I’m wondering if I’ve wandered into trouble with my Ecto schemas. I have one that describes files in S3. For convenience (e.g. for REST endpoints), it’s nice to have a singular primary key. But it’s also sane to have a unique constraint on the combination of bucket + object_key. There are a couple ways of describing this for PostGres in the migration. My question is how to describe this in the Ecto schema?

One way is this:

  @primary_key {:id, :binary_id, autogenerate: true}
  schema "files" do
    field(:bucket, :string, primary_key: true)
    field(:object_key, :string, primary_key: true)
    # etc.
    timestamps()
  end

Is that correct? Or is it incorrect because it infers that the composite key is actually id + bucket + object_key?

or would this be more correct:

  @primary_key {:id, :binary_id, autogenerate: true}
  schema "files" do
    field(:bucket, :string)
    field(:object_key, :string)
    # etc.
    timestamps()
  end

How to describe the critical importance of the bucket + object_key constraint?

Thanks in advance for clarity on this!