Merge branch 'release/2.3.1'

monitor/__init__.py

@@ -13,7 +13,7 @@
 
 # package metadata
 __appname__     = 'monitor'
-__version__     = '2.3.0'
+__version__     = '2.3.1'
 __authors__     = 'Geoffrey Lentner'
 __contact__     = 'glentner@purdue.edu'
 __license__     = 'Apache Software License 2.0'

monitor/cli/__init__.py

@@ -23,20 +23,23 @@
     from signal import signal, SIGPIPE, SIG_DFL
     signal(SIGPIPE, SIG_DFL)
 
+# external libs
+from cmdkit.app import Application
+from cmdkit.cli import Interface, ArgumentError
+
 # internal libs
 from ..core.logging import Logger
 from ..core.exceptions import CompletedCommand
 from .. import (__appname__, __version__, __description__,
                 __copyright__, __license__, __website__)
 
-# external libs
-from cmdkit.app import Application
-from cmdkit.cli import Interface, ArgumentError
-
 # resource commands
 from .cpu import CPUDevice
 from .gpu import GPUDevice
 
+# public interface
+__all__ = ['ResourceMonitor', 'main', ]
+
 
 DEVICES = {
     'cpu': CPUDevice,

monitor/cli/cpu/__init__.py

@@ -14,19 +14,22 @@
 # standard libs
 import sys
 
-# internal libs
-from ...core.logging import Logger
-from ...core.exceptions import CompletedCommand
-from ... import __appname__
-
 # external libs
 from cmdkit.app import Application
 from cmdkit.cli import Interface, ArgumentError
 
+# internal libs
+from ... import __appname__
+from ...core.logging import Logger
+from ...core.exceptions import CompletedCommand
+
 # resource commands
 from .percent import CPUPercent
 from .memory import CPUMemory
 
+# public interface
+__all__ = ['CPUDevice', ]
+
 
 RESOURCES = {
     'percent': CPUPercent,

monitor/cli/cpu/memory.py

@@ -18,15 +18,18 @@
 import time
 import functools
 
+# external libs
+import psutil
+from cmdkit.app import Application, exit_status
+from cmdkit.cli import Interface, ArgumentError
+
 # internal libs
 from ...core.exceptions import log_and_exit
 from ...core.logging import Logger, PLAIN_HANDLER, CSV_HANDLER
 from ... import __appname__
 
-# external libs
-import psutil
-from cmdkit.app import Application, exit_status
-from cmdkit.cli import Interface, ArgumentError
+# public interface
+__all__ = ['CPUMemory', ]
 
 
 PROGRAM = f'{__appname__} cpu memory'

monitor/cli/cpu/percent.py

@@ -18,15 +18,18 @@
 import time
 import functools
 
+# external libs
+import psutil
+from cmdkit.app import Application, exit_status
+from cmdkit.cli import Interface, ArgumentError
+
 # internal libs
 from ...core.exceptions import log_and_exit
 from ...core.logging import Logger, PLAIN_HANDLER, CSV_HANDLER
 from ... import __appname__
 
-# external libs
-import psutil
-from cmdkit.app import Application, exit_status
-from cmdkit.cli import Interface, ArgumentError
+# public interface
+__all__ = ['CPUPercent', ]
 
 
 PROGRAM = f'{__appname__} cpu percent'

monitor/cli/gpu/__init__.py

@@ -14,21 +14,24 @@
 # standard libs
 import sys
 
+# external libs
+from cmdkit.app import Application
+from cmdkit.cli import Interface, ArgumentError
+
 # internal libs
 from ... import __appname__
 from ...core.logging import Logger
 from ...core.exceptions import CompletedCommand
 
-# external libs
-from cmdkit.app import Application
-from cmdkit.cli import Interface, ArgumentError
-
 # resource commands
 from .percent import GPUPercent
 from .memory import GPUMemory
 from .power import GPUPower
 from .temp import GPUTemp
 
+# public interface
+__all__ = ['GPUDevice', ]
+
 
 RESOURCES = {
     'percent': GPUPercent,

monitor/cli/gpu/memory.py

@@ -18,15 +18,18 @@
 import time
 import functools
 
+# external libs
+from cmdkit.app import Application, exit_status
+from cmdkit.cli import Interface, ArgumentError
+
 # internal libs
 from ... import __appname__
 from ...contrib import SMIData
 from ...core.exceptions import log_and_exit
 from ...core.logging import Logger, PLAIN_HANDLER, CSV_HANDLER
 
-# external libs
-from cmdkit.app import Application, exit_status
-from cmdkit.cli import Interface, ArgumentError
+# public interface
+__all__ = ['GPUMemory', ]
 
 
 PROGRAM = f'{__appname__} gpu memory'

monitor/cli/gpu/percent.py

@@ -18,15 +18,18 @@
 import time
 import functools
 
+# external libs
+from cmdkit.app import Application, exit_status
+from cmdkit.cli import Interface, ArgumentError
+
 # internal libs
 from ... import __appname__
 from ...contrib import SMIData
 from ...core.exceptions import log_and_exit
 from ...core.logging import Logger, PLAIN_HANDLER, CSV_HANDLER
 
