Q: How to read in traces generated by Caliper? about hatchet HOT 4 CLOSED

Hi @cielling, At this time, Hatchet's caliper reader only supports loading caliper profiles, not traces. To generate profiles for Hatchet, you can use caliper's built-in profiling configurations hatchet-region-profile and hatchet-sample-profile. The error you're seeing when loading in a trace is because Hatchet's reader is looking for a path or source.function#callpath.address column. We're also working on a docs page to provide more details into how to generate caliper, HPCToolkit, etc. datasets for Hatchet. Hope this helps, please let us know if we can answer additional questions about generating caliper data for hatchet.

from hatchet.

Hi @slabasan, thanks for the reply! This is good to know, although not quite the answer I was hoping for :) I'm looking forward to that docs page, so far the things I've tried that were documented (e.g. the hatchet-region-profile) work quite nicely!
I'm going to see if I can find some other way to analyze my trace data, there is probably some tool out there that can help me.
Thanks!

from hatchet.

Hi @cielling. Would you be able to send me a copy of the trace data you are trying to analyze? Thanks!

from hatchet.

Hi @cielling ,

It's a bit late, but for reference, to convert a Caliper trace into a profile that Hatchet can read you'll have to aggregate it over the region hierarchy, like so:

cali-query -q "select *,sum(time.duration) as time group by prop:nested format json-split" <file(s).cali>

The sum function aggregates time, and group by prop:nested groups by any Caliper region attribute ("annotation", "function", and so on). That'll produce the "path" column Hatchet is looking for.

from hatchet.