mtnlog is a simple multinode performance logger for Python. It is designed to be used in a similar way to Python's built-in logging module, but with a focus on performance logging. It provides a simple API for logging performance data, including start and end times, and allows for easy integration with other logging systems.
You can install mtnlog using pip:
pip install mtnlogTo use mtnlog, you have two features: JSONLogger and PerformanceLogger and PerformancePlotter.
The JSONLogger class is a simple logger that writes performance data to a JSON file. You can create a new JSONLogger instance by passing a file path to the constructor:
from mtnlog import JSONLogger
logger = JSONLogger(log_dir='logs') # logs is the directory where the log file will be savedYou can then use the log method to log performance data:
logger.log('<your_dict>', filename='log') # your_dict is a dictionary with the data you want to log / filename is the name of the fileyour_dict is a dictionary with the data you want to log.
filename is the name of the file where the data will be saved
The PerformanceLogger class is a logger for system performance data. It logs the the time taken to execute the block, as well as the CPU, memory, and GPU usage. You can create a new PerformanceLogger instance by passing a file path to the constructor:
from mtnlog import PerformanceLogger
collector = PerformanceLogger(log_dir="<your_log_dir>", log_node="<current_node>")your_log_dir is the directory where the log file will be saved.
current_node is the number of the node you are logging.
You can then use the change_tag method to change the tag of the log:
collector.change_tag("<new_tag>")new_tag is the new tag you want to use.
To stop logging, you can use the stop method:
collector.stop()The PerformancePlotter class is a plotter for system performance data. It plots the time taken to execute the block, as well as the CPU, memory, GPU, and network usage. You can create a new PerformancePlotter instance by passing a file path to the constructor:
from mtnlog import PerformancePlotter
plotter = PerformancePlotter(base_dir="<your_base_dir>", log_node="<current_node>")your_base_dir is the base directory where the log file will be saved.
current_node is the number of the node you are logging.
You can then use the plot method to plot the data:
plotter.plot()Here is an example of how to use mtnlog:
from mtnlog import JSONLogger, PerformanceLogger, PerformancePlotter
# Create a JSONLogger instance
logger = JSONLogger(log_dir='logs')
# Log some data
logger.log({'message': 'Hello, world!'}, filename='log')
# Create a PerformanceLogger instance
collector = PerformanceLogger(log_dir='logs', log_node="0")
# Change the tag
collector.change_tag('new_tag')
# Stop logging
collector.stop()
# Create a PerformancePlotter instance
plotter = PerformancePlotter(base_dir='logs', log_node="0")
# Plot the data
plotter.plot()