HPI Emergency Shutdown Server

This is a service provided by HPI's Scientific Compute Infrastructure Team, which ensures that servers in the data center can be shut down quickly, reliably and in a coordinated manner in the event of an emergency. Alarms can be triggered both manually and through integration into existing monitoring solutions. In the event of an emergency, affected administrators can be automatically informed of incidents. When clients register, they can provide additional metadata about themselves (location in the data center, server type, department, etc.), which enables granular control and monitoring.

Clients

To avoid damage and hard shutdowns in case of emergency, we strongly recommend that the “HPI Emergency Shutdown Client” is installed on all servers in the data center. This client communicates with the emergency shutdown server and calls up the hooks you want to be called in an emergency.

Instructions for installing and using the client, as well as the source code, can be viewed here:
https://gitlab.hpi.de/sci/emergency-shutdown/emergency-shutdown-client

Please reach out to us if you have any questions or need help with the installation. Also let us know if you start using the client, so we can reach out to you in case of updates or maintenance.

Status Types

The current status is communicated via a traffic light system. A client can receive the following statuses:

• Green: Operations may be resumed.
• Yellow: Limit resource usage. All operations that require significant power must be immediately ceased, e.g., killing HPC jobs. This stage is optional and should only be used if it makes sense in this use case.
• Red: The host will shut down in 60 seconds. Prepare everything accordingly; if necessary, initiate self-shutdown.
• Black: Shutdown command is executed immediately by the daemon.

Contact

This service is provided by the Scientific Compute Infrastructure Team. If you have any problems, suggestions for improvement, or questions - feel free to leave us an e-mail at sci.helpdesk@hpi.de.

Architecture

Abuse

Please note that all access to this system is logged. Since the service is only available within the HPI network - where you have authenticated yourself - every visit can be traced back to you as a person. Misuse of this service can be punished with exmatriculation.