Tests for RAPL and ZeusMonitor (#100)

Co-authored-by: Jae-Won Chung <[email protected]>

pyproject.toml

@@ -94,6 +94,7 @@ pydocstyle.convention = "google"
 "zeus/optimizer/batch_size/common.py" = ["N805"]
 "zeus/device/gpu/*.py" = ["N802", "N803"]
 "zeus/device/cpu/*.py" = ["N802"]
+"zeus/utils/testing.py" = ["N802"]
 
 [tool.pytest.ini_options]
 addopts = "--numprocesses auto"

tests/device/cpu/test_rapl.py

@@ -4,14 +4,12 @@
 import os
 import pytest
 from typing import Generator, TYPE_CHECKING, Sequence
-from time import sleep
 from unittest.mock import patch, mock_open, create_autospec, MagicMock
-import unittest.mock as mock
-from sys import stdout, stderr
 import warnings
 
 import multiprocessing as mp
 
+
 if TYPE_CHECKING:
     from pathlib import Path
 
@@ -56,24 +54,32 @@ def __exit__(self, exc_type, exc_value, traceback):
 @pytest.fixture
 def mock_rapl_values():
     rapl_values = [
-        "1000",
-        "900",
-        "800",
-        "700",
-        "600",
-        "500",
-        "400",
-        "500",
-        "200",
-        "100",
+        "100000",
+        "90000",
+        "80000",
+        "70000",
+        "60000",
+        "50000",
+        "40000",
+        "50000",
+        "20000",
+        "10000",
     ]
     mocked_rapl_file = MockRaplFile(RAPL_DIR + "/intel-rapl:0/energy_uj", rapl_values)
+    mocked_rapl_file_name = mock_open()
+    mocked_rapl_file_name.return_value.read.return_value = "package"
+    mocked_rapl_file_max = mock_open()
+    mocked_rapl_file_max.return_value.read.return_value = "100000"
 
     real_open = builtins.open
 
     def mock_file_open(filepath, *args, **kwargs):
         if filepath == (RAPL_DIR + "/intel-rapl:0/energy_uj"):
             return mocked_rapl_file
+        if filepath == (RAPL_DIR + "/intel-rapl:0/name"):
+            return mocked_rapl_file_name()
+        if filepath == (RAPL_DIR + "/intel-rapl:0/max_energy_range_uj"):
+            return mocked_rapl_file_max()
         else:
             return real_open(filepath, *args, **kwargs)
 
@@ -92,8 +98,155 @@ def mock_file_open(filepath, *args, **kwargs):
     patch_sleep.stop()
 
