How does the estimation for GitLab work?

[Johardt]

Hello,
This is a really helpful project. I am currently in the process of experimenting with different approaches to estimating CO2 emissions of GitLab pipelines, and this tool seems like a very easy drop-in solution.

However, from the documentation I've read so far, Eco-CI uses precalculated power curves from the cloud energy project.
As far as I understand, in the case of GitHub it uses precalculated power curves from Azure.
But the story is different for GitLab, which can be hosted privately. The GitLab instances I work with are hosted on a third-party cloud provider, with very different hardware setups than Azure.
Thus, I am wondering how the Eco-CI tool measures the total system power in the case of a privately hosted GitLab instance? And also how it calculates the embodied carbon?

[ArneTR]

Thanks for opening a discussion on this!

When using custom runners (or simply "your box") you have to create power profiles yourself and also plug in the embodied carbon yourself (in this case through a PR atm ... )

In the readme we have some samples how to do that for power profiles. See: https://github.com/green-coding-solutions/eco-ci-energy-estimation?tab=readme-ov-file#support-for-dedicated-runners--non-standard-machines

The embodied carbon as said is currently not so good in terms of modularity. We do that for the already supported machines here:

eco-ci-energy-estimation/scripts/vars.sh

Line 46 in c4f6d0a

if [[ "$machine_power_data" == "github_EPYC_7763_4_CPU_shared.sh" ]]; then

All data comes from DataVizta bc we need to use averages as we do not know the vendor of many components. However if you know your vendor specifically also proper PCF (Product Carbon Footprints) from the vendors can be used.

To summarize: If you want to configure that for your local box we are happy to support on that and in the process create a PR that makes plugging in custom embodied carbon more modular :)