Alert when application traffic drops to zero, but don't alert at quiet times.

[DannyH]

We had an issue where one of our applications had no traffic for 2 hours due to a genuine problem. We got no alert from the anomaly detection we had set up for Dynatrace traffic drops.

Dynatrace support said that was because we had no traffic at all rather than a drop from one figure to another so DT will not alert on 0 traffic.

I was advised to set up a metric event using auto-adaptive setting. I have it set to alert if metric is below and to alert on missing data.

This is the metric selector code used - builtin:apps.web.actionsPerSession:filter(in("dt.entity.application",entitySelector("type(application),entityName.equals(~"Aspire~")")))

Trouble is it is alerting a lot, every time there are 0 user actions in the application. I've messed about with sliding windows and signal fluctuations but still alert spam.

I was hoping and told the auto-adaptive would know not to alert at quiet times. 

Any suggestions how to get this to work? I'm surprised Dynatrace doesn't alert out of the box on zero traffic to an application for a sustained period.

[Akhil-Jayendran]

Hi @DannyH ,

You may want to experiment with the traffic drop setting on the frontend app. but if it is  no traffic at all rather than a drop might not be useful . 

https://community.dynatrace.com/t5/Real-User-Monitoring/How-application-traffic-drops-is-calculated/....

If you have the data in Grail, maybe we could use DQL to run a comparison of the previous 1-hour timeframe and send an alert if there is any drop

[DannyH]

Hi Akhil,

Thanks for the reply. Yeah it was to get an alert when there was no traffic at all that I needed, thats why it has to be a metric event with, "Alert on missing data" turned on.

I've actually got a better metric selector string now based on total count rather than average. It works better, but still alerts at night when the application is quiet. I thought the auto-adaptive threshold would know when the app is quiet from historic data, but it doesn't seem to. Any idea why that is? I've set up a maintenance window anyway for OOH.

This is the metric selector

builtin:apps.web.action.count.load.browser:filter(and(or(in("dt.entity.application_method",entitySelector("type(application_method),entityName.equals(~"loading of page /auth/dashboard~")"))))):splitBy():sort(value(auto,descending)):limit(20)

[g_kat]

Hi @DannyH ,

From experience, auto-adaptive threshold isn't all that smart. It performs a statistical calculation daily for the last 7 days. If there are no data, I believe it does nothing to identify a threshold for that window.

Maybe, if you want to fiddle around, you could set up in your query selector the default(0) transformation. This way, instead of NULL data you will have 0 which can be evaluated for a threshold and then auto-adaptive maybe adjust the threshold for night.

Maintenance windows are a good backup plan too.

George