Environment

Tracarbon version: v0.7.2

Environment:

• Source: pypi
• OS: Linux (NixOS)
• Other:

Bug

What happened: The server I tested on has multiple physical CPUs. The resulting energy usage is completely wrong.

What you expected to happen: The energy usage is correct.

How to reproduce it: Run the Tracarbon on a server with multiple physical CPUs

More details:
The problem seems to be that multipe CPUs means that there are multiple RAPL directories. This is handled mostly correctly, but there is the issue that there are multiple dram or core sensors, which is a problem when storing previous results in a dictionary in the RAPL implementation. There is no distinction between core and dram or CPU 1 or CPU 2.
This then causes problems with the calculation of differences, since the old CPU 2 value could be used for the new CPU 1 value or vice versa.