feat(test): add --local to run tests without loading state (#6060)

Signed-off-by: Adegbite Ayoade <tripleaceme@gmail.com>
Co-authored-by: Cortland Goffena <30168413+cmgoffena13@users.noreply.github.com>

Author: tripleaceme

docs/concepts/tests.md

@@ -463,6 +463,23 @@ You can also run tests that match a pattern or substring using a glob pathname e
 $ sqlmesh test tests/test_*
 ```
 
+You can pass `--local` to run tests without loading state from the configured state connection:
+
+``` bash
+$ sqlmesh test --local
+```
+
+This keeps offline runs and commit hooks from opening a connection to the state backend.
+
+In multi-repository setups, or when running tests for only a subset of projects, models that exist only in remote state are not loaded under `--local`. Unlike [`sqlmesh lint --local`](../guides/linter.md), which reports additional errors in that situation, a test whose model is missing is **skipped with a warning and the run still succeeds**:
+
+```
+[WARNING] Model '"memory"."bronze"."a"' was not found at tests/test_a.yaml
+.**Successfully Ran `1` Tests Against `duckdb`**
+```
+
+So a passing exit code alone does not mean every test you expected actually ran. Watch the output for these warnings, and keep in mind that a hook using `--local` will not fail on them.
+
 ### Testing using notebooks
 
 You can execute tests on demand using the `%run_test` notebook magic as follows:

docs/reference/cli.md

@@ -630,6 +630,9 @@ Options:
                        useful for debugging.
   --select-model TEXT  Select specific models to run unit tests for. Can be
                        specified multiple times.
+  --local              Run tests using only locally loaded project files
+                       without loading state. Tests whose model is not loaded
+                       are skipped with a warning rather than failing.
   --help               Show this message and exit.
 ```
 

sqlmesh/cli/main.py

@@ -42,6 +42,8 @@
 )
 SKIP_CONTEXT_COMMANDS = ("init", "ui")
 LOCAL_ONLY_COMMANDS = ("format",)
+# Commands that are local-only when they're passed --local.
+OPTIONAL_LOCAL_COMMANDS = ("lint", "test")
 
 
 class _SQLMeshGroup(click.Group):
@@ -129,8 +131,12 @@ def cli(
     load = True
     # Local-only gating must hold for any number of --paths, so it stays outside the block below.
     load_state = ctx.invoked_subcommand not in LOCAL_ONLY_COMMANDS
-    # The parent callback constructs Context before Click invokes `lint`, so inspect its parsed args here.
-    if ctx.invoked_subcommand == "lint" and "--local" in ctx.meta["subcommand_args"]:
+    # The parent callback constructs Context before Click invokes the subcommand, so inspect its
+    # parsed args here.
+    if (
+        ctx.invoked_subcommand in OPTIONAL_LOCAL_COMMANDS
+        and "--local" in ctx.meta["subcommand_args"]
+    ):
         load_state = False
 
     if len(paths) == 1:
@@ -811,6 +817,12 @@ def create_test(
     multiple=True,
     help="Select specific models to run unit tests for.",
 )
+@click.option(
+    "--local",
+    is_flag=True,
+    expose_value=False,
+    help="Run tests using only locally loaded project files without loading state. Tests whose model is not loaded are skipped with a warning rather than failing.",
+)
 @click.argument("tests", nargs=-1)
 @click.pass_obj
 @error_handler

tests/cli/test_cli.py

@@ -2530,6 +2530,59 @@ def test_lint_local_runs_without_state(runner: CliRunner, tmp_path: Path, mocker
     mock.assert_not_called()
 
