Displays an action count (per second) and total for a time window from a Temporal dev server

temporal-dev-server-actions-count.py

import time
import requests
import argparse
from prometheus_client.parser import text_string_to_metric_families

# Prometheus scrape endpoint
PROMETHEUS_SCRAPE_URL = "http://localhost:63626/metrics"
EXCLUDED_NAMESPACE = "temporal_system"  # Namespace to exclude

def scrape_metrics():
    """Fetches metrics from the Prometheus endpoint and returns parsed data."""
    response = requests.get(PROMETHEUS_SCRAPE_URL)
    response.raise_for_status()  # Raise an error if the request failed
    return list(text_string_to_metric_families(response.text))

def get_action_count(metrics_data):
    """Extracts and returns the total 'action' count from the parsed metrics data, excluding the specified namespace."""
    total_actions = 0
    for family in metrics_data:
        if family.name == 'action':
            for sample in family.samples:
                # Exclude the specified namespace
                if sample.labels.get('namespace') != EXCLUDED_NAMESPACE:
                    total_actions += sample.value
    return total_actions

def print_readme():
    """Prints a README message when no argument is provided."""
    readme_message = """
    Usage: python temporal-dev-server-actions-count.py --time-window-seconds <time_window_in_seconds>
    
    This script monitors Prometheus action metrics and calculates the average actions per second 
    and the total number of actions over the specified time window, excluding the 'temporal_system' namespace.

    Arguments:
      --time-window-seconds: The time period in seconds over which the total actions will be 
                             calculated and reported.
      
    Example:
      python temporal-dev-server-actions-count.py --time-window-seconds 120
    """
    print(readme_message)

def main(time_window):
    print(f"Starting Prometheus action metric monitoring with a time window of {time_window} seconds...")
    print(f"Monitoring Prometheus endpoint: {PROMETHEUS_SCRAPE_URL} (excluding namespace: {EXCLUDED_NAMESPACE})")
    print(f"Please wait, the total number of actions will be reported after {time_window} seconds...")

    last_action_count = None
    total_action_delta = 0
    start_time = time.time()

    while True:
        try:
            # Scrape and parse the metrics data
            metrics_data = scrape_metrics()

            # Get the current action count from the metrics data
            current_action_count = get_action_count(metrics_data)

            # Calculate the difference between the current and last action count
            if last_action_count is not None:
                action_delta = current_action_count - last_action_count
                total_action_delta += action_delta
                actions_per_second = action_delta / 1  # Calculating actions per second

                # Report the average actions per second every second
                print(f"Current average actions per second: {actions_per_second:.2f}")

            # Update the last action count
            last_action_count = current_action_count

            # Wait for 1 second before the next scrape
            time.sleep(1)

            # Check if the total time has reached the specified time window
            if time.time() - start_time >= time_window:
                break

        except Exception as e:
            print(f"An error occurred: {e}")
            break

    # Print the total number of actions after the time window
    print(f"Total actions in the last {time_window} seconds: {total_action_delta}")
    print("Monitoring completed.")

if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="Monitor Prometheus action metrics and calculate average actions per second and total actions over a given time window, excluding the 'temporal_system' namespace.")
    parser.add_argument("--time-window-seconds", type=int, help="The time period in seconds to capture the action metrics.")

    args = parser.parse_args()

    if args.time_window_seconds:
        main(args.time_window_seconds)
    else:
        print_readme()