sustainability-measurement-agent 0.1.9

Sustainability Measurement Agent - A tool for collecting and analyzing sustainability metrics in cloud native enviroments.

Sustainability Measurements Agent (SMA)

|

SMA is an open-source tool designed to help deploy, collect and report sustainability measurements on cloud-native applications. It is not itself a measurement tool, but rather a framework to orchestrate, combine and aggregate results from emerging suite of sustainability measurement tools.

Capabilities (planned)

• Configuration driven deployment and operation
• Support for multiple measurement tools
  • Kepler
  • Scraphandre
  • cAdvisor
  • KubeWatt
  • Cloud Provider APIs (AWS, GCP, Azure)
  • KubeMetrics
• Support for multiple scenarios
  • Continuous monitoring and reporting
  • Experiment Measurements / Benchmarking
  • Ad-hoc measurements
  • Programmatic measurements via API
  • Kubernetes Operator
• Multiple report formats
  • CSV
  • HDF5
• Post processing and aggregation
  • Pandas
  • Grafana Dashboards

Use / Install

Make sure you have Python 3.8+ as well as hdf5 and jupyter installed. You may need to install additional dependencies depending on your setup, e.g., c-blosc lzo bzip2 gcc.

pip install sustainability-measurement-agent

touch main.py

import sys
from sma import SustainabilityMeasurementAgent, Config, SMAObserver

config = Config.from_file("examples/minimal.yaml")
sma = SustainabilityMeasurementAgent(config)
sma.connect()
def wait_for_user() -> None:
    input("Hit Enter to to Stop Measuring...")

sma.run(wait_for_user)
sma.teardown()
report_location = config.report.get("location")
if report_location:
    print(f"Report written to {report_location}")

$ python main.py
$ Hit Enter to to Stop Measuring...

Architecture and Lifecycle

SMA follows a simple lifecycle:

Fully driven by configuration files, SMA connects to external services (e.g., Prometheus), deploys measurement tools (e.g., Kepler), and once instrcuted via the API (run()), starts the observation according to the configured mode (trigger, timer, continuous). Finally, SMA collects the measurement data, generates reports, and tears down any deployed infrastructure (if configured to do so).

SMA exposes an observer API (SMAObserver), which allows to hook into the lifecycle events (see the diagram) to extend SMA's functionality programmatically, or to integrate into experiment frameworks.

Configuration

SMA is configured via YAML files. See the examples/ folder for sample configurations. The configuration allows to specify which measurement tools to use, how to deploy them, and how to collect and report the measurements.

Configuration Structure

The configuration file consists of the following main sections:

1. Version

sma:
  version: 0.0.1

Specifies the SMA configuration schema version.

2. Deployment (Optional)

deployment:
  namespace: sma
  install: true

Controls automatic deployment of measurement infrastructure.

Field	Type	Description
namespace	String	Kubernetes namespace where SMA components will be deployed
install	Boolean	Whether to automatically install/deploy measurement tools

3. Services

services:
  prometheus:
    address: http://130.149.158.132:32426
    # Optional: Deploy Prometheus via Helm
    chart: prometheus-community/prometheus
    version: 14.6.0
    values:
      - service.type: NodePort
        service.nodePort: 30900

Defines external services used for data collection, sofar we only support Prometheus.

Field	Description
prometheus.address	URL of the Prometheus server for metrics collection
prometheus.chart	(Optional) Helm chart repository and name for deploying Prometheus
prometheus.version	(Optional) Helm chart version to deploy
prometheus.values	(Optional) List of Helm values to override default configuration

4. Sensors (Optional)

sensors:
  kepler:
    chart: kepler/kepler
    version: 0.8.1
    values:
      - global:
          prometheus:
            enabled: true

Defines measurement sensors/tools to deploy via Helm charts.

