Calculating processing time percentiles for a custom service metric

[swin12]

Hi,

I have the following criteria that I am looking to visualize in a dashboard:

Metric: builtin:service.totalProcessingTime

Aggregation(s): Percentile (50, 75, 90, 95, 99)

Split by Dimension: {RequestAttribute:flow}

Idea being that I have a number of requests hitting my service that differ by the value of the request attribute 'flow'. One request might have 'flow = foo' and another request might have 'flow = bar'. I want to know, for example:

"For requests hitting my service that have a 'flow' request attribute value of 'foo', what is the 99th percentile of processing time for those requests?"

Given the "builtin:service.totalProcessingTime" metric does not allow me to split/filter the data by the value of a request attribute, I was instructed in another thread to create a calculated service metric (which could be split by the {RequestAttribute:flow} dimension).

I can do this, however the percentile aggregation is not supported by calculated service metrics.

I am brainstorming some ideas as to what other aggregations I could use (to then be filtered further in Data Explorer), but not getting any great ideas.

Is what I'm asking for possible?

[zietho · Accepted solution]

Hey, 

1. yes, calc metrics would be the way to go to filter by request attribute. 

2. yes, you're also right with calc. metrics not to support percentiles, sadly

3. I am, however, not an expert in the realm of services, and I would not know of any workaround to this 

Maybe @tom_rothschaedl @balint_kelen can chime in here?