S3 - Summation

The S3 connector exposes objects in an AWS S3 bucket as datasets. Summation reads Parquet, CSV, and JSON files directly, with no warehouse needed.

What you’ll need

An S3 bucket in a known AWS region. See AWS Regions and Zones for region codes.
An IAM user (or role with access keys) with read permission on the prefix you want exposed. See Managing access keys for IAM users. The minimum policy is in the IAM policy section below.

Form fields

Field	Required	Stored as	Notes
Region	Yes	Config	AWS region of the bucket, e.g. `us-west-2`. See Bucket regions.
Base Path	Yes	Config	An `s3://` prefix to scope the connection to, e.g. `s3://my-bucket/path`. Browsing starts from this prefix. The connection saves without it, but browsing and attaching datasets both fail until it is set, so treat it as required.
Access Key ID	Yes	Secret	AWS access key ID. See Managing access keys for IAM users.
Secret Access Key	Yes	Secret	AWS secret access key matching the key ID.

IAM policy

Minimum policy for read-only access to a single prefix. Replace my-bucket and the path/ prefix with your own. For more, see Bucket policy examples and Identity-based policies for Amazon S3.

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": ["s3:ListBucket"],
      "Resource": "arn:aws:s3:::my-bucket",
      "Condition": {
        "StringLike": { "s3:prefix": ["path/*"] }
      }
    },
    {
      "Effect": "Allow",
      "Action": ["s3:GetObject"],
      "Resource": "arn:aws:s3:::my-bucket/path/*"
    }
  ]
}

Adding datasets

Each dataset is a single file or a glob within the bucket. Summation infers the schema from the file format. Source references use the form:

s3://bucket/path/to/data/

Common problems

Error or symptom	Likely cause
`AccessDenied`	The IAM user/role lacks `s3:GetObject` on the key. Check both the IAM policy and the bucket policy.
`NoSuchBucket`	Wrong region. Region must match the bucket’s region exactly.
CSV columns inferred wrong	Re-check the file’s encoding and delimiter. Parquet is the most reliable format for repeatable analyses.