 
+@pytest.fixture()
+def mock_rapl_wraparound_tracker():
+    patch_tracker = patch("zeus.device.cpu.rapl.RaplWraparoundTracker")
+    MockRaplWraparoundTracker = patch_tracker.start()
+
+    mock_tracker = MockRaplWraparoundTracker.return_value
+    mock_tracker.get_num_wraparounds.side_effect = [0, 5]
+
+    yield mock_tracker
+
+    patch_tracker.stop()
+
+
 def test_rapl_polling_process(mock_rapl_values):
     wraparound_counter = mp.Value("i", 0)
     with pytest.raises(MockRaplFileOutOfValues) as exception:
         _polling_process(RAPL_DIR + "/intel-rapl:0/energy_uj", 1000, wraparound_counter)
     assert wraparound_counter.value == 8
+
+
+# RAPLFile tests
+@pytest.fixture
+@patch("os.path.exists", return_value=False)
+def test_rapl_available(mock_exists):
+    assert rapl_is_available() == False
+
+
+def test_rapl_file_class(mock_rapl_values, mock_rapl_wraparound_tracker):
+    """Test the `RAPLFile` class."""
+    # Test initialization
+    raplFile = RAPLFile("/sys/class/powercap/intel-rapl/intel-rapl:0")
+    assert raplFile.name == "package"
+    assert raplFile.last_energy == 100000.0
+    assert raplFile.max_energy_range_uj == 100000.0
+
+    # Test read method where get_num_wraparounds is 0
+    assert raplFile.read() == 90.0
+
+    # Test read method where get_num_wraparounds is 5
+    assert raplFile.read() == 580.0  # (80000+5*100000)/1000
+
+
+def test_rapl_file_class_exceptions():
+    """Test `RAPLFile` Init errors"""
+    with patch("builtins.open", mock_open()) as mock_file:
+        # Fails to open name file
+        mock_file.side_effect = FileNotFoundError
+        with pytest.raises(ZeusRAPLFileInitError):
+            RAPLFile("/sys/class/powercap/intel-rapl/intel-rapl:0")
+
+        # Fails to read energy_uj file
+        mock_file.side_effect = [
+            mock_open(read_data="package").return_value,
+            FileNotFoundError,
+        ]
+        with pytest.raises(ZeusRAPLFileInitError):
+            RAPLFile("/sys/class/powercap/intel-rapl/intel-rapl:0")
+
+        # Fails to read max_energy_uj file
+        mock_file.side_effect = [
+            mock_open(read_data="package").return_value,
+            mock_open(read_data="1000000").return_value,
+            FileNotFoundError,
+        ]
+        with pytest.raises(ZeusRAPLFileInitError):
+            RAPLFile("/sys/class/powercap/intel-rapl/intel-rapl:0")
+
+
+# RAPLCPU tests
+@pytest.fixture()
+def mock_os_listdir_cpu(mocker):
+    return mocker.patch("os.listdir", return_value=["intel-rapl:0", "intel-rapl:0:0"])
+
+
+def create_rapl_file_mock(name="package", read_value=1000.0):
+    """Create a mock `RAPLFile` class"""
+    mock_rapl_file = create_autospec(RAPLFile, instance=True)
+    mock_rapl_file.name = name
+    mock_rapl_file.read.return_value = read_value
+    return mock_rapl_file
+
+
+def test_rapl_cpu_class(mocker, mock_os_listdir_cpu):
+    """Test `RAPLCPU` with `DRAM`"""
+    mock_rapl_file_package = create_rapl_file_mock()
+    mock_rapl_file_dram = create_rapl_file_mock(name="dram", read_value=500.0)
+
+    def rapl_file_side_effect(path):
+        if "0:0" in path:
+            return mock_rapl_file_dram
+        return mock_rapl_file_package
+
+    mocker.patch("zeus.device.cpu.rapl.RAPLFile", side_effect=rapl_file_side_effect)
+    cpu = RAPLCPU(cpu_index=0)
+    measurement = cpu.getTotalEnergyConsumption()
+
+    assert cpu.path == os.path.join(RAPL_DIR, "intel-rapl:0")
+    assert cpu.rapl_file == mock_rapl_file_package
+    assert cpu.dram == mock_rapl_file_dram
+    assert measurement.cpu_mj == mock_rapl_file_package.read.return_value
+    assert measurement.dram_mj == mock_rapl_file_dram.read.return_value
+
+
+def test_rapl_cpu_class_exceptions(mocker, mock_os_listdir_cpu):
+    """Test `RAPLCPU` subpackage init error"""
+    mock_rapl_file_package = create_rapl_file_mock()
+    mock_rapl_file_dram = create_rapl_file_mock(name="dram", read_value=500.0)
+
+    def rapl_file_side_effect(path):
+        if "0:0" in path:
+            raise ZeusRAPLFileInitError("Initilization Error")
+        return mock_rapl_file_package
+
+    mocker.patch("zeus.device.cpu.rapl.RAPLFile", side_effect=rapl_file_side_effect)
+    with warnings.catch_warnings(record=True) as w:
+        cpu = RAPLCPU(cpu_index=0)
+        assert "Failed to initialize subpackage" in str(w[-1].message)
+
+    assert cpu.path == os.path.join(RAPL_DIR, "intel-rapl:0")
+    assert cpu.rapl_file == mock_rapl_file_package
+    assert cpu.dram is None
+
+
+# RAPLCPUs tests
+def test_rapl_cpus_class(mocker):
+    """Test initialization when RAPL is available."""
+    mocker.patch("zeus.device.cpu.rapl.rapl_is_available", return_value=True)
+    mocker.patch(
+        "zeus.device.cpu.rapl.glob",
+        return_value=[f"{RAPL_DIR}/intel-rapl:0", f"{RAPL_DIR}/intel-rapl:1"],
+    )
+    mock_rapl_cpu_constructor = mocker.patch("zeus.device.cpu.rapl.RAPLCPU")
+    mock_rapl_cpu_instance = MagicMock(spec=RAPLCPU)
+    mock_rapl_cpu_constructor.side_effect = [
+        mock_rapl_cpu_instance,
+        mock_rapl_cpu_instance,
+    ]
+    rapl_cpus = RAPLCPUs()
+
+    assert len(rapl_cpus.cpus) == 2
+    assert all(isinstance(cpu, MagicMock) for cpu in rapl_cpus.cpus)
+    assert mock_rapl_cpu_constructor.call_count == 2
+
+
+def test_rapl_cpus_class_init_error(mocker):
+    """Test initialization when RAPL is not available."""
+    mocker.patch("zeus.device.cpu.rapl.rapl_is_available", return_value=False)
+
+    with pytest.raises(
+        ZeusRAPLNotSupportedError, match="RAPL is not supported on this CPU."
+    ):
+        RAPLCPUs()

tests/test_monitor.py

@@ -25,6 +25,8 @@
 from zeus.monitor import Measurement, ZeusMonitor
 from zeus.utils.testing import ReplayZeusMonitor
 import zeus.device.gpu
+import zeus.device.cpu
+from zeus.device.cpu.common import CpuDramMeasurement, CPUs, CPU
 
