Merge pull request #140 from fvaleye/fix/rapl-negative-value

Fix negative value with RAPL wrap-around for Linux hardware

tests/hardwares/data/intel-rapl/intel-raplT0/intel-raplT0T0/max_energy_range_uj

@@ -0,0 +1 @@
+65532610987

tests/hardwares/data/intel-rapl/intel-raplT0/intel-raplT0T1/max_energy_range_uj

@@ -0,0 +1 @@
+65532610987

tests/hardwares/data/intel-rapl/intel-raplT0/max_energy_range_uj

@@ -0,0 +1 @@
+65532610987

tests/hardwares/data/intel-rapl/intel-raplT1/intel-raplT1T0/max_energy_range_uj

@@ -0,0 +1 @@
+65532610987

tests/hardwares/data/intel-rapl/intel-raplT1/intel-raplT1T1/max_energy_range_uj

@@ -0,0 +1 @@
+65532610987

tests/hardwares/data/intel-rapl/intel-raplT1/max_energy_range_uj

@@ -0,0 +1 @@
+65532610987

tests/hardwares/data/intel-rapl2/intel-raplT0/intel-raplT0T0/max_energy_range_uj

@@ -1 +1 @@
-50000
+70000

tests/hardwares/test_rapl.py

@@ -47,19 +47,28 @@ def by_energy_uj(rapl_result: RAPLResult) -> str:
 @pytest.mark.asyncio
 @pytest.mark.linux
 @pytest.mark.darwin
-async def test_get_rapl_power_usage_max_when_0():
+async def test_get_rapl_power_wrap_around_when_0():
     path = f"{pathlib.Path(__file__).parent.resolve()}/data/intel-rapl2"
+    two_seconds_ago = datetime.datetime.now() - datetime.timedelta(seconds=2)
     rapl_separator_for_windows = "T"
+    rapl_results = dict()
+    rapl_results["package-0"] = RAPLResult(
+        name="package", energy_uj=2, max_energy_uj=70000, timestamp=two_seconds_ago
+    )
+    rapl_results["core"] = RAPLResult(
+        name="core", energy_uj=1, max_energy_uj=70000, timestamp=two_seconds_ago
+    )
+    rapl = RAPL(
+        path=path, rapl_separator=rapl_separator_for_windows, rapl_results=rapl_results
+    )
+    host_energy_usage_expected = 35
+    cpu_energy_usage_expected = 35
 
-    rapl_results = await RAPL(
-        path=path, rapl_separator=rapl_separator_for_windows
-    ).get_rapl_power_usage()
-    assert rapl_results[0].name == "package-0"
-    assert rapl_results[0].energy_uj == 70000.0
-    assert rapl_results[0].timestamp is not None
-    assert rapl_results[1].name == "core"
-    assert rapl_results[1].energy_uj == 50000.0
-    assert rapl_results[1].timestamp is not None
+    energy_report = await rapl.get_energy_report()
+    energy_report.convert_unit(EnergyUsageUnit.MILLIWATT)
+    assert round(energy_report.host_energy_usage, 0) == host_energy_usage_expected
+    assert round(energy_report.cpu_energy_usage, 0) == cpu_energy_usage_expected
+    assert energy_report.memory_energy_usage is None
 
 
 @pytest.mark.asyncio
@@ -71,19 +80,19 @@ async def test_get_total_uj_one_call():
     one_minute_ago = datetime.datetime.now() - datetime.timedelta(seconds=60)
     rapl_results = dict()
     rapl_results["package-0"] = RAPLResult(
-        name="package", energy_uj=50000, timestamp=one_minute_ago
+        name="package", energy_uj=50000, max_energy_uj=70000, timestamp=one_minute_ago
     )
     rapl_results["core"] = RAPLResult(
-        name="core", energy_uj=40000, timestamp=one_minute_ago
+        name="core", energy_uj=40000, max_energy_uj=70000, timestamp=one_minute_ago
     )
     rapl = RAPL(
         path=path, rapl_separator=rapl_separator_for_windows, rapl_results=rapl_results
     )
-    host_energy_usage_expected = 0.333
-    cpu_energy_usage_expected = 0.167
+    host_energy_usage_expected = 0.33
+    cpu_energy_usage_expected = 0.5
 
     energy_report = await rapl.get_energy_report()
     energy_report.convert_unit(EnergyUsageUnit.MILLIWATT)
-    assert round(energy_report.host_energy_usage, 3) == host_energy_usage_expected
-    assert round(energy_report.cpu_energy_usage, 3) == cpu_energy_usage_expected
+    assert round(energy_report.host_energy_usage, 2) == host_energy_usage_expected
+    assert round(energy_report.cpu_energy_usage, 2) == cpu_energy_usage_expected
     assert energy_report.memory_energy_usage is None

tracarbon/exporters/exporter.py

@@ -7,6 +7,7 @@
 from pydantic import BaseModel
 
 from tracarbon.hardwares.hardware import HardwareInfo
+from tracarbon.locations import Location
 
 
 class Exporter(BaseModel, metaclass=ABCMeta):
@@ -126,6 +127,7 @@ class MetricGenerator(BaseModel):
 
     metrics: List[Metric]
     platform: str = HardwareInfo.get_platform()
+    location: Optional[Location] = None
 
     async def generate(self) -> AsyncGenerator[Metric, None]:
         """

tracarbon/general_metrics.py

@@ -12,7 +12,6 @@ class EnergyConsumptionGenerator(MetricGenerator):
     Energy consumption generator for energy consumption.
     """
 
