carbon-ops

carbon-ops provides tooling for measuring host energy use, estimating carbon emissions, and writing canonical ledger records. The repository contains the Python library, a privileged polling daemon, and helper scripts for audit workflows.

Components

• Library: energy logging helpers, carbon-estimation routines, and ledger utilities that emit canonical JSON lines with hash chaining.
• Governance daemon: polls RAPL counters at 10 Hz, maintains a monotonic accumulator, and serves readings over a Unix domain socket so unprivileged processes have access to wrap-safe totals.
• Examples: reference scripts that show how to assemble the pieces, including the aliasing demo that compares the governance daemon to a 15 s Prometheus-style scraper.
• Documentation: Markdown files and notes that cover deployment, limitations, and planned work.

Use cases

• Collect host telemetry (CPU, GPU, memory, RAPL) for training or inference jobs and store the results as time-stamped metrics.
• Estimate carbon emissions for a completed span using location-aware grid intensity data. The estimator records the method (analytical_truncated or monte_carlo) and confidence interval metadata in the output.
• Append audit ledger entries where each JSON object includes the hash of the previous entry, enabling tamper detection.
• Attribute package-level energy to individual processes by combining governance-daemon readings with per-process CPU time (allocation_ratio).

Installation

The project targets Python 3.10 and later on Linux hosts. RAPL-based telemetry requires access to /sys/class/powercap or /dev/cpu/*/msr (kernel modules intel_rapl_common and msr). Install with all extras to enable testing and optional providers.

python -m pip install carbon-ops[all,dev]

Local development clones can use the supplied virtual environment helpers in scripts/ or standard tooling such as pip and venv.

Quick start

from carbon_ops import CarbonEstimator, EnergyLogger

try:
  logger = EnergyLogger()
except Exception as exc:
  raise SystemExit(f"EnergyLogger initialisation failed: {exc}")

metric = logger.log_metrics("training_step")

estimator = CarbonEstimator()
record = estimator.estimate_over_span(
  start_ts=metric["timestamp_start"],
  end_ts=metric["timestamp_end"],
  energy_wh=metric["energy"]["energy_wh_total"],
)
print(record.to_dict())

If the governance daemon is not available, the energy summary will report attribution_mode="monitor_only" and execution continues without raising exceptions.

Telemetry modes

Two RAPL backends are available:

• Sysfs mode (default): reads /sys/class/powercap/.../energy_uj files. All samples are masked to 32 bits before they are accumulated to prevent high-bit noise.
• MSR mode (privileged): reads /dev/cpu/<n>/msr directly. The governance daemon fetches MSR 0x606 (RAPL power unit register), applies a 32-bit mask to RAPL counters (for example MSR 0x611), converts to microjoules, and maintains the same accumulator used by the sysfs path. Enable with --rapl-mode=msr [--msr-cpus=…]. Root or CAP_SYS_RAWIO and the msr kernel module are required.

Governance daemon

Start the daemon with the default sysfs reader:

sudo python -m carbon_ops.governor.daemon

A minimal MSR launch targeting the first two sockets:

sudo python -m carbon_ops.governor.daemon --rapl-mode=msr --msr-cpus=0,1

The daemon listens on /var/run/carbon-ops.sock with root:carbon-users ownership and 0o660 permissions. The library degrades to monitor-only mode if the socket is unavailable.

Aliasing demonstration

examples/aliasing_demo.py contrasts the 10 Hz governance-daemon output with a mock Prometheus scraper that polls every 15 seconds. The script exercises a CPU busy loop, records energy via the Unix socket, and reports the energy that the slow scraper misses when RAPL wraps (the “11-second anomaly”).

sudo python examples/aliasing_demo.py --duration 120

Ledger output

Ledger entries are written as canonical JSON lines. Each record includes the previous entry hash (prev_hash), producing a simple hash chain for tamper resistance. The ledger functions guarantee atomic writes using a temporary file followed by os.replace.

Testing

Run the test matrix after installing the development dependencies:

python -m pytest
python -m ruff check .
python -m black --check .
python -m mypy --strict .
python -m bandit -r src

The pyproject.toml file contains the exact tool versions used in CI.

Contributing

See CONTRIBUTING.md.

License

Licensed under the terms of LICENSE.

Support

• Issues: https://github.com/scrrlt/carbon-ops/issues
• Discussions: https://github.com/scrrlt/carbon-ops/discussions
• Email: s@scrrlt.dev