 if TYPE_CHECKING:
     from pathlib import Path
@@ -37,12 +39,56 @@
     pynvml.NVML_DEVICE_ARCH_VOLTA,
     pynvml.NVML_DEVICE_ARCH_AMPERE,
 ]
+NUM_CPUS = 2
+
+
+class MockCPU(CPU):
+    """Control a single MOCK CPU for testing."""
+
+    def __init__(self, index):
+        """Initialize the MOCKCPU with a specified index for testing."""
+        self.index = index
+        self.cpu_energy = itertools.count(start=1000, step=10)
+        self.dram_energy = (
+            itertools.count(start=200, step=5) if self.index % 2 == 0 else None
+        )
+
+    def getTotalEnergyConsumption(self):
+        """Returns the total energy consumption of the specified powerzone. Units: mJ."""
+        return CpuDramMeasurement(
+            cpu_mj=float(next(self.cpu_energy)),
+            dram_mj=float(next(self.dram_energy))
+            if self.dram_energy is not None
+            else None,
+        )
+
+    def supportsGetDramEnergyConsumption(self):
+        """Returns True if the specified CPU powerzone supports retrieving the subpackage energy consumption."""
+        return self.dram_energy is not None
+
+
+class MockCPUs(CPUs):
+    """MOCK CPU Manager object, containing individual MOCKCPU objects for testing."""
+
+    def __init__(self):
+        """Instantiates MOCKCPUs object for testing."""
+        self._cpus = [MockCPU(i) for i in range(NUM_CPUS)]
+
+    @property
+    def cpus(self) -> Sequence[CPU]:
+        """Returns a list of CPU objects being tracked."""
+        return self._cpus
+
+    def __del__(self) -> None:
+        """Shuts down the Mock CPU monitoring."""
+        return
 
 
 @pytest.fixture(autouse=True, scope="function")
-def reset_gpus() -> None:
-    """Reset the global variable `_gpus` to None on every test."""
+def reset_gpus_and_cpus() -> None:
+    """Reset the global variable `_gpus` and `_cpus` to None on every test."""
     zeus.device.gpu._gpus = None
+    zeus.device.cpu._cpus = None
 
 
 @pytest.fixture
@@ -209,11 +255,14 @@ def get_energy(self, start: float, end: float) -> dict[int, float]:
 
     # want to make zeus.device.gpu.nvml_is_available is a function, want it to always return true when testing
     mocker.patch("zeus.device.gpu.nvml_is_available", return_value=True)
+    mocker.patch("zeus.device.cpu._cpus", new=MockCPUs())
 
     ########################################
     # Test ZeusMonitor initialization.
     ########################################
-    monitor = ZeusMonitor(gpu_indices=gpu_indices, log_file=log_file)
+    monitor = ZeusMonitor(
+        gpu_indices=gpu_indices, cpu_indices=list(range(NUM_CPUS)), log_file=log_file
+    )
 
     # Check GPU index parsing from the log file.
     replay_monitor = ReplayZeusMonitor(gpu_indices=None, log_file=log_file)
@@ -227,6 +276,10 @@ def tick():
         next(time_counter)
         for counter in energy_counters.values():
             next(counter)
+        for i in range(len(monitor.cpu_indices)):
+            next(monitor.cpus._cpus[i].cpu_energy)
+            if i % 2 == 0:
+                next(monitor.cpus._cpus[i].dram_energy)
 
     def assert_window_begin(name: str, begin_time: int):
         """Assert monitor measurement states right after a window begins."""
@@ -237,6 +290,16 @@ def assert_window_begin(name: str, begin_time: int):
             for i in torch_gpu_indices
             if not is_old_torch[i]
         }
+        assert monitor.measurement_states[name].cpu_energy == {
+            i: pytest.approx((1000 + 10 * (begin_time - 4)) / 1000.0)
+            for i in range(len(monitor.cpu_indices))
+        }
+        assert monitor.measurement_states[name].dram_energy == {
+            i: pytest.approx((200 + 5 * (begin_time - 4)) / 1000.0)
+            if i % 2 == 0
+            else None
+            for i in range(0, len(monitor.cpu_indices), 2)
+        }
         pynvml_mock.nvmlDeviceGetTotalEnergyConsumption.assert_has_calls(
             [call(f"handle{i}") for i in nvml_gpu_indices if not is_old_nvml[i]]
         )
@@ -269,6 +332,18 @@ def assert_measurement(
                     elapsed_time * 3 / 1000.0
                 )
 
+        if measurement.cpu_energy is not None:
+            for i in measurement.cpu_energy.keys():
+                assert measurement.cpu_energy[i] == pytest.approx(
+                    elapsed_time * 10 / 1000.0
+                )
+
+        if measurement.dram_energy is not None:
+            for i in measurement.dram_energy.keys():
+                assert measurement.dram_energy[i] == pytest.approx(
+                    elapsed_time * 5 / 1000.0
+                )
+
         if not assert_calls:
             return
 

zeus/device/exception.py

@@ -16,6 +16,7 @@ class ZeusBaseCPUError(ZeusBaseError):
 
     def __init__(self, message: str) -> None:
         """Initialize Base Zeus Exception."""
+        super().__init__(message)
 
 
 class ZeusdError(ZeusBaseGPUError):