More sample of Anomaly Detection for PrometheusDB

[wars...@gmail.com]

Hi,

I've been trying to create an anomaly detection based on my count metric. I already have a recording rule that does compute the sum and rate of the count metric.

A sample problem would be: anomaly detection on a daily visit to a specific page. May I know or have a sample on which I can make this kind of detection?

I only found this sample anomaly detection:
https://prometheus.io/blog/2015/06/18/practical-anomaly-detection/

thanks!

[wars...@gmail.com]

Is it possible to use the holt_winters() on this? Is it also okay to use holt_winters() in a recording rule result which has sum(rate())?

Can I also use my recording rule result based on this sample: https://prometheus.io/blog/2015/06/18/practical-anomaly-detection/

If not, any recommendations? thank you in advance.

[Julius Volz]

If you want the kind of anomaly detection where you do not formulate a precise rule that detects the anomaly, but have some kind of machine learning that tells you that "something doesn't look as usual", then no, Prometheus does not support this.

You will need to define what you see as anomalous behavior and formulate a PromQL query expression that catches that condition in an alerting rule. holt_winters() is a function that could help you do this, but it totally depends on what anomaly case you want to detect and what your data is.

--
You received this message because you are subscribed to the Google Groups "Prometheus Developers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to prometheus-devel...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.