refactor: idiomatic test sandbox setup (manual mode, single reset hook)

Replace the non-idiomatic :auto sandbox mode with :manual, eliminating
the root cause of stale data across test runs. The :auto mode let any
process auto-checkout and commit outside test transactions — Oban in
testing:inline already disables plugins/queues, so :auto was never
needed.

Changes:
- test_helper: Sandbox.mode(Repo, :manual); drop manual app boot (mix
  now starts the app idiomatically), drop --no-start alias, drop
  table-cleanup loop
- DataCase: single reset_test_state/0 consolidates GridCache.clear,
  ScoreCache.clear, score-file wipe, and default HTTP stubs; drop
  the :auto-restore on_exit hack; setup_sandbox now just start_owner
- ConnCase: delegates shared reset to DataCase, removes duplicate
  cache clears and stub installs
- config: pool Ecto.Adapters.SQL.Sandbox (idiomatic canonical form)

Note: GridCache/ScoreCache are globally-registered GenServers, so
the architecturally clean per-test start_supervised! approach
conflicts with async:true (parallel tests can't register the same
name). The centralized clear in reset_test_state is the practical
compromise — one place, documented, single call.

config/test.exs

@@ -34,7 +34,7 @@ config :microwaveprop, Microwaveprop.Repo,
   password: "postgres",
   hostname: "localhost",
   database: "microwaveprop_test#{System.get_env("MIX_TEST_PARTITION")}",
-  pool: DBConnection.Ownership,
+  pool: Sandbox,
   pool_size: System.schedulers_online() * 2,
   queue_target: 500,
   queue_interval: 5000,

mix.exs

@@ -167,7 +167,7 @@ defmodule Microwaveprop.MixProject do
       "ecto.setup": ["ecto.create", "ecto.migrate", "run priv/repo/seeds.exs"],
       "ecto.reset": ["ecto.drop", "ecto.setup"],
       "test.setup": ["ecto.create --quiet", "ecto.migrate --quiet"],
-      test: ["test --no-start"],
+      test: ["test"],
       "assets.setup": ["tailwind.install --if-missing", "esbuild.install --if-missing"],
       "assets.build": [
         "compile",

test/microwaveprop/workers/backfill_enqueue_worker_test.exs

@@ -93,7 +93,7 @@ defmodule Microwaveprop.Workers.BackfillEnqueueWorkerTest do
     test "enqueues all pending contacts when limit is omitted" do
       Phoenix.PubSub.subscribe(Microwaveprop.PubSub, "backfill:enqueue_complete")
 
-      our_ids =
+      _our_ids =
         for i <- 1..7 do
           ts = DateTime.shift(~U[2026-01-01 00:00:00Z], hour: i)
           create_contact(%{qso_timestamp: ts}).id
@@ -111,7 +111,7 @@ defmodule Microwaveprop.Workers.BackfillEnqueueWorkerTest do
     test "broadcasts enqueue_complete with count" do
       Phoenix.PubSub.subscribe(Microwaveprop.PubSub, "backfill:enqueue_complete")
 
-      contact = create_contact()
+      _contact = create_contact()
 
       assert :ok =
                BackfillEnqueueWorker.perform(%Oban.Job{

test/support/conn_case.ex

@@ -1,17 +1,6 @@
 defmodule MicrowavepropWeb.ConnCase do
   @moduledoc """
   Test case template for HTTP connection tests.
-
-  Such tests rely on `Phoenix.ConnTest` and also
-  import other functionality to make it easier
-  to build common data structures and query the data layer.
-
-  Finally, if the test case interacts with the database,
-  we enable the SQL sandbox, so changes done to the database
-  are reverted at the end of every test. If you are using
-  PostgreSQL, you can even run database tests asynchronously
-  by setting `use MicrowavepropWeb.ConnCase, async: true`, although
-  this option is not recommended for other databases.
   """
 
   use ExUnit.CaseTemplate
@@ -21,8 +10,6 @@ defmodule MicrowavepropWeb.ConnCase do
   alias Microwaveprop.Accounts.User
   alias Microwaveprop.AccountsFixtures
   alias Microwaveprop.DataCase
-  alias Microwaveprop.Propagation.ScoreCache
-  alias Microwaveprop.Weather.GridCache
   alias Phoenix.Ecto.SQL.Sandbox
 
   using do
@@ -32,60 +19,33 @@ defmodule MicrowavepropWeb.ConnCase do
       import MicrowavepropWeb.ConnCase
       import Phoenix.ConnTest
       import Plug.Conn
-      # The default endpoint for testing
-      @endpoint MicrowavepropWeb.Endpoint
 
-      # Import conveniences for testing with connections
+      @endpoint MicrowavepropWeb.Endpoint
     end
   end
 
   setup tags do
     owner = DataCase.setup_sandbox(tags)
-    DataCase.reset_score_files()
-    DataCase.stub_nexrad_default()
-    GridCache.clear()
-    ScoreCache.clear()
+    DataCase.reset_test_state()
 
