fix: Improve test reliability

Makefile

@@ -20,7 +20,10 @@ build:  ## Build the python package
 
 tests: ${TESTS}  ## Run tests for each package
 ${TESTS}: %/tests:
-	uv run coverage run --parallel -m pytest -v  $*/tests
+	uv run coverage run --parallel -m pytest -v $*/tests
+
+quick-core-tests:  ## Run core tests excluding long_running
+	uv run coverage run --parallel -m pytest -v -m "not long_running" core/tests
 
 coverage:  ## Target to combine and report coverage.
 	uv run coverage combine
@@ -61,7 +64,7 @@ clean-all: clean ## Remove all generated files and reset the local virtual envir
 	rm -rf .venv
 
 # Targets that do not generate file-level artifacts.
-.PHONY: clean docs doctests image tests ${TESTS}
+.PHONY: clean docs doctests image tests quick-core-tests ${TESTS}
 
 
 # Implements this pattern for autodocumenting Makefiles:

core/testcontainers/core/config.py

@@ -88,6 +88,8 @@ def read_tc_properties() -> dict[str, str]:
 
 @dataclass
 class TestcontainersConfiguration:
+    __test__ = False
+
     def _render_bool(self, env_name: str, prop_name: str) -> bool:
         env_val = environ.get(env_name, None)
         if env_val is not None:

core/tests/test_compose.py

@@ -1,3 +1,4 @@
+import logging
 import subprocess
 from pathlib import Path
 from re import split
@@ -150,7 +151,7 @@ def test_compose_logs():
         assert not line or container.Service in next(iter(line.split("|")))
 
 
-def test_compose_volumes():
+def test_compose_volumes(caplog):
     _file_in_volume = "/var/lib/example/data/hello"
     volumes = DockerCompose(context=FIXTURES / "basic_volume", keep_volumes=True)
     with volumes:
@@ -167,8 +168,11 @@ def test_compose_volumes():
     assert "hello" in stdout
 
     # third time we expect the file to be missing
-    with volumes, pytest.raises(subprocess.CalledProcessError):
-        volumes.exec_in_container(["cat", _file_in_volume], "alpine")
+    with caplog.at_level(
+        logging.CRITICAL, logger="testcontainers.compose.compose"
+    ):  # suppress expected error logs about missing volume
+        with volumes, pytest.raises(subprocess.CalledProcessError):
+            volumes.exec_in_container(["cat", _file_in_volume], "alpine")
 
 
 # noinspection HttpUrlsUsage

core/tests/test_docker_in_docker.py

@@ -179,6 +179,7 @@ def test_find_host_network_in_dood() -> None:
     assert DockerClient().find_host_network() == os.environ[EXPECTED_NETWORK_VAR]
 
 
+@pytest.mark.long_running
 @pytest.mark.skipif(
     is_mac(),
     reason="Docker socket mounting and container networking do not work reliably on Docker Desktop for macOS",
@@ -219,6 +220,7 @@ def test_dood(python_testcontainer_image: str) -> None:
     assert status["StatusCode"] == 0
 
 
+@pytest.mark.long_running
 @pytest.mark.skipif(
     is_mac(),
     reason="Docker socket mounting and container networking do not work reliably on Docker Desktop for macOS",

core/tests/test_ryuk.py

@@ -1,4 +1,4 @@
-from time import sleep
+from time import sleep, perf_counter
 import pytest
 from pytest import MonkeyPatch
 
@@ -12,6 +12,27 @@
 from testcontainers.core.waiting_utils import wait_for_logs
 
 
+def _wait_for_container_removed(client: DockerClient, container_id: str, timeout: float = 30) -> None:
+    """Poll until a container is fully removed (raises NotFound)."""
+    start = perf_counter()
+    while perf_counter() - start < timeout:
+        try:
+            client.containers.get(container_id)
+        except NotFound:
+            return
+        sleep(0.5)
+
+    try:
+        c = client.containers.get(container_id)
+        name = c.name
+        status = c.status
+        started_at = c.attrs.get("State", {}).get("StartedAt", "unknown")
+        detail = f"name={name}, status={status}, started_at={started_at}"
+    except NotFound:
+        detail = "container disappeared just after timeout"
+    raise TimeoutError(f"Container {container_id} was not removed within {timeout}s ({detail})")
+
+
 @pytest.mark.skipif(
     is_mac(),
     reason="Ryuk container reaping is unreliable on Docker Desktop for macOS due to VM-based container lifecycle handling",
@@ -39,8 +60,11 @@ def test_wait_for_reaper(monkeypatch: MonkeyPatch):
     assert rs
     rs.close()
 
