Broken test for multi-tenancy and Triplex

Implement complex logic for multi-tenancy solution: create team → create schema → run migration on new schema. Found a thread and a mention in Triplex documentation and moved the migrations from the transaction. Method look like

  def create_team(attrs \\ %{}) do
    changeset =
      %Team{}
      |> change(%{uid: generate_token(type: :schema)})
      |> Team.changeset(attrs)

    Multi.new()
    |> Multi.insert(:team, changeset)
    |> Multi.run(:schema, fn repo, %{team: team} ->
      Triplex.create_schema(team.uid, repo)
    end)
    |> Repo.transaction()
    |> case do
      {:ok, %{team: team}} ->
        {:ok, _} = Triplex.migrate(team.uid)

        {:ok, team}

      {:error, _, reason, _} ->
        {:error, reason}
    end
  end

It work flawless inside iex

iex(97)> Runa.Teams.create_team(%{title: “test”})
[debug] QUERY OK db=0.4ms idle=1327.3ms
begin
↳ Runa.Teams.create_team/1, at: lib/domains/teams/teams.ex:82
[debug] QUERY OK source=“teams” db=0.5ms
INSERT INTO “teams” (“title”,“uid”,“inserted_at”,“updated_at”) VALUES ($1,$2,$3,$4) RETURNING “id” [“test”, “nm8qnu26poykd2y02e0dup5wnf168qz2”, ~U[2024-05-20 17:55:36Z], ~U[2024-05-20 17:55:36Z]]
↳ Runa.Teams.create_team/1, at: lib/domains/teams/teams.ex:82
[debug] QUERY OK db=0.1ms
CREATE SCHEMA “nm8qnu26poykd2y02e0dup5wnf168qz2”
[debug] QUERY OK db=14.5ms
commit
↳ Runa.Teams.create_team/1, at: lib/domains/teams/teams.ex:82
[info] == Running 20240420161100 Runa.Repo.Migrations.CreateContributors.up/0 forward
[info] execute “CREATE TABLE nm8qnu26poykd2y02e0dup5wnf168qz2.contributors (\n id serial PRIMARY KEY,\n user_id integer REFERENCES public.users(id) ON DELETE CASCADE NOT NULL,\n team_id integer REFERENCES public.teams(id) ON DELETE CASCADE NOT NULL,\n role_id integer REFERENCES public.roles(id) ON DELETE CASCADE NOT NULL,\n inserted_at timestamp NOT NULL,\n updated_at timestamp NOT NULL\n);\n”
[info] execute “CREATE UNIQUE INDEX contributors_user_id_team_id_role_id_index\nON nm8qnu26poykd2y02e0dup5wnf168qz2.contributors (user_id, team_id, role_id);\n”
[info] == Migrated 20240420161100 in 0.0s
[info] == Running 20240520093137 Runa.Repo.Migrations.CreateProjectsTable.up/0 forward
[info] create table nm8qnu26poykd2y02e0dup5wnf168qz2.projects
[info] == Migrated 20240520093137 in 0.0s
{:ok,
%Runa.Teams.Team{
meta: ecto.Schema.Metadata<:loaded, “teams”>,
id: 53,
title: “test”,
uid: “nm8qnu26poykd2y02e0dup5wnf168qz2”,
team_roles: ecto.Association.NotLoaded,
users: ecto.Association.NotLoaded,
inserted_at: ~U[2024-05-20 17:55:36Z],
updated_at: ~U[2024-05-20 17:55:36Z]
}}

But inside test in returns error, I use create_project inside the fixture and create a entity, schema and migrations before each test

as I see it migrations cannot be performed in a test environment, but Triplex.exists?(team.uid) return true, Triplex.migrate("public") return {:ok, _}

  1. test projects returns all projects (Runa.ProjectsTest)
    test/domains/projects_test.exs:16
    ** (MatchError) no match of right hand side value: {:error, “could not checkout the connection owned by #PID<0.610.0>. When using the sandbox, connections are shared, so this may imply another process is using a connection. Reason: connection not available and request was dropped from queue after 822ms. You can configure how long requests wait in the queue using :queue_target and :queue_interval. See DBConnection.start_link/2 for more information”}
    stacktrace:
    (runa 0.1.0) lib/domains/teams/teams.ex:71: Runa.Teams.create_team/1
    (runa 0.1.0) test/support/fixtures/teams_fixtures.ex:25: Runa.TeamsFixtures.create_aux_team/1
    Runa.ProjectsTest.__ex_unit_describe_0/1

Is this a Triplex problem or a test environment problem?

Hi @ravecat I’ve stumbled upon this too but not the testing part. I didn’t think this would work in iex (or in a transaction) but then it seems it’s only Triplex.create that cannot run inside of a transaction since Ecto 3 :thinking:

However, running the Triplex.create_schema this way inside a multi transaction means it may leave unclean teams’ schemas should the Triplex.migrate fail later on.
If you use the Triplex.create_schema func argument to wrap the transaction instead, you might be able to avoid that side effect.

Regarding the failure in your test, I wonder whether your fixture is spawning a process that’s not dropping the connection before it’s checked out again by the test process? It would be useful to share your fixture and test code here. Also a total speculation is whether you have multiple of these tests in different modules with async turned on …

and apparently Sandboxed SQL in tests share the connection with the parent process while transaction spawned functions will not be part of the parent process and will hold their own connection (if they can?)

The transaction is per process. A separate process started inside a transaction won’t be part of the same transaction and will use a separate connection altogether.

… the parent process will typically hold the connection until the transaction completes. This may lead to a deadlock if the child process attempts to use the same connection.

So maybe your test is getting stuck in that deadlock?