Keeping track of your VM

After creating a VM and deploying an application on it, if disk/memory/cpu usage increases drastically then your application may stop responding. In order to handle such kind of situation, monitoring such components is important. Usually, cloud provider provide a monitoring service which can keep track of various resources created on the cloud. In case of AWS, CloudWatch is the monitoring solution.

Here is a list of items that can be tracked for a VM using CloudWatch.

Screenshot from AWS Console showing items that can be monitored for a VM

You can define a threshold such as if CPU utilization is above 50% then send out an email. We need to create an alarm when an event occurs. In this case, the event is CPU utilization exceeding 50% threshold. CPU utilization — the item that we want to monitor is known as a metric.

Here is the specification for the alarm:

Metric = CPU utilization
Event = CPU utilization > 50
Action = Send an email notification

It is possible that VM’s CPU utilization goes up for a minute and then comes back down. If we were to send an email every time event occurs then you can get 100s of emails within few days. This is why a metric we should use average over X amount of time or period (such as 10 second, 1 minute and so on). Using this, we can refine event

Event : average (CPU utilization of 1 minute) is > 50

You can also configure to send a notification if above event occurs 3 times out of 5.

Rule to notify only when an event occurs 3 times in last 5 periods

A metric can be in one of the following states:

• Ok = event has not occurred
• Alarm = event has occurred
• Insufficient data = self explanatory

We are interested to send an email notification when the metric is in alarm state. Let’s say we had 1000s of VMs, 100s of alarms. If we had a program which can send out an email then we need some kind of pipe like mechanism.

• On one end of the pipe, alarms can send or produce necessary data for the email. Here, alarms can be considered a producer.
• On the other, a program would read the data from the pipe and send out an email. This program can be considered a consumer.
• The pipe is known as a topic.

AWS has a service called Simple Notification Service (SNS) where one can create topics. Consumer program needs to subscribe to the topic. This is why it’s called a subscription in SNS which can be created using UI.

Types of notifications supported by SNS

In this post, we looked at

• CloudWatch
• Metric
• Alarm
• Simple Notification Server
• Topic
• Subscription

References

Basic vs Detailed monitoring — using-cloudwatch-new.html

CloudWatch alarm — using-cloudwatch-createalarm.html

States of an alarm — AlarmThatSendsEmail.html

Simple Notification Service — welcome.html