-    sleep(0.6)  # Sleep until Ryuk reaps all dangling containers. 0.5 extra seconds for good measure.
+    # Ryuk will reap containers then auto-remove itself.
+    # Wait for the reaper container to disappear and once it's gone, all labeled containers are guaranteed reaped.
+    _wait_for_container_removed(docker_client, reaper_id)
 
+    # Verify both containers were reaped
     with pytest.raises(NotFound):
         docker_client.containers.get(container_id)
     with pytest.raises(NotFound):

core/tests/test_wait_strategies.py

@@ -1,3 +1,4 @@
+import logging
 import re
 import time
 from datetime import timedelta
@@ -528,7 +529,7 @@ def test_file_exists_wait_strategy_initialization(self, file_path):
     @patch("pathlib.Path.is_file")
     @patch("time.time")
     @patch("time.sleep")
-    def test_wait_until_ready(self, mock_sleep, mock_time, mock_is_file, file_exists, expected_behavior):
+    def test_wait_until_ready(self, mock_sleep, mock_time, mock_is_file, file_exists, expected_behavior, caplog):
         strategy = FileExistsWaitStrategy("/tmp/test.txt").with_startup_timeout(1)
         mock_container = Mock()
 
@@ -547,7 +548,8 @@ def test_wait_until_ready(self, mock_sleep, mock_time, mock_is_file, file_exists
             mock_is_file.assert_called()
         else:
             with pytest.raises(TimeoutError, match="File.*did not exist within.*seconds"):
-                strategy.wait_until_ready(mock_container)
+                with caplog.at_level(logging.CRITICAL, logger="testcontainers.core.wait_strategies"):
+                    strategy.wait_until_ready(mock_container)
 
 
 class TestCompositeWaitStrategy:
@@ -615,7 +617,7 @@ def test_wait_until_ready_all_strategies_succeed(self):
         strategy2.wait_until_ready.assert_called_once_with(mock_container)
         strategy3.wait_until_ready.assert_called_once_with(mock_container)
 
-    def test_wait_until_ready_first_strategy_fails(self):
+    def test_wait_until_ready_first_strategy_fails(self, caplog):
         """Test that execution stops when first strategy fails."""
         strategy1 = Mock()
         strategy2 = Mock()
@@ -628,7 +630,8 @@ def test_wait_until_ready_first_strategy_fails(self):
         strategy1.wait_until_ready.side_effect = TimeoutError("First strategy failed")
 
         with pytest.raises(TimeoutError, match="First strategy failed"):
-            composite.wait_until_ready(mock_container)
+            with caplog.at_level(logging.CRITICAL, logger="testcontainers.core.wait_strategies"):
+                composite.wait_until_ready(mock_container)
 
         # Only first strategy should be called
         strategy1.wait_until_ready.assert_called_once_with(mock_container)

pyproject.toml

@@ -98,7 +98,7 @@ rabbitmq = ["pika>=1"]
 redis = ["redis>=7"]
 registry = ["bcrypt>=5"]
 selenium = ["selenium>=4"]
-scylla = ["cassandra-driver>=3"]
+scylla = ["cassandra-driver>=3; python_version < '3.14'"]
 sftp = ["cryptography"]
 vault = []
 weaviate = ["weaviate-client>=4"]
@@ -120,7 +120,7 @@ test = [
   "psycopg2-binary==2.9.11",
   "pg8000==1.31.5",
   "psycopg>=3",
-  "cassandra-driver>=3",
+  "cassandra-driver>=3; python_version < '3.14'",
   "kafka-python-ng>=2",
   "hvac>=2; python_version < '4.0'",
   "pymilvus>=2",
@@ -277,6 +277,7 @@ log_cli = true
 log_cli_level = "INFO"
 markers = [
   "inside_docker_check: mark test to be used to validate DinD/DooD is working as expected",
+  "long_running: mark test as very long running",
 ]
 filterwarnings = [
   "ignore:The @wait_container_is_ready decorator is deprecated.*:DeprecationWarning",