I’m implementing an Ecto adapter for the Dgraph graph db. I’m able to successfully insert a new record from a Schema or a Changeset. I’ve implemented the callback Ecto.Adapter.Schema.insert/6 and this callback is expected to return {:ok, fields()}.

If I return {:ok, []} from the callback, Ecto loads the associated Schema struct with the changeset that was given by the user. Dgraph autogenerates UIDs as the primary key for each Dgraph Node created (equivalent to a record), but this doesn’t get included in the Schema struct returned by Ecto to the user.

I assumed incorrectly that if I return {:ok, [uid: autogenerated_id, ...additional_fields]} from my callback, it would use those values to load the Schema struct. (This is a minimal implementation for the example.)

  @impl Ecto.Adapter.Schema
  def insert(
        %{pid: conn} = _repo_meta,
        schema_meta,
        fields,
        _on_conflict,
        _returning,
        _opts
      ) do
    mutation = Query.new_mutation(:set, schema_meta, fields)

    case DBConnection.prepare_execute(conn, mutation, fields, []) do
      {:ok, _, %{uids: uids}} ->
        {:ok, Keyword.put(fields, :uid, find_uid(fields, uids))}

      {:error, reason} ->
        {:invalid, [{:unique, "primary_key"}]}
    end
  end

Instead, I’m getting the following error in Ecto after my callback was called and it is attempting to load the Schema:

     ** (FunctionClauseError) no function clause matching in Ecto.Repo.Schema.load_each/4

     The following arguments were given to Ecto.Repo.Schema.load_each/4:

         # 1
         %DgraphEx.Test.User{__meta__: #Ecto.Schema.Metadata<:loaded, "User">, uid: "_:m0vRzOFL9c", email: "kartch@dgrex.com", handle: "kartch"}

         # 2
         [uid: "0x33"]

         # 3
         []

         # 4
         DgraphEx.Ecto.Adapter

     Attempted function clauses (showing 2 out of 2):

         defp load_each(struct, [{_, value} | kv], [{key, type} | types], adapter)
         defp load_each(struct, [], _types, _adapter)

     code: TestRepo.insert(user_changeset)
     stacktrace:
       (ecto 3.12.5) lib/ecto/repo/schema.ex:1090: Ecto.Repo.Schema.load_each/4
       (ecto 3.12.5) lib/ecto/repo/schema.ex:1068: Ecto.Repo.Schema.load_changes/8
       (ecto 3.12.5) lib/ecto/repo/schema.ex:509: anonymous fn/15 in Ecto.Repo.Schema.do_insert/4
       test/insert_test.exs:18: (test)

The problem with the unmatched function clause is due to an empty 3rd argument being passed to Ecto.Repo.Schema.load_each/4. It appears to need a Keyword list of field keys to types as the 3rd argument but it’s empty. The return clause from my callback doesn’t allow for anything other than {:ok, fields()} so I’m unsure how to apply the new uid to the loaded struct.

Does the fields keyword list being returned by my callback need to be in a different format than the fields supplied in the form of [uid: uid_value, field_key1: field_value1, field_key2: field_value2, ...]?

Thank you for any help.

Edit for clarification:

I’m using a custom type for the UID field that autogenerates a temporary ID. I need to generate a temporary UID to associate the Node predicates on insert, and Dgraph assigns a new UID. In short, my custom type autogenerates a temp UID and I need the DB generated UID to be loaded into the struct.