Is it possible to detect bad instances using datadog outlier detection?

Question

I would like to detect bad/faulty aws instances using datadog's outlier detection. Is that possible? I'm trying to create an automatic failover scenario using datadog. Any suggestions would be appreciated.

Answer 1

You could use Datadog's Outlier detection to identify instances which exhibit behavior outside the normal for it's peer set. As an example, you could create an outlier detection monitor:

http://docs.datadoghq.com/guides/outliers/#alerts

Which would be scoped to a system metric like aws.ec2.cpuutilization and be alerted if any host spiked abnormally or had very low utilization in comparison to its group.

There are some additional blog posts which discuss the use of the algorithms that can be found here:

https://www.datadoghq.com/blog/introducing-outlier-detection-in-datadog/

https://www.datadoghq.com/blog/outlier-detection-algorithms-at-datadog/

https://www.datadoghq.com/blog/scaling-outlier-algorithms/

That said, if you find yourself needing additional assistance with outlier detection you can always reach out to the Support team at support@datadoghq.com or by using the internal support features found here:

https://app.datadoghq.com/help

Hope this helps!