CarbonCopy Anchor Link

Carbon attributes energy, emissions, water, and cost to the workloads that consume them. It lives at /observability/carbon, with breakdowns by node, namespace, host, and job, plus surfaces for budgets, recommendations, regressions, and a telemetry diet.

What it isCopy Anchor Link

Carbon answers a question observability usually ignores: what does this system cost the planet, and which of my workloads is responsible? Owlpane Carbon follows the Green Software Foundation Software Carbon Intensity (SCI) specification (ISO/IEC 21031:2024), whose rate form is:

SCI = (E × I + M) / R

where E is energy consumed, I is the grid carbon intensity, M is embodied (hardware manufacturing and amortised) emissions, and R is the functional unit of the software. Carbon reports each term so the result is auditable rather than a single opaque score:

• Energy in kWh.
• Operational CO₂e.
• Embodied CO₂e.
• Water.
• Cost, at your own blended vCPU-hour rate.

Attribution starts from kubeletstats node and pod metrics, so the same fleet that powers Infrastructure also powers Carbon.

Key capabilitiesCopy Anchor Link

• Overview — fleet energy, operational and embodied CO₂e, water, and cost for the window, with a comparison against the prior period and a per-figure uncertainty band.
• Scope breakdowns — attribute down to Nodes, Namespaces, Hosts, and Jobs.
• Services and waste — workloads ranked by attributed energy, with a waste finder that shows the static energy of CPU requested but not used, per workload owner.
• Endpoints — grams of CO₂e per million requests, using the unsampled request counter where it exists and duration-weighting otherwise.
• GPU — emissions and energy for GPU workloads.
• Budget and regressions — set a budget and catch a workload whose emissions regressed.
• Recommendations and diet — suggestions to reduce emissions, including trimming the telemetry you do not need.
• Factors and compare — the grid factors behind the numbers, and a comparison across zones.
• Billing — cost of the attributed compute alongside the emissions.

Where it livesCopy Anchor Link

Plans and query windowCopy Anchor Link

Carbon is available on every plan. The query window is 7 days on Starter and 30 days on Team and above. Carbon attribution and Carbon report export are available on all plans. See Plan and billing for the workspace plan details.

How data gets inCopy Anchor Link

Carbon is derived, not ingested directly. It attributes from the node and pod metrics your cluster already reports, so it starts to populate once kubeletstats node and pod metrics are flowing to ingest.owlpane.com — install the Kubernetes agent with hostmetrics and resourcedetection enabled. Endpoint-level attribution additionally needs server spans from an Owlpane SDK.

• Kubernetes Helm agent — install the agent that supplies node and pod metrics.
• Metric ingestion — the series Carbon reads.
• Distributed trace ingestion — server spans for per-endpoint attribution.
• Integration library — NVIDIA GPU and hostmetric collectors.
• OTLP ingest reference — the forwarding endpoints.

RelatedCopy Anchor Link

• Infrastructure — the fleet whose resources are attributed.
• Kubernetes — nodes and namespaces behind the breakdowns.
• Dashboards — put Carbon panels beside cost and health.
• Metrics — the metric primitives underneath.
• Features index — the full surface map, including Beta surfaces.