Alerting
Questions about alerting and problem detection in Dynatrace.
cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

Delta between Dashboard UI Davis and Workflow Davis Timeframes

Rudolph_Sedlin
Participant

Hey all,

We issue alerts via scheduled Workflows that run Davis Anomaly Detection, scanning recent data against the training data and issuing alerts subsequently. We prototype these workflows visually via simulated alerts in Dashboards, however those pin the training and scanning data timeframes for Davis to match the viewing timeframe in the dashboard itself, preventing accuracy (and for that matter causing increased variance and ahistoricity for more granular viewing intervals, reminds me a bit of Heisenberg Uncertainty of all things, and even occurring for somewhat similar statistical reasons). I wonder if there is a way to overcome or sidestep the issue to make the simulated alerts in Dashboards more faithful to what Davis would see when invoked from a specialized Workflow, especially since in the most extreme scenarios slightly changing the viewing timeframe in Dashboards can cause simulated alerts to vanish due to these inconsistencies, even if both viewing timeframes comfortably envelop the simulated alert lifecycle.

1 REPLY 1

t_pawlak
Leader

Hi,
What you’re hitting is a context mismatch: Dashboard “simulation” is anchored to the currently viewed timeframe, while Davis Anomaly Detection executed by a Workflow runs on predefined sliding/evaluation windows plus a reference (baseline) period and controls what data it fetches. That’s why small timeframe changes in Dashboards can make simulated alerts disappear, even if the Workflow execution would still detect the anomaly.

Ways to make Dashboard prototyping closer to Workflow reality:

Don’t “manually” control timeframe inside DQL for Davis AD
Dynatrace explicitly recommends: avoid from:/to: and don’t override the timeframe field in DQL used for anomaly detection, because the detector pulls data according to its own sliding windows—overrides lead to mismatched data/failures
davis-ad-dql-best-practice 

Or you can use tile-specific timeframe (instead of global timeframe/zoom) matching your Workflow scan window. Set the tile to the exact window your Workflow evaluates (e.g., last 30m / last 2h) and avoid zooming during validation

I recomended use Anomally Detection app if you need check how this it will run in production.
Anomaly Detection app 

Featured Posts