A solution which scales both at the container and the IaaS levels for providing true compute elasticity, on-demand. Ideal for deploying workloads with variable footprint. Comes with an example monitoring stack.
- Realized using Azure VM Scale Sets
- Linux Diagnostics extension used for getting the guest VM metrics used for triggering scaling in/out based on CPU and Memory use.
- CPU bound tested with stress tool encapsulated as a docker image
- Azure deployment jsons created with acs-engine chosing DockerCE (Swarm Mode) orchestrator.
- Grafana dashboard json in /grafana directory.
- Docker experimental mode enabled before installing monitoring stack.
- All scripts in /scripts directory.
- cputest (cputest.sh) is itself deployed as a swarm mode service in the global mode.
- clean_swarm.sh utility for cleaning up "Down" nodes from swarm master after cluster scales in.
- Additional container allocation visualization tool can be deployed using setup_visualizer.sh script. This needs to be run on the master on the "local" docker daemon bound to the docker0 eth interface.
- deploy_monitoring.sh for deploying the monitoring stack consisting of the following components:
- cAdvisor for container metrics,
- node-exporter voor VM metrics,
- Prometheus for making the time series data stream,
- Grafana dashboard for displaying the metrics.
Architecture diagram is as follows:
Sample dashboard looks like following when cputest is running, demonstrating scaling out of the smarm mode cluster automatically and scaling in when cputest is stopped:
This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments. All credit goes to contributors of the individual components used in this project, where applicable.