Python¶

Use boto3. The compatibility suite pins boto3==1.43.77 and runs it against a real Record Store server.

pip install boto3

Client¶

import boto3

s3 = boto3.client(
    "s3",
    endpoint_url="https://storage.example.com",
    aws_access_key_id="<service account access key>",
    aws_secret_access_key="<service account secret key>",
    region_name="us-east-1",
    config=boto3.session.Config(s3={"addressing_style": "path"}),
)

addressing_style="path" is required. Region is arbitrary but must be consistent — SigV4 signs it.

Prefer environment variables

Read credentials from AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY or your own variables rather than committing them.

Buckets¶

s3.create_bucket(Bucket="uploads")

for bucket in s3.list_buckets()["Buckets"]:
    print(bucket["Name"])

Upload¶

s3.put_object(
    Bucket="uploads",
    Key="invoices/2026/03/inv-1.pdf",
    Body=open("inv-1.pdf", "rb"),
    ContentType="application/pdf",
)

upload_file handles large objects with multipart automatically:

s3.upload_file("big.bin", "uploads", "big.bin")

Download¶

s3.download_file("uploads", "invoices/2026/03/inv-1.pdf", "inv-1.pdf")

Or stream it:

response = s3.get_object(Bucket="uploads", Key="invoices/2026/03/inv-1.pdf")
for chunk in response["Body"].iter_chunks():
    ...

List¶

paginator = s3.get_paginator("list_objects_v2")
for page in paginator.paginate(Bucket="uploads", Prefix="invoices/2026/"):
    for obj in page.get("Contents", []):
        print(obj["Key"], obj["Size"])

Use the paginator. Listing is always bounded.

Delete¶

s3.delete_object(Bucket="uploads", Key="invoices/2026/03/inv-1.pdf")

Presigned URLs¶

url = s3.generate_presigned_url(
    "put_object",
    Params={"Bucket": "uploads", "Key": key, "ContentType": "application/pdf"},
    ExpiresIn=900,
)

Sign server-side only. See Presigned URLs.

Versioning¶

s3.put_bucket_versioning(
    Bucket="uploads",
    VersioningConfiguration={"Status": "Enabled"},
)

versions = s3.list_object_versions(Bucket="uploads")

Ranges¶

response = s3.get_object(Bucket="uploads", Key="big.bin", Range="bytes=0-1023")

Checksums¶

If uploads fail with NotImplemented, boto3 is using AWS's aws-chunked trailing checksums. Turn that off:

export AWS_REQUEST_CHECKSUM_CALCULATION=WHEN_REQUIRED
export AWS_RESPONSE_CHECKSUM_VALIDATION=WHEN_REQUIRED