-    # Stamp the sandbox metadata onto the "user-agent" header so a
-    # LiveView's connected mount (a separate process from this one, per
-    # `MicrowavepropWeb.SandboxHook`) can see this test's transaction.
     metadata = Sandbox.metadata_for(Microwaveprop.Repo, owner)
-
     conn = Plug.Conn.put_req_header(Phoenix.ConnTest.build_conn(), "user-agent", Sandbox.encode_metadata(metadata))
 
     {:ok, conn: conn}
   end
 
-  @doc """
-  Setup helper that registers and logs in users.
-
-      setup :register_and_log_in_user
-
-  It stores an updated connection and a registered user in the
-  test context.
-  """
   @spec register_and_log_in_user(map()) :: %{conn: Plug.Conn.t(), user: User.t(), scope: Scope.t()}
   def register_and_log_in_user(%{conn: conn} = context) do
     user = AccountsFixtures.user_fixture()
     scope = Scope.for_user(user)
 
-    opts =
-      context
-      |> Map.take([:token_authenticated_at])
-      |> Enum.to_list()
-
+    opts = context |> Map.take([:token_authenticated_at]) |> Enum.to_list()
     %{conn: log_in_user(conn, user, opts), user: user, scope: scope}
   end
 
-  @doc """
-  Logs the given `user` into the `conn`.
-
-  It returns an updated `conn`.
-  """
   @spec log_in_user(Plug.Conn.t(), User.t(), Keyword.t()) :: Plug.Conn.t()
   def log_in_user(conn, user, opts \\ []) do
     token = Accounts.generate_user_session_token(user)
-
     maybe_set_token_authenticated_at(token, opts[:token_authenticated_at])
 
     conn

test/support/data_case.ex

@@ -1,21 +1,13 @@
 defmodule Microwaveprop.DataCase do
   @moduledoc """
   Test case template for tests that need database access.
-
-  You may define functions here to be used as helpers in
-  your tests.
-
-  Finally, if the test case interacts with the database,
-  we enable the SQL sandbox, so changes done to the database
-  are reverted at the end of every test. If you are using
-  PostgreSQL, you can even run database tests asynchronously
-  by setting `use Microwaveprop.DataCase, async: true`, although
-  this option is not recommended for other databases.
   """
 
   use ExUnit.CaseTemplate
 
   alias Ecto.Adapters.SQL.Sandbox
+  alias Microwaveprop.Propagation.ScoreCache
+  alias Microwaveprop.Weather.GridCache
 
   using do
     quote do
@@ -29,79 +21,50 @@ defmodule Microwaveprop.DataCase do
   end
 
   setup tags do
-    Microwaveprop.DataCase.setup_sandbox(tags)
-    Microwaveprop.DataCase.reset_score_files()
-    Microwaveprop.DataCase.stub_nexrad_default()
+    pid = Microwaveprop.DataCase.setup_sandbox(tags)
+    Microwaveprop.DataCase.reset_test_state()
 
-    # sync tests (async: false) need shared Req.Test mode so stubs
-    # are visible to Tasks spawned by async_stream_nolink.
     Req.Test.set_req_test_from_context(tags)
 
     :ok
   end
 
   @doc """
-  Installs a default 404 stub for the IEM NEXRAD client so tests that
-  trigger `CommonVolumeRadarWorker` via `enqueue_for_contact/1` don't
-  crash for lack of a Req.Test plug. Individual tests can override with
-  their own `Req.Test.stub/2`.
+  Reset state that persists across tests within a single ExUnit run.
+
+  The sandbox isolates the database, but ETS-backed caches
+  (GridCache, ScoreCache) and on-disk score files survive across
+  test boundaries because the cache GenServers and filesystem live
+  in the app's BEAM process tree, not per-test. A single reset
+  here avoids per-file setup duplication.
   """
-  @spec stub_nexrad_default :: :ok
-  def stub_nexrad_default do
+  @spec reset_test_state :: :ok
+  def reset_test_state do
+    dir = Application.get_env(:microwaveprop, :propagation_scores_dir)
+    if is_binary(dir), do: File.rm_rf(dir)
+
+    GridCache.clear()
+    ScoreCache.clear()
+
     Req.Test.stub(Microwaveprop.Weather.NexradClient, fn conn ->
       Plug.Conn.send_resp(conn, 404, "not found")
     end)
+
+    Req.Test.stub(Microwaveprop.Weather.HrrrClient, fn conn ->
+      Plug.Conn.send_resp(conn, 404, "not found")
+    end)
+
+    :ok
   end
 
