ecs-task-metrics.py

gistfile1.py

#!/usr/bin/env python

from __future__ import print_function

from datetime import datetime
import boto3

print('Loading function...')

now = datetime.utcnow()

ecs_client = boto3.client('ecs')
cw_client = boto3.client('cloudwatch')

namespace = 'CG/ECS'

def list_clusters():
    cluster_arns = ecs_client.list_clusters()['clusterArns']
    clusters_info = ecs_client.describe_clusters(clusters=cluster_arns)['clusters']
    
    return [{'clusterName': cluster['clusterName'], 'clusterArn': cluster['clusterArn']} for cluster in clusters_info]

def list_services(cluster_arn):
    paginator = ecs_client.get_paginator('list_services')
    response_iterator = paginator.paginate(cluster=cluster_arn)
    
    return [serviceArn
        for service in response_iterator
        for serviceArn in service['serviceArns']]
        
def get_metric_data(cluster_arn, cluster_name, services):
    response = ecs_client.describe_services(cluster=cluster_arn, services=services)
    
    cluster_desired_count = 0.0
    cluster_running_count = 0.0
    cluster_pending_count = 0.0
    cluster_waiting_count = 0.0
    
    metrics = []
    for service in response['services']:
        service_name  = service['serviceName']
        service_desired_count = float(service['desiredCount'])
        service_running_count = float(service['runningCount'])
        service_pending_count = float(service['pendingCount'])
        service_waiting_count = float(service_desired_count - service_running_count) if service_desired_count > service_running_count else 0.0
        
        cluster_desired_count += service_desired_count
        cluster_running_count += service_running_count
        cluster_pending_count += service_pending_count
        cluster_waiting_count += service_waiting_count
        
        service_dimensions = [{'Name': 'ClusterName',
                       'Value': cluster_name},
                      {'Name': 'ServiceName',
                       'Value': service_name}]

        service_metric_data = [{
                'MetricName': 'DesiredCount',
                'Dimensions': service_dimensions,
                'Value': service_desired_count,
                'Unit': 'None'
            },{
                'MetricName': 'RunningCount',
                'Dimensions': service_dimensions,
                'Value': service_pending_count,
                'Unit': 'None'
            },{
                'MetricName': 'PendingCount',
                'Dimensions': service_dimensions,
                'Value': service_pending_count,
                'Unit': 'None'
            },{
                'MetricName': 'WaitingCount',
                'Dimensions': service_dimensions,
                'Value': service_waiting_count,
                'Unit': 'None'
            }]
        
        service_metric = {'Namespace': namespace, 'MetricData': service_metric_data}
        metrics.append(service_metric)
    
    cluster_dimensions = [{'Name': 'ClusterName', 'Value': cluster_name}]
    cluster_metric_data = [{
                'MetricName': 'DesiredCount',
                'Dimensions': cluster_dimensions,
                'Value': cluster_desired_count,
                'Unit': 'None'
            },{
                'MetricName': 'RunningCount',
                'Dimensions': cluster_dimensions,
                'Value': cluster_pending_count,
                'Unit': 'None'
            },{
                'MetricName': 'PendingCount',
                'Dimensions': cluster_dimensions,
                'Value': cluster_pending_count,
                'Unit': 'None'
            },{
                'MetricName': 'WaitingCount',
                'Dimensions': cluster_dimensions,
                'Value': cluster_waiting_count,
                'Unit': 'None'
            }]
    cluster_metric = {'Namespace': namespace, 'MetricData': cluster_metric_data}
    metrics.append(cluster_metric)
    
    return metrics

def lambda_handler(event, context):
    dimension_count = 0
    metric_count = 0
    for cluster in list_clusters():
        services = list_services(cluster_arn=cluster['clusterArn'])
        metrics = get_metric_data(cluster_arn=cluster['clusterArn'],
                                  cluster_name=cluster['clusterName'],
                                  services=services)

        for metric_kwargs in metrics:
            try:
                cw_client.put_metric_data(**metric_kwargs)
                dimension_count += 1
                metric_count += len(metric_kwargs['MetricData'])
            except Exception as e:
                print('ERROR: {error}'.format(error=e))
    msg = 'Pushed {dimension_count} dimensions and {metric_count} metrics'.format(dimension_count=dimension_count, metric_count=metric_count)
    print(msg)
    return msg