Field	Description
<sensor_name>.chart	Helm chart repository and name for the sensor
<sensor_name>.version	Helm chart version to deploy
<sensor_name>.values	List of Helm values to override default configuration

Supported Sensors:

• Kepler: Kubernetes-based Efficient Power Level Exporter
• Scaphandre: Host-level power consumption metrics
• cAdvisor: Container resource usage and performance metrics
• Kube Metrics: Kubernetes metrics server

5. Observation

observation:
  mode: trigger 
  window:
    left: 10s   
    right: 10s 
    duration: 10s
  targets:
    - name: sut
      namespace: kubeai

Field	Options/Type	Description
mode	trigger, timer, continuous	trigger: Measurement starts/stops based on a Python function that blocks timer: Measurement based on time window settings continuous: Collects data until stopped, with a blocking Python function
window.left	Duration (e.g., 10s)	Time to capture before the measurement trigger
window.right	Duration (e.g., 10s)	Time to capture after the measurement trigger
window.duration	Duration (e.g., 10s)	Total measurement duration
targets[].name	String	Name identifier for the target workload
targets[].namespace	String	Kubernetes namespace of the target workload

Depoending on the mode, SMA will not use of all of the windowing parameters, e.g., duration is only considered in timer mode. Both left and right windows are optional and mean, SMA will wait before handing off the measurement to the next API call.

6. Measurements

measurements:
  # Custom PromQL query
  - measurement_name:
      type: aggregate
      layer: pod         
      query: <PromQL query>
      step: 30           
      unit: watts       
      target:
        - all        
  
  # Prepared query (predefined)
  - pod-consumption:
      type: aggregate
      layer: pod
      prepared_query: pod_kepler_energy_consumption
      step: 30
      unit: watts
      sensors:
        - kepler
      target:
        - sut

Field	Options/Type	Description
type	raw, aggregate	raw: Unprocessed metric data aggregate: Aggregated/computed metrics
layer	substrate,pod, node, process	The abstraction level being measured
query	PromQL string	Prometheus query to retrieve the metric. Supports template variables: ${namespace}: Replaced with target namespace ${SMA_SELECTORS}: Additional label filters
prepared_query	String	Use a predefined query from SMA for common metrics (alternative to query)
sensors	List of strings	List of sensors required for this measurement (used with prepared_query)
step	Integer (seconds)	Sampling interval in seconds
unit	String	Measurement unit (e.g., watts, joules, cpu_time_seconds, metadata)
target	List of strings	Target names to measure, or all for all defined targets

Query Methods:

• Custom Query: Use query field with custom PromQL
• Prepared Query: Use prepared_query field with predefined SMA queries (e.g., pod_kepler_energy_consumption, node_scaphandre_energy_consumption)

Example Measurements:

• CPU Usage: Container CPU usage across namespaces
• Wall Power: Physical power consumption from smart plugs (e.g., Shelly)
• Pod Power (Kepler): Per-pod energy consumption via Kepler
• Node Power (Scaphandre): Host-level power measurements
• Container Info: Metadata about running containers

7. Report

report:
  format: csv          # Output format
  location: reports/${startTime}_${runHash}/
  filename: "${name}_${startTime}.csv"
  post_processor:
    - name: overview
      type: notebook
      notebook: reports/overview.ipynb
    - name: validation
      type: sma
      module: sma.validation.Validator

Field	Type	Description
format	String	Output format (currently supports csv)
location	String	Directory path for reports. Supports template variables: ${startTime}: Timestamp when measurement started ${runHash}: Unique hash for the measurement run
filename	String	File naming pattern. Supports template variables: ${name}: Measurement name ${startTime}: Measurement start time
post_processor	List	(Optional) Post-processing steps to run after data collection
post_processor[].name	String	Identifier for the post-processor
post_processor[].type	notebook, sma	Type of post-processor: notebook for Jupyter notebooks, sma for built-in SMA modules
post_processor[].notebook	String	Path to Jupyter notebook (when type: notebook)
post_processor[].module	String	Python module path (when type: sma)

API

Development

1. Install pipenv: pip install pipenv
2. Install dependencies: pipenv install --dev
3. Activate virtual environment: pipenv shell
4. Do your thing. Eventually with tests.
5. If you add new dependencies, run pipenv lock --pre to update the Pipfile.lock.
6. If you want to release a new version, update the version in pyproject.toml and run pipenv run pyproject-pipenv --fix and commit the changes.
7. Tag your release: git tag vx.y.z and git push --tags
8. Create a new release on GitHub, this will trigger the upload to PyPI via GitHub Actions.

Acknowledgements

This project builds upon prior work done in the OXN project, GOXN project and CLUE and is part of the research at ISE-TU Berlin.