 
+def test_test_still_loads_state(runner: CliRunner, tmp_path: Path, mocker):
+    """Guard that `test` explicitly passes `load_state=True` and still reaches state sync."""
+    mock = _setup_local_only_project(tmp_path, mocker)
+    init_spy = mocker.spy(Context, "__init__")
+
+    runner.invoke(cli, ["--paths", str(tmp_path), "test"])
+
+    assert init_spy.called, "Context was never constructed"
+    for call in init_spy.call_args_list:
+        assert "load_state" in call.kwargs, (
+            "CLI didn't pass load_state= explicitly; missing kwarg defaults to True silently"
+        )
+        assert call.kwargs["load_state"] is True, (
+            f"Context was constructed with load_state={call.kwargs['load_state']} for `test`"
+        )
+    assert mock.called, "state-sync was never accessed during `test`"
+
+
+def test_test_local_runs_without_state(runner: CliRunner, tmp_path: Path, mocker):
+    mock = _setup_local_only_project(tmp_path, mocker)
+    init_spy = mocker.spy(Context, "__init__")
+
+    result = runner.invoke(cli, ["--paths", str(tmp_path), "test", "--local"])
+
+    assert result.exit_code == 0, f"Test failed: {result.output}\nException: {result.exception}"
+    assert init_spy.called, "Context was never constructed"
+    for call in init_spy.call_args_list:
+        assert "load_state" in call.kwargs, (
+            "CLI didn't pass load_state= explicitly; missing kwarg defaults to True silently"
+        )
+        assert call.kwargs["load_state"] is False, (
+            f"Context was constructed with load_state={call.kwargs['load_state']} for `test --local`"
+        )
+    mock.assert_not_called()
+
+
+def test_test_local_runs_without_state_multiple_paths(
+    runner: CliRunner, tmp_path: Path, mocker
+) -> None:
+    """`--local` gating must hold for any number of --paths, matching `lint --local`."""
+    project_a = tmp_path / "a"
+    project_b = tmp_path / "b"
+    _create_local_only_project(project_a, "proj_a")
+    _create_local_only_project(project_b, "proj_b")
+    mock = _patch_state_access(mocker)
+
+    result = runner.invoke(
+        cli, ["--paths", str(project_a), "--paths", str(project_b), "test", "--local"]
+    )
+    assert result.exit_code == 0, f"Test failed: {result.output}\nException: {result.exception}"
+    mock.assert_not_called()
+
+
 @pytest.mark.parametrize("command", ["format"])
 def test_local_only_commands_skip_state_multiple_paths(
     runner: CliRunner, tmp_path: Path, mocker, command: str
@@ -2611,3 +2664,115 @@ def test_format_does_not_open_state_connection(
     result = runner.invoke(cli, ["--paths", str(tmp_path), "format"])
     assert result.exit_code == 0, f"Format failed: {result.output}\nException: {result.exception}"
     mock.assert_not_called()
+
+
+def test_test_local_runs_project_unit_tests(runner: CliRunner, tmp_path: Path, mocker) -> None:
+    """A real unit test from the project's YAML runs under `--local` without touching state."""
+    create_example_project(tmp_path)
+    mock = _patch_state_access(mocker)
+
+    result = runner.invoke(cli, ["--paths", str(tmp_path), "test", "--local"])
+
+    assert result.exit_code == 0, f"Test failed: {result.output}\nException: {result.exception}"
+    assert "Successfully Ran 1 tests" in " ".join(result.output.split())
+    mock.assert_not_called()
+
+
+def test_test_local_does_not_open_state_connection(
+    runner: CliRunner, tmp_path: Path, mocker, monkeypatch
+) -> None:
+    """`test --local` must not open a configured remote Postgres state connection."""
+    pytest.importorskip("psycopg2")
+
+    for var in ("PG_HOST", "PG_USER", "PG_PASSWORD", "PG_DATABASE"):
+        monkeypatch.delenv(var, raising=False)
+
+    create_example_project(tmp_path)
+    (tmp_path / "config.yaml").write_text(
+        """project: cli_test
+
+gateways:
+  prod:
+    state_connection:
+      type: postgres
+      host: "{{ env_var('PG_HOST', 'postgres.internal.example.com') }}"
+      port: 5432
+      user: "{{ env_var('PG_USER') }}"
+      password: "{{ env_var('PG_PASSWORD') }}"
+      database: "{{ env_var('PG_DATABASE', 'sqlmesh_state') }}"
+    connection:
+      type: duckdb
+      database: "warehouse.db"
+
+default_gateway: prod
+
+model_defaults:
+  dialect: duckdb
+""",
+        encoding="utf-8",
+    )
+
+    mock = _patch_state_access(mocker)
+
+    result = runner.invoke(cli, ["--paths", str(tmp_path), "test", "--local"])
+    assert result.exit_code == 0, f"Test failed: {result.output}\nException: {result.exception}"
+    mock.assert_not_called()
+
+
+def test_test_local_multi_repo_partial(runner: CliRunner, copy_to_temp_path, mocker) -> None:
+    """Run tests for one repo of a multi-repo project whose upstream models live only in state.
+
+    Pins the behavioral difference against `lint --local`: a model that isn't loaded produces a
+    warning and its test is skipped, rather than turning into an error.
+    """
+    repo_2 = copy_to_temp_path("examples/multi")[0] / "repo_2"
+
+    # silver.c lives in repo_2 and its upstream bronze.a is supplied as a test input.
+    (repo_2 / "tests" / "test_c.yaml").write_text(
+        """test_silver_c:
+  model: silver.c
+  inputs:
+    bronze.a:
+      rows:
+      - col_a: 1
+      - col_a: 1
+      - col_a: 2
+  outputs:
+    query:
+      rows:
+      - col_a: 1
+      - col_a: 2
+""",
+        encoding="utf-8",
+    )
+    # bronze.a itself is defined in repo_1, so it is not loaded when only repo_2 is given.
+    (repo_2 / "tests" / "test_a.yaml").write_text(
+        """test_bronze_a:
+  model: bronze.a
+  outputs:
+    query:
+      rows:
+      - col_a: 1
+""",
+        encoding="utf-8",
+    )
+
+    mock = _patch_state_access(mocker)
+    args = ["--gateway", "memory", "--paths", str(repo_2), "test"]
+
+    # Without --local the same run reaches the state backend.
+    runner.invoke(cli, args)
+    assert mock.called, "state-sync was never accessed during `test`"
+
+    mock.reset_mock()
+
+    result = runner.invoke(cli, [*args, "--local"])
+
+    assert result.exit_code == 0, f"Test failed: {result.output}\nException: {result.exception}"
+    # Console output wraps, so compare against whitespace-normalized text.
+    output = " ".join(result.output.split())
+    assert 'Model \'"memory"."bronze"."a"\' was not found' in output, (
+        "the unloaded model should warn rather than fail"
+    )
+    assert "Successfully Ran 1 tests" in output, "the repo_2 test should still run"
+    mock.assert_not_called()