Hierarchical storage management

S3 can be configured to use global or local buckets. Investigate your latency requirements before choosing the global option. You can use replication to synchronize your objects across regions if eventual consistency is acceptable—the following code sample shows how this might be achieved, but should not be used verbatim:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Action": [
        "s3:GetReplicationConfiguration",
        "s3:ListBucket"
      ],
      "Effect": "Allow",
      "Resource": [
        "${aws_s3_bucket.bucket.arn}"
      ]
    },
-----------------------------------------------------------------------

resource "aws_iam_policy_attachment" "replication" {
  name = "tf-iam-role-attachment-replication-12345"
  roles = ["${aws_iam_role.replication.name}"]
  policy_arn = "${aws_iam_policy.replication.arn}"
}

resource "aws_s3_bucket" "destination" {
  bucket = "tf-test-bucket-destination-12345"
  region = "us-west-1"

  versioning {
    enabled = true
  }
}
-----------------------------------------------------------------------

  replication_configuration {
    role = "${aws_iam_role.replication.arn}"

    rules {
      id = "foobar"
      prefix = "foo"
      status = "Enabled"

      destination {
        bucket = "${aws_s3_bucket.destination.arn}"
        storage_class = "STANDARD"
      }
    }
  }
}

Storage life cycle policies can also be attached to your buckets, which will move infrequently accessed object to less available and more inexpensive storage:

resource "aws_s3_bucket" "versioning_bucket" {
  bucket = "my-versioning-bucket"
  acl = "private"

  versioning {
    enabled = true
  }

  lifecycle_rule {
    prefix = "config/"
    enabled = true
    noncurrent_version_transition {
      days = 30
      storage_class = "STANDARD_IA"
    }

    noncurrent_version_transition {
      days = 60
      storage_class = "GLACIER"
    }

    noncurrent_version_expiration {
      days = 90
    }
  }
}