-# external libs
-from cmdkit.app import Application, exit_status
-from cmdkit.cli import Interface, ArgumentError
+# public interface
+__all__ = ['GPUPercent', ]
 
 
 PROGRAM = f'{__appname__} gpu percent'

monitor/cli/gpu/power.py

@@ -18,15 +18,18 @@
 import time
 import functools
 
+# external libs
+from cmdkit.app import Application, exit_status
+from cmdkit.cli import Interface, ArgumentError
+
 # internal libs
 from ... import __appname__
 from ...contrib import SMIData
 from ...core.exceptions import log_and_exit
 from ...core.logging import Logger, PLAIN_HANDLER, CSV_HANDLER
 
-# external libs
-from cmdkit.app import Application, exit_status
-from cmdkit.cli import Interface, ArgumentError
+# public interface
+__all__ = ['GPUPower', ]
 
 
 PROGRAM = f'{__appname__} gpu power'

monitor/cli/gpu/temp.py

@@ -18,15 +18,18 @@
 import time
 import functools
 
+# external libs
+from cmdkit.app import Application, exit_status
+from cmdkit.cli import Interface, ArgumentError
+
 # internal libs
 from ... import __appname__
 from ...contrib import SMIData
 from ...core.exceptions import log_and_exit
 from ...core.logging import Logger, PLAIN_HANDLER, CSV_HANDLER
 
-# external libs
-from cmdkit.app import Application, exit_status
-from cmdkit.cli import Interface, ArgumentError
+# public interface
+__all__ = ['GPUTemp', ]
 
 
 PROGRAM = f'{__appname__} gpu temp'

monitor/contrib/__init__.py

@@ -24,6 +24,9 @@
 from .nvidia import NvidiaPercent, NvidiaMemory, NvidiaTemperature, NvidiaPower
 from .rocm import RocmPercent, RocmMemory, RocmTemperature, RocmPower
 
+# public interface
+__all__ = ['SMIData', ]
+
 
 class SMIData:
     """High-level interface to external smi tool for GPU telemetry."""

monitor/contrib/nvidia.py

@@ -21,6 +21,9 @@
 # internal libs
 from ..core.extern import ExternalMetric
 
+# public interface
+__all__ = ['NvidiaMetric', 'NvidiaPercent', 'NvidiaMemory', 'NvidiaPower', 'NvidiaTemperature', ]
+
 
 class NvidiaMetric(ExternalMetric, ABC):
     """Status object for Nvidia GPU resource."""

monitor/contrib/rocm.py

@@ -22,6 +22,9 @@
 # internal libs
 from ..core.extern import ExternalMetric
 
+# public interface
+__all__ = ['RocmMetric', 'RocmPercent', 'RocmMemory', 'RocmPower', 'RocmTemperature', ]
+
 
 class RocmMetric(ExternalMetric, ABC):
     """Run rocm-smi to collect metrics on GPU usage."""

monitor/core/exceptions.py

@@ -14,6 +14,9 @@
 # type annotations
 from typing import Callable
 
+# public interface
+__all__ = ['CompletedCommand', 'log_and_exit', ]
+
 
 class CompletedCommand(Exception):
     """Lift exit_status of sub-commands `main` method."""

monitor/core/extern.py

@@ -8,7 +8,7 @@
 # You should have received a copy of the Apache License along with this program.
 # If not, see <https://www.apache.org/licenses/LICENSE-2.0>.
 
-"""Stat class implementations for external tools (e.g., nvidia-smi, rocm-smi)."""
+"""Simple support for external metrics (subprocess call to external commands)."""
 
 
 # type annotations
@@ -19,10 +19,15 @@
 from abc import ABC, abstractclassmethod
 from subprocess import check_output
 
+# public interface
+__all__ = ['ExternalMetric', ]
+
+
 # value type for metrics
 # either individual value, a list, or indexed collection
 T = TypeVar('T', float, List[float], Dict[int, float])
 
+
 class ExternalMetric(ABC):
     """Structured object from text-block."""
 

monitor/core/logging.py

@@ -29,11 +29,14 @@
 # internal library
 from .. import __appname__
 
+# public interface
+__all__ = ['HOSTNAME', 'Logger', 'PLAIN_HANDLER', 'CSV_HANDLER', ]
 
-# get hostname from `socket` instead of `.config`
+
+# cache hostname
 HOSTNAME = socket.gethostname()
 
-# NOTICE messages won't actually be formatted with color.
+
 LEVELS = levels.Level.from_names(['DEBUG', 'INFO', 'WARNING', 'ERROR', 'CRITICAL'])  # noqa: typing is broken
 COLORS = colors.Color.from_names(['blue', 'green', 'yellow', 'red', 'magenta'])  # noqa: typing is broken
 RESET = colors.Color.reset
@@ -67,7 +70,7 @@ def with_name(cls, name: str) -> Logger:
 
     @property
     def handlers(self) -> List[handlers.Handler]:
-        """Overide of local handlers to global list."""
+        """Override of local handlers to global list."""
         global _handlers
         return _handlers