NOTE: This repository is deprecated. If you are searching for a cluster-wide job-specific monitoring solution, check ClusterCockpit.
The LIKWID Monitoring Stack (LMS) is a set of tools and scripts to employ job specific monitoring on a small to midsized HPC cluster. It contains components running on the compute nodes, a tool to route the measurements as middleware and agents for Grafana that automatically create dashboards.
It is thought to allow easy integration without changes to other software or the creation of interfaces to them. There are so many solutions to gather data at the job nodes that the interface to the upper stack should be as simple and globally available as possible, thus the decision was made for HTTP.
The middleware tags the measurements with additional information like jobid or username and forwards the data into a central database. The current version provides only InfluxDB as database (mainly because it supports numbers AND strings as metric values). If configured, the middleware can also duplicate the measurements and send them in other databases (e.g. one per user) so that users can view their own data without being able to see others' data.
The stack top is the web frontend which presents the job data for admins/users/ ... The decision was made for Grafana as is supports many visualization options, is scriptable and has a good HTTP API for management. It would be tedious to create dashboards for each job manually, thus agents provide automatic generation or manipulations of dashboards.
It is never complete! Things are changing, some people require features that are currently not provided by the LMS: New metrics? New node agent? New grafana agent?
Each folder contains a README file about the content in the folder.
All stuff in this repository is GPLv3. This might not fit for third-party components like diamond, InfluxDB or Grafana.
If you have problems, please open an issue For other things, please write me an email: Thomas.Roehl@fau.de