Create status check and 90% CPU alert on all EC2 machines in AWS

create_status_check_90_cpu_ec2.py

#!/usr/bin/env python

# find REPLACE_WITH_ARN and replace with real SNS ARN

import boto3
from dateutil.parser import parse
import datetime
aws_profile_name = 'prod'

def get_instance_name(fid):
    instancename = ''
    for tags in fid['Tags']:
        if tags["Key"] == 'Name':
            instancename = tags["Value"]
    return instancename

def delete_cloudwatch_alarm(instance_id, instance_name):
    cloudwatch_conn = boto3.client('cloudwatch')
    status_name = instance_id + '-' + instance_name + "-status-alarm"
    cpu_name = instance_id + '-' + instance_name + "-90%-CPU_Utilization"
    cloudwatch_conn.delete_alarms(
        AlarmNames=[
            status_name, cpu_name
        ]
    )
    print "Deleting - " + status_name
    print "Deleting - " + cpu_name

def create_status_alarm(instance_id, instance_name):
    cloudwatch_conn = boto3.client('cloudwatch')
    cloudwatch_conn.put_metric_alarm(
        AlarmName = instance_id + '-' + instance_name + "-status-alarm",
        ComparisonOperator='GreaterThanOrEqualToThreshold',
        EvaluationPeriods=2,
        MetricName='StatusCheckFailed',
        Namespace='AWS/EC2',
        Period=60,
        Statistic='Maximum',
        Threshold=1.0,
        AlarmDescription='Status check for ' + instance_id + '-' + instance_name,
        OKActions = ['REPLACE_WITH_ARN'],
        AlarmActions = ['REPLACE_WITH_ARN'],
        Dimensions=[
            {
              'Name': 'InstanceId',
              'Value': instance_id
            },
        ],
    )
    print "Creating - " + instance_id + '-' + instance_name + "-status-alarm"

    cloudwatch_conn.put_metric_alarm(
        AlarmName = instance_id + '-' + instance_name + "-90%-CPU_Utilization",
        ComparisonOperator='GreaterThanOrEqualToThreshold',
        EvaluationPeriods=1,
        MetricName='CPUUtilization',
        Namespace='AWS/EC2',
        Period=300,
        Statistic='Maximum',
        Threshold=90.0,
        OKActions = ['arn:aws:sns:ap-southeast-1:928401551325:AllowSNSToSlackExecutionFromSNS'],
        AlarmActions = ['arn:aws:sns:ap-southeast-1:928401551325:AllowSNSToSlackExecutionFromSNS'],
        AlarmDescription='Alarm when server CPU exceeds 90% for ' + instance_id + '-' + instance_name,
        Dimensions=[
            {
              'Name': 'InstanceId',
              'Value': instance_id
            },
        ],
        Unit='Percent'
    )
    print "Creating - " + instance_id + '-' + instance_name + "-90%-CPU_Utilization"


boto3.setup_default_session(profile_name = aws_profile_name)
ec2 = boto3.client('ec2')
instances = ec2.describe_instances()

for reservation in instances["Reservations"]:
    for instance in reservation["Instances"]:
        instance_name = get_instance_name(instance)
        if(instance["State"]["Name"] == "running"):
            create_status_alarm(instance["InstanceId"], instance_name)
        else:
            delete_cloudwatch_alarm(instance["InstanceId"], instance_name)
    # break