	"log"
	"os"
	"path/filepath"

	"github.com/aws/aws-sdk-go-v2/aws"
	"github.com/aws/aws-sdk-go-v2/config"
	"github.com/aws/aws-sdk-go-v2/feature/s3/manager"
	"github.com/aws/aws-sdk-go-v2/service/s3"
)

var (
	Bucket         = "MyBucket" // Download from this bucket
	Prefix         = "logs/"    // Using this key prefix
	LocalDirectory = "s3logs"   // Into this directory
)

func main() {
	cfg, err := config.LoadDefaultConfig(context.TODO())
	if err != nil {
		log.Fatalln("error:", err)
	}

	client := s3.NewFromConfig(cfg)
	manager := manager.NewDownloader(client)

	paginator := s3.NewListObjectsV2Paginator(client, &s3.ListObjectsV2Input{
		Bucket: &Bucket,
		Prefix: &Prefix,
	})

	for paginator.HasMorePages() {
		page, err := paginator.NextPage(context.TODO())
		if err != nil {
			log.Fatalln("error:", err)
		}
		for _, obj := range page.Contents {
			if err := downloadToFile(manager, LocalDirectory, Bucket, aws.ToString(obj.Key)); err != nil {
				log.Fatalln("error:", err)
			}
		}
	}
}

func downloadToFile(downloader *manager.Downloader, targetDirectory, bucket, key string) error {
	// Create the directories in the path
	file := filepath.Join(targetDirectory, key)
	if err := os.MkdirAll(filepath.Dir(file), 0775); err != nil {
		return err
	}

	// Set up the local file
	fd, err := os.Create(file)
	if err != nil {
		return err
	}
	defer fd.Close()

	// Download the file using the AWS SDK for Go
	fmt.Printf("Downloading s3://%s/%s to %s...\n", bucket, key, file)
	_, err = downloader.Download(context.TODO(), fd, &s3.GetObjectInput{Bucket: &bucket, Key: &key})

	return err
}

GetBucketRegion

The GetBucketRegion is a utility function for determining the AWS Region location of an Amazon S3 Bucket. GetBucketRegion takes an Amazon S3 client and uses it to determine the location of the requested Bucket within the AWS Partition associated with the client’s configured Region.

For example to find the Region for the Bucket my-bucket:

cfg, err := config.LoadDefaultConfig(context.TODO())
if err != nil {
	log.Println("error:", err)
	return
}

bucket := "my-bucket"
region, err := manager.GetBucketRegion(ctx, s3.NewFromConfig(cfg), bucket)
if err != nil {
	var bnf manager.BucketNotFound
	if errors.As(err, &bnf) {
		log.Printf("unable to find bucket %s's Region\n", bucket)
	} else {
		log.Println("error:", err)
	}
	return
}
fmt.Printf("Bucket %s is in %s region\n", bucket, region)

If GetBucketRegion is not able to resolve the location of a Bucket, the function returns a BucketNotFound error type as shown in the example.

Unseekable Streaming Input

For API operations like PutObject and UploadPart, the Amazon S3 client expects the value of the Body input parameter to implement the io.Seeker interface by default. The io.Seeker interface is used by the client to determine the length of the value to upload, and to compute payload hash for the request signature. If the Body input parameter value does not implement io.Seeker, your application will receive an error.

operation error S3: PutObject, failed to compute payload hash: failed to seek
body to start, request stream is not seekable

You can change this behavior by modifying the operation method’s Middleware using functional options. The WithAPIOptions helper returns a functional option for zero or more middleware mutators. To disable the client computing the payload hash and use Unsigned Payload request signature add v4.SwapComputePayloadSHA256ForUnsignedPayloadMiddleware.

resp, err := client.PutObject(context.TODO(), &s3.PutObjectInput{
    Bucket: &bucketName,
    Key: &objectName,
    Body: bytes.NewBuffer([]byte(`example object!`)),
    ContentLength: 15, // length of body
}, s3.WithAPIOptions(
    v4.SwapComputePayloadSHA256ForUnsignedPayloadMiddleware,
))

Amazon S3 requires the content length to be provided for all object’s uploaded to a bucket. Since the Body input parameter does not implement io.Seeker interface the client will not be able to compute the ContentLength parameter for the request. The parameter must be provided by the application. The request will fail if the ContentLength parameter is not provided.

Use the SDK’s Amazon S3 Upload Manager for uploads that are not seekable, and do not have a known length.

Last modified February 16, 2023: doc: fix incorrectly link url (#2024) (01cae8c31e)