boto3 S3 Multipart Upload

s3_multipart_upload.py

import argparse
import os

import boto3


class S3MultipartUpload(object):
  # AWS throws EntityTooSmall error for parts smaller than 5 MB
  PART_MINIMUM = int(5e6)

  def __init__(self,
               bucket,
               key,
               local_path,
               part_size=int(15e6),
               profile_name=None,
               region_name="eu-west-1",
               verbose=False):
    self.bucket = bucket
    self.key = key
    self.path = local_path
    self.total_bytes = os.stat(local_path).st_size
    self.part_bytes = part_size
    assert part_size > self.PART_MINIMUM
    assert (self.total_bytes % part_size == 0
            or self.total_bytes % part_size > self.PART_MINIMUM)
    self.s3 = boto3.session.Session(
        profile_name=profile_name, region_name=region_name).client("s3")
    if verbose:
      boto3.set_stream_logger(name="botocore")

  def abort_all(self):
    mpus = self.s3.list_multipart_uploads(Bucket=self.bucket)
    aborted = []
    print("Aborting", len(mpus), "uploads")
    if "Uploads" in mpus:
      for u in mpus["Uploads"]:
        upload_id = u["UploadId"]
        aborted.append(
            self.s3.abort_multipart_upload(
                Bucket=self.bucket, Key=self.key, UploadId=upload_id))
    return aborted

  def create(self):
    mpu = self.s3.create_multipart_upload(Bucket=self.bucket, Key=self.key)
    mpu_id = mpu["UploadId"]
    return mpu_id

  def upload(self, mpu_id):
    parts = []
    uploaded_bytes = 0
    with open(self.path, "rb") as f:
      i = 1
      while True:
        data = f.read(self.part_bytes)
        if not len(data):
          break
        part = self.s3.upload_part(
            Body=data, Bucket=self.bucket, Key=self.key, UploadId=mpu_id, PartNumber=i)
        parts.append({"PartNumber": i, "ETag": part["ETag"]})
        uploaded_bytes += len(data)
        print("{0} of {1} uploaded ({2:.3f}%)".format(
            uploaded_bytes, self.total_bytes,
            as_percent(uploaded_bytes, self.total_bytes)))
        i += 1
    return parts

  def complete(self, mpu_id, parts):
    result = self.s3.complete_multipart_upload(
        Bucket=self.bucket,
        Key=self.key,
        UploadId=mpu_id,
        MultipartUpload={"Parts": parts})
    return result


# Helper
def as_percent(num, denom):
  return float(num) / float(denom) * 100.0


def parse_args():
  parser = argparse.ArgumentParser(description='Multipart upload')
  parser.add_argument('--bucket', required=True)
  parser.add_argument('--key', required=True)
  parser.add_argument('--path', required=True)
  parser.add_argument('--region', default="eu-west-1")
  parser.add_argument('--profile', default=None)
  return parser.parse_args()


def main():
  args = parse_args()
  mpu = S3MultipartUpload(
      args.bucket,
      args.key,
      args.path,
      profile_name=args.profile,
      region_name=args.region)
  # abort all multipart uploads for this bucket (optional, for starting over)
  mpu.abort_all()
  # create new multipart upload
  mpu_id = mpu.create()
  # upload parts
  parts = mpu.upload(mpu_id)
  # complete multipart upload
  print(mpu.complete(mpu_id, parts))


if __name__ == "__main__":
  main()

had to roll a custom multipart upload (awscli erroring out on long upload with faulty network), and found boto3 multipart upload poorly documented so storing example code here

We are working off your code for a Lambda function that pulls data from an FTP site, caches in memory and uploads chunks in a multipart, based on your code. We notice that for large files (1GB for eg) the upload process repeats. We are thinking maybe a part fails?

Thanks a lot for this gist, it has been a fantastic resource for me.

Great job! It will be very helpful to me.

super, A+++, 👍
used this script to upload 50GB files from a kube pod to S3.

Thanks a lot, this has been most useful!

I have created a modified version able to resume the upload after a failure, useful if the network fails or your session credentials expire.

This is super helpful and very clean, thanks.
However, while searching for this, I also found a dead simple way of doing it where you can force multipart by setting a size threshold, in the AWS docs, just 2 lines of code:
https://boto3.amazonaws.com/v1/documentation/api/latest/guide/s3.html

How would this be modified to generated a presigned URL? I'm able to generate one, but it has a signature verification error, so I'm thinking that I'm missing something that sets the algorithm / version. Here are details if anyone can help! I'm trying to use the s3 boto3 client for a minio server for multipart upload with a presigned url because the minio-py doesn't support that.

---

Update - I think I figured out how to add the key - the config parameter below is newly added

from botocore.client import Config
...
s3_external = session.client(
    "s3",
    use_ssl=MINIO_SSL,
    region_name=MINIO_REGION,
    endpoint_url=MINIO_HTTP_PREFIX + MINIO_EXTERNAL_SERVER,
    verify=False,
    config=Config(signature_version='s3v4'),
)

The signed url generated now has the (previously missing) algorithm, etc. headers, however the signature doesn't match, so I'm wondering if the key generated by the client (Singularity / Sylabs scs-library-client) is different than what I am specifying - that almost must be it...

---

Update: i think the issue is that the signature includes the host, which is different inside (minio:9000) as opposed to outside (127.0.0.1:9000) the container, reading this post. boto/boto3#1982 (comment)

Nice brother, great great job

@balukrishnans are you talking to me or @teasherm? To follow up with my question above for future lurkers, it was a non-trivial thing that wound up needing a PR to the Minio Python client. Details about my particular implementation are here. And if you are referencing @teasherm, I agree, great job and thank you for posting this!

How to get the content type of object after complete action?