How to heatmap a timeseries?

[ronmacdonald]

I have timeseries metrics with common meta-data.  The first is the count of "calls", the second is duration of calls.  I'd like to create a heat map with the data to quickly identify distribution of call volume by response time across a full week.  To do this would have x-axis as `day of the week`, y-axis as `range of durations`, and the value the number of calls.

I'm thinking something like:

Starting with this: `timeseries { avg=avg(otel.tyk.duration), calls=sum(otel.tyk.calls) } 
...
Finishing with this: `| summarize calls=sum(calls), by:{range(duration), 100ms, day_of_week}`

Unsure of the middle and how to unwind the 2 timeseries metrics to get the day of week range, that would allow me to finish...

[Michal_Gebacki]

Hello, @ronmacdonald!

There's a documentation page about Heatmap visualization in the Dynatrace Documentation. Can you take a look at this please and let us know whether is it helping your issue in any way?

Thank you for collaboration!

[t_pawlak]

Hi,

I tried to build the heatmap you described and here are the findings.

Using metrics alone (e.g., avg(duration) and sum(calls)) it’s not possible to derive a distribution of calls across duration buckets. Timeseries metrics don’t contain per-call durations, so DQL can’t “unwind” them into individual events.

To calculate the distribution correctly you need to use span-level data, which contains the actual duration of every call.
For example:

fetch spans
| fieldsAdd
    day_of_week     = getDayOfWeek(start_time),
    duration_bucket = range(duration, duration(100, "ms"))
| summarize calls = count(), by: { day_of_week, duration_bucket }
| sort day_of_week asc

This produces a clean distribution table (day of week × duration bucket × calls).
However, Dynatrace heatmap does not support span-based queries, so even with correctly shaped data this cannot be visualized as a heatmap.

I also tried converting the buckets into strings:

fetch spans
| fieldsAdd
    day = formatTimestamp(start_time, format:"EEE"),
    bucket = concat(
        toLong(duration / duration(100,"ms")) * 100,
        "-",
        toLong(duration / duration(100,"ms")) * 100 + 100,
        " ms"
    )
| summarize calls = count(), by: { day, bucket }
| sort day asc, bucket asc

This gives the right structure (string/string/number), but heatmap still rejects it because the heatmap visualization currently works only for metric-based queries, not for span/log/event queries.

You can try this on your data