Sample python code for a autoscaler

AutoScaler.py

import datetime
import boto.ec2.cloudwatch
import boto.ec2.autoscale
import time

import numpy as np


def gradient(foo):
    graient_list = []
    previous = foo[0]
    for i in range(1, len(foo)):
        graient_list.append(foo[i] - previous)
        previous = foo[i]
    return graient_list


debug = 0

AWS_ACCESS_KEY_ID = 'YOUR_KEY_ID'
AWS_SECRET_ACCESS_KEY = 'YOUR_ACCESS_KEY'

# boto.ec2.connect_to_region('us-east-1', aws_access_key_id=AWS_ACCESS_KEY_ID, aws_secret_access_key=AWS_SECRET_ACCESS_KEY)
cloudwatch = boto.ec2.cloudwatch.connect_to_region('us-west-2', aws_access_key_id=AWS_ACCESS_KEY_ID,
                                                   aws_secret_access_key=AWS_SECRET_ACCESS_KEY)
autoscale = boto.ec2.autoscale.connect_to_region('us-west-2', aws_access_key_id=AWS_ACCESS_KEY_ID,
                                                 aws_secret_access_key=AWS_SECRET_ACCESS_KEY)

metrics = cloudwatch.list_metrics()
end = datetime.datetime.utcnow()
start = end - datetime.timedelta(minutes=1)

counter = 1

while ('true'):

    try:
        data_4_mins = cloudwatch.get_metric_statistics(
            60,
            datetime.datetime.utcnow() - datetime.timedelta(seconds=180),
            datetime.datetime.utcnow(),
            'CPUUtilization',
            'AWS/EC2',
            'Average',
            dimensions={'ImageId': ['ami-59055369']}
        )

        data_3_mins = cloudwatch.get_metric_statistics(
            60,
            datetime.datetime.utcnow() - datetime.timedelta(seconds=180),
            datetime.datetime.utcnow(),
            'CPUUtilization',
            'AWS/EC2',
            'Average',
            dimensions={'ImageId': ['ami-59055369']}
        )

        cpu_levels_4 = []
        for dataPoint in data_4_mins:
            avg = dataPoint['Average']
            cpu_levels_4.append(avg)

        cpu_levels_3 = []
        for dataPoint in data_3_mins:
            avg = dataPoint['Average']
            cpu_levels_3.append(avg)

        average_cpu_level = 0
        average_cpu_change = 0

        if 0 != len(cpu_levels_3):
            average_cpu_level = sum(cpu_levels_3) / len(cpu_levels_3)

        if 0 != len(cpu_levels_4):
            average_cpu_change = np.average(gradient(cpu_levels_4))

        print "average", average_cpu_level
        if counter == 2:
            print "average change", average_cpu_change
        print "cpu levels", cpu_levels_3

        instances = autoscale.get_all_autoscaling_instances()

        if len(data_3_mins) > 0:

            desired_capacity = 0
            # first iteration
            if counter == 2 and average_cpu_change >= 40 and average_cpu_level > 30:
                print("increasing the capacity by 2 instances due to sudden change")
                desired_capacity += 2
            elif counter == 2 and average_cpu_change >= 20 and average_cpu_level > 30:
                print("increasing the capacity by 1 due to sudden change")
                desired_capacity += 1

            if average_cpu_level > 70:
                print("increasing the capacity due to high cpu average ")
                desired_capacity += 1

            autoscale.set_desired_capacity("70-node-app-ac", len(instances) + desired_capacity)

        print("================================")
        time.sleep(60)
        if counter == 2:
            counter = 1
        else:
            counter += 1

    except Exception, e:
        print "error occured...!!!"
        time.sleep(60)

# print(datapoints[0])