-    location: Location
     energy_consumption: EnergyConsumption
 
     def __init__(self, location: Location, **data: Any) -> None:
@@ -50,7 +49,6 @@ class CarbonEmissionGenerator(MetricGenerator):
     Carbon emission generator to generate carbon emissions.
     """
 
-    location: Location
     carbon_emission: CarbonEmission
     co2signal_api_key: Optional[str] = None
 
@@ -98,7 +96,6 @@ class EnergyConsumptionKubernetesGenerator(MetricGenerator):
         Energy consumption generator for energy consumption of the containers.
         """
 
-        location: Location
         energy_consumption: EnergyConsumption
         kubernetes: Kubernetes
 
@@ -148,7 +145,6 @@ class CarbonEmissionKubernetesGenerator(MetricGenerator):
         Carbon emission generator to generate carbon emissions of the containers.
         """
 
-        location: Location
         carbon_emission: CarbonEmission
         kubernetes: Kubernetes
         co2signal_api_key: Optional[str] = None

tracarbon/hardwares/rapl.py

@@ -18,6 +18,7 @@ class RAPLResult(BaseModel):
 
     name: str
     energy_uj: float
+    max_energy_uj: float
     timestamp: datetime
 
 
@@ -82,15 +83,16 @@ async def get_rapl_power_usage(self) -> List[RAPLResult]:
                         f"{file_path}/energy_uj", "r"
                     ) as rapl_energy:
                         energy_uj = float(await rapl_energy.read())
-                        max_energy_uj_value_reached = energy_uj < 1
-                        if max_energy_uj_value_reached:
-                            async with aiofiles.open(
-                                f"{file_path}/max_energy_range_uj", "r"
-                            ) as max_energy_rapl_file:
-                                energy_uj = float(await max_energy_rapl_file.read())
+                    async with aiofiles.open(
+                        f"{file_path}/max_energy_range_uj", "r"
+                    ) as rapl_max_energy:
+                        max_energy_uj = float(await rapl_max_energy.read())
                     rapl_results.append(
                         RAPLResult(
-                            name=name, energy_uj=energy_uj, timestamp=datetime.now()
+                            name=name,
+                            energy_uj=energy_uj,
+                            max_energy_uj=max_energy_uj,
+                            timestamp=datetime.now(),
                         )
                     )
         except Exception as exception:
@@ -115,6 +117,13 @@ async def get_energy_report(self) -> EnergyUsage:
             time_difference = (
                 rapl_result.timestamp - previous_rapl_result.timestamp
             ).total_seconds()
+            if previous_rapl_result.energy_uj > rapl_result.energy_uj:
+                logger.debug(
+                    f"Wrap-around detected in RAPL {rapl_result.name}. The current RAPL energy value ({rapl_result.energy_uj}) is lower than previous value ({previous_rapl_result.energy_uj})."
+                )
+                rapl_result.energy_uj = (
+                    rapl_result.energy_uj + rapl_result.max_energy_uj
+                )
             watts = Power.watts_from_microjoules(
                 (
                     (rapl_result.energy_uj - previous_rapl_result.energy_uj)

tracarbon/hardwares/sensors.py

@@ -153,17 +153,19 @@ def __init__(self, instance_type: str, **data: Any) -> None:
 
                     for row in reader:
                         if row[0] == instance_type:
+                            data["cpu_idle"] = float(row[14].replace(",", "."))
+                            data["cpu_at_10"] = float(row[15].replace(",", "."))
+                            data["cpu_at_50"] = float(row[16].replace(",", "."))
+                            data["cpu_at_100"] = float(row[17].replace(",", "."))
+                            data["memory_idle"] = float(row[18].replace(",", "."))
+                            data["memory_at_10"] = float(row[19].replace(",", "."))
+                            data["memory_at_50"] = float(row[20].replace(",", "."))
+                            data["memory_at_100"] = float(row[21].replace(",", "."))
+                            data["has_gpu"] = float(row[22].replace(",", ".")) > 0
+                            data["delta_full_machine"] = float(
+                                row[26].replace(",", ".")
+                            )
                             super().__init__(
-                                cpu_idle=float(row[14].replace(",", ".")),
-                                cpu_at_10=float(row[15].replace(",", ".")),
-                                cpu_at_50=float(row[16].replace(",", ".")),
-                                cpu_at_100=float(row[17].replace(",", ".")),
-                                memory_idle=float(row[18].replace(",", ".")),
-                                memory_at_10=float(row[19].replace(",", ".")),
-                                memory_at_50=float(row[20].replace(",", ".")),
-                                memory_at_100=float(row[21].replace(",", ".")),
-                                has_gpu=float(row[22].replace(",", ".")) > 0,
-                                delta_full_machine=float(row[26].replace(",", ".")),
                                 **data,
                             )
                             return