Hello everyone,
I'm currently working on a use case where we need to evaluate the number of users impacted by different problems affecting different applications of a client.
To achieve this, I’m using the metric builtin:apps.web.activeUsersEst to show the number of users actively using the application at any given moment. (Session capture is configured at 100%.) According to the Dynatrace documentation, this metric is calculated using the HyperLogLog algorithm (Wikipedia). Based on the original paper (Flajolet et al., 2007) and other sources, HyperLogLog provides a relative error of about 2% for cardinalities greater than 10⁹.
In parallel, I’m also using the dt.davis.affected_users_count field from the dt.davis.problems bucket, which I ingest as a metric filtered by the affected application.
Regarding its calculation, Dynatrace offers the following brief description:
Business impact analysis shows the number of potentially affected real users (collected from the incoming traces), the number of traces, and their service endpoints.
Now, here’s where the issue arises: when comparing the two charts, the value for dt.davis.affected_users_count is significantly higher than the value for builtin:apps.web.activeUsersEst.
This naturally raises a question from the client:
How is it possible that both metrics are estimations of user count, yet the number of affected users is so much higher than the number of active users for the application at the same point in time?
For the active users metric, I can support the estimation method based on HyperLogLog and its documented accuracy. However, I lack further technical details to explain or justify the behavior of the affected users count.
Given this context, I’d like to know:
Has anyone faced a similar situation?
Is there any Dynatracer who can provide more insight into how affected_users_count is calculated (e.g., what algorithm or heuristics are involved), so I can provide a clearer explanation to the client?
I’m attaching an example comparison where the red line represents affected_users_count and the purple one shows builtin:apps.web.activeUsersEst.
Any help or clarification would be greatly appreciated.
Featured Posts
