Understanding Evaluation Periods and Datapoints to Alarm

Question

I kinda struggle understanding what happens in our AWS CloudWatch alarms setup.

We're watching the number of available nodes in ElasticSearch. The goal is to raise an alarm when available nodes is less than three for more than 2 hours.

Here's the alarm definition:

But still the alarm is triggered even when the available nodes number drops down for just a while:

When period is set to 1 hour the graph looks as follows:

According to the AWS documentation I'd expect the alarm to be triggered when the available nodes is <= 2 for two periods (2 hours).

Can anyone explain what is wrong here?

What does the graph look like if you set the period to 1 hour, are you using min, max, avg for the alarm? — Chris Williams, Aug 26 '20 at 15:51
I'm using minimum (can be seen in first image under "Statistic"). Updated the question to contain the graph when period is set to 1 hour. — Ikar Pohorský, Aug 26 '20 at 15:59

score 2 · Answer 1 · edited Aug 11 '21 at 13:07

2

I think the problem here is that the period is 1 hour. If it goes down to two nodes within one hour it will go on the alarm state.

If you want to evaluate every two hours, you can try to set your evaluation periods to 2 with period of 1 hour or to 1 with a period of 2 hours.

edited Aug 11 '21 at 13:07

General Grievance

4,555
31
31
45

answered Aug 10 '21 at 18:11

KPAP

21
3

Understanding Evaluation Periods and Datapoints to Alarm

1 Answers1