将对象上传到AWS S3，而无需使用aws-sdk-go创建文件

The Body field of the UploadInput struct is just an io.Reader. So pass any io.Reader you want--it doesn't need to be a file.

Here is what I ended up writing

func (s *S3Sink) upload() {
    now := time.Now()
    key := s.getNewKey(now)

    _, err := s.uploader.Upload(&s3manager.UploadInput{
        Bucket: aws.String(s.bucket),
        Key:    aws.String(key),
        Body:   s.bodyBuf,
    })

    if err != nil {
        glog.Errorf("Error uploading %s to s3, %v", key, err)
    }
    glog.Infof("Uploaded at %s", key)
    s.lastUploadTimestamp = now.UnixNano()

    s.bodyBuf.Truncate(0)
}

More details below: https://github.com/heptiolabs/eventrouter/blob/20edca33bc6e20465810d49bdb213119464eb440/sinks/s3sink.go#L185-L201

In this answer, I will post all the things that worked for me that related to this question. Many thanks to @ThunderCat and @Flimzy that alerted me that the body parameter of the upload request was already an io.Reader. I will post some sample codes commenting on what I've learned from this question and how it helped me solve this problem. Perhaps this will help others like me and @AlokKumarSingh.

Case 1: You already have the data in memory (e.g. receiving data from a streaming/messaging service like Kafka, Kinesis or SQS)

func main() {
    if len(os.Args) != 3 {
        fmt.Printf(
            "bucket and file name required
Usage: %s bucket_name filename",
            os.Args[0],
        )
    }

    bucket := os.Args[1]
    filename := os.Args[2]

    // this is your data that you have in memory
    // in this example it is hard coded but it may come from very distinct
    // sources, like streaming services for example.
    data := "Hello, world!"

    // create a reader from data data in memory
    reader := strings.NewReader(data)

    sess, err := session.NewSession(&aws.Config{
        Region: aws.String("us-east-1")},
    )
    uploader := s3manager.NewUploader(sess)

    _, err = uploader.Upload(&s3manager.UploadInput{
        Bucket: aws.String(bucket),
        Key: aws.String(filename),
        // here you pass your reader
        // the aws sdk will manage all the memory and file reading for you
        Body: reader,
    })
    if err != nil {.
        fmt.Printf("Unable to upload %q to %q, %v", filename, bucket, err)
    }

    fmt.Printf("Successfully uploaded %q to %q
", filename, bucket)
}

Case 2: You already have a persisted file and you want to upload it but you dont want to maintain the whole file in memory:

func main() {
    if len(os.Args) != 3 {
        fmt.Printf(
            "bucket and file name required
Usage: %s bucket_name filename",
            os.Args[0],
        )
    }

    bucket := os.Args[1]
    filename := os.Args[2]

    // open your file
    // the trick here is that the method os.Open just returns for you a reader
    // for the desired file, so you will not maintain the whole file in memory.
    // I know this might sound obvious, but for a starter (as I was at the time
    // of the question) it is not.
    fileReader, err := os.Open(filename)
    if err != nil {
        fmt.Printf("Unable to open file %q, %v", err)
    }
    defer fileReader.Close()

    sess, err := session.NewSession(&aws.Config{
        Region: aws.String("us-east-1")},
    )
    uploader := s3manager.NewUploader(sess)

    _, err = uploader.Upload(&s3manager.UploadInput{
        Bucket: aws.String(bucket),
        Key:    aws.String(filename),
        // here you pass your reader
        // the aws sdk will manage all the memory and file reading for you
        Body: fileReader,
    })
    if err != nil {
        fmt.Printf("Unable to upload %q to %q, %v", filename, bucket, err)
    }

    fmt.Printf("Successfully uploaded %q to %q
", filename, bucket)
}

Case 3: This is how I implemented it on the final version of my system, but to understand why I did it I must give you some background.

My use case evolved a bit. The upload code was going to be a function in Lambda and the files turned out to be huge. What do this changes mean: If I uploaded the file via an entry point in API Gateway attached to a Lambda function I would have to wait for the whole file to complete the upload in Lambda. Since lambda is priced by the duration and memory usage of the invocation, this could be a really big problem.

So, to solve this problem I used a pre-signed post URL for the upload. How does this affect the architecture/workflow?

Instead of upload to S3 from my backend code, I just create and authenticate a URL for posting the object to S3 in the backend and send this URL to the frontend. With that, I just implemented a multipart upload to that URL. I know that this is a lot more specific than the question, but it wasn't easy to discover this solution, so I think it would be a good idea to document it here for others.

Here is a sample of how to create that pre-signed URL in nodejs.

const AWS = require('aws-sdk');

module.exports.upload = async (event, context, callback) => {

  const s3 = new AWS.S3({ signatureVersion: 'v4' });
  const body = JSON.parse(event.body);

  const params = {
    Bucket: process.env.FILES_BUCKET_NAME,
    Fields: {
      key: body.filename,
    },
    Expires: 60 * 60
  }

  let promise = new Promise((resolve, reject) => {
    s3.createPresignedPost(params, (err, data) => {
      if (err) {
        reject(err);
      } else {
        resolve(data);
      }
    });
  })

  return await promise
    .then((data) => {
      return {
        statusCode: 200,
        body: JSON.stringify({
          message: 'Successfully created a pre-signed post url.',
          data: data,
        })
      }
    })
    .catch((err) => {
      return {
        statusCode: 400,
        body: JSON.stringify({
          message: 'An error occurred while trying to create a pre-signed post url',
          error: err,
        })
      }
    });
};

If you want to use go it's the same idea, you just have to change de sdk.