After reading this:

I tried to use term_to_binary together with postgresql:

Migration:

[...]
      add :owner, :text
      add :version, :text
      add :state, :bytea
[...]

Schema:

[...]
     field :state, :binary
[...]

Code:

update_or_insert(%State{owner: owner, state: :erlang.term_to_binary(state)})

  def update_or_insert(state) do
    state
    |> Repo.insert(
      on_conflict: {:replace_all_except, [:id, :owner, :created_at]},
      conflict_target: [:owner],
      set: [updated_at: DateTime.utc_now()]
    )
  end

Error:

07:11:53.186 [error] GenServer #PID<0.15770.0> terminating
** (Postgrex.Error) ERROR 22021 (character_not_in_repertoire) invalid byte sequence for encoding "UTF8": 0x83

How to solve this?
Change encoding for the resulting binary?
Leave term_to_binary alone and use JSON encoding/decoding?

I was also looking at the docs for erlang term_to_binary but there are only compression levels mentioned