Refactor Ecto transaction for efficiency using Ecto.Multi - ecto

Question

I have a function that assigns multiple skills to a user. The skills already exist and have a many_to_many relationship with the user.
# skills will look like [%{id: "1"}, %{id: "4"}, %{id: "5"}] etc.
def reset(user, skills) do
user
|> Repo.preload(:skills)
|> Ecto.Changeset.change()
|> Ecto.Changeset.put_assoc(:skills, [])
|> Repo.update()
ids = Enum.map(skills, &Map.get(&1, :id))
query = Repo.all(from(p in Skill, where: p.id in ^ids))
skills = Enum.each(query, &Skill.create(user, &1))
end
Currently this works but feels inefficient:
We may be removing skills that will just be added back again
Might not need to run Repo.all to fetch skills, can just join them by ID if they exist
Could be wrapped in Ecto.Multi for database efficiency
In addition to this, it would be nice to return the created skills rather than just the :ok atom that Enum.each returns.
What would be the best way to refactor the code?

Answer 1

One improvement would be not assigning each skill one by one
def reset(user, skills) do
ids = Enum.map(skills, &Map.get(&1, :id))
skills = Repo.all(from(p in Question.Skill, where: p.id in ^ids))
user
|> Repo.preload(:skills)
|> Ecto.Changeset.change()
|> Ecto.Changeset.put_assoc(:skills, skills)
|> Repo.update()
end
If no change is needed (that is any deletion or addition to the user's skills), this makes two query to the database (assuming :skills hasn't been loaded, otherwise there'll be just one query, which is fetching the skills).
Also because we're making a changeset, it doesn't reset the whole thing. It only deletes or adds what is necessary.