-  @doc """
-  Sets up the sandbox based on the test tags. Returns the owner pid so
-  callers (e.g. `MicrowavepropWeb.ConnCase`) can build Ecto Sandbox
-  metadata for out-of-process access (LiveView's connected mount).
-  """
   @spec setup_sandbox(map()) :: pid()
   def setup_sandbox(tags) do
     shared? = not tags[:async]
     pid = Sandbox.start_owner!(Microwaveprop.Repo, shared: shared?)
-
-    on_exit(fn ->
-      Sandbox.stop_owner(pid)
-
-      # Ecto reverts the pool to :manual mode when a `{:shared, pid}`
-      # owner terminates (see Ecto.Adapters.SQL.Sandbox docs). Since
-      # test_helper.exs relies on global :auto mode so Oban's
-      # background plugins (Stager, Peers, Met.Reporter) can
-      # auto-checkout, every `async: false` test permanently flips the
-      # pool to :manual once it exits — starving those processes of
-      # connections and crash-looping Oban until it exceeds its
-      # restart intensity and takes the whole app (and Repo) down with
-      # it. Restore :auto after any shared owner exits.
-      if shared?, do: Sandbox.mode(Microwaveprop.Repo, :auto)
-    end)
-
+    on_exit(fn -> Sandbox.stop_owner(pid) end)
     pid
   end
 
-  @doc """
-  Wipe the propagation ScoresFile tree between tests so files don't
-  leak between cases sharing the same tmp dir.
-  """
-  @spec reset_score_files :: :ok
-  def reset_score_files do
-    dir = Application.get_env(:microwaveprop, :propagation_scores_dir)
-
-    if is_binary(dir) do
-      File.rm_rf(dir)
-    end
-  end
-
-  @doc """
-  A helper that transforms changeset errors into a map of messages.
-
-      assert {:error, changeset} = Accounts.create_user(%{password: "short"})
-      assert "password is too short" in errors_on(changeset).password
-      assert %{password: ["password is too short"]} = errors_on(changeset)
-
-  """
   @spec errors_on(Ecto.Changeset.t()) :: map()
   def errors_on(changeset) do
     Ecto.Changeset.traverse_errors(changeset, fn {message, opts} ->

test/test_helper.exs

@@ -17,35 +17,14 @@ Code.require_file("test/support/fixtures/accounts_fixtures.ex")
 Code.require_file("test/support/fixtures/beacons_fixtures.ex")
 Code.require_file("test/support/fixtures/contacts_fixtures.ex")
 
-# ExUnit.start/1 with capture_log: true starts :logger, which cascades
-# into the full OTP application. mix.exs now runs `mix test --no-start`
-# so *we* control when the app boots. Initialize ExUnit first (no logger
-# cascade), force the Sandbox pool into the application environment,
-# then start the application manually.
 ExUnit.start()
 
-# With ExUnit initialized but no app running, the Repo supervisor hasn't
-# started yet. Force the Sandbox ownership pool into the app env before
-# the Repo child reads it.
-for repo <- [Microwaveprop.Repo] do
-  existing = Application.get_env(:microwaveprop, repo, [])
-
-  if existing[:url] || existing[:database] do
-    Application.put_env(:microwaveprop, repo, Keyword.put(existing, :pool, DBConnection.Ownership))
-  end
-end
-
-# Now start the full OTP app. The Repo supervisor reads pool:
-# DBConnection.Ownership from the app env and initializes with it.
-{:ok, _} = Application.ensure_all_started(:microwaveprop)
-
-# Clean up any rows left behind by a crashed or killed prior run.
-# The sandbox rolls back per-test, but if the suite itself crashes
-# mid-run, committed data persists and causes count-mismatch failures
-# in the next run. Wipe known test-data tables before each suite.
-for table <- ~w(contacts hrrr_fetch_tasks iemre_observations oban_jobs) do
-  Microwaveprop.Repo.delete_all(table)
-end
+# :manual mode: only the test process (via start_owner! / checkout) and
+# processes explicitly allowed via Sandbox.allow/4 or Caller Tracking
+# ($callers) can access the database. Background processes (Oban plugins,
+# boot Tasks) that auto-checked out during app boot are checked in here,
+# preventing any stale data from leaking across tests or across runs.
+Sandbox.mode(Microwaveprop.Repo, :manual)
 
 # stream_data is a test-only dep — its ebin is on the code path from the
 # wildcard above, but the application isn't started. `check all` etc.
@@ -53,12 +32,6 @@ end
 # the app to be loaded.
 Application.ensure_all_started(:stream_data)
 
-# :auto mode lets every process (including background Tasks started by
-# Application.init and Oban workers) auto-checkout from the sandbox.
-# DataCase.setup_sandbox/1 still calls start_owner!/2 for per-test
-# isolation, which is compatible with :auto mode.
-Sandbox.mode(Microwaveprop.Repo, :auto)
-
 # lazy_html provides an Enumerable protocol implementation for the
 # LazyHTML struct. When protocol consolidation is enabled (the default),
 # the consolidated Elixir.Enumerable.beam file is generated during