Qlustar Feature Update 07/26

Qlustar Feature Update 07/26

A large bunch of new features were completed and published since the last Qlustar feature release 11/25. Here is an overview.

QluMan

QluMan versions up to 14.1.5.x have added the following:

Dynamic Booting

Adding the possibilty to dynamically boot cluster nodes in the local network adds many new exciting possibilities, e.g.:

• Quickly diagnose or stress-test new hardware before final integration into the cluster.
• Convert nodes having booted dynamically into real cluster nodes with just a few mouse clicks.
• Do fast and simple pre-qualifying of test hardware without polluting the production cluster setup.

Node/MAC detection from Network Switches

Adding and configuring Network Switches is one of these Enterprise features that many admins of large clusters will highly appreciate. It makes flawless and precise mass integration of hundreds of nodes possible at an incredible pace by allowing to:

• Add new hosts via consultation of the switches MAC tables.
• Clearly display used switch ports as a property of each host.
• Automatically filter out IPMI and real host MAC addresses.
• Test-boot and easily analyze the properties of new nodes.

Prometheus Alerts

The creation of Prometheus Alerts was highly faciliated by:

• Adding a new warning/critical severity alert type that covers alerts where the same expression can be used with different severity levels depending on the value of the expression. This significantly reduces the number of alerts to be configured, e.g. with CPU temperatures where it is usually desirable to have warning and critical values.
• Improved parsing and converting metric values in the alert expression editor.
• Filtering out linebreaks and repeated spaces in expressions.
• Improved silencing of alerts:
  • Add ErrorDialog when fetching silences fails.
  • Reset the input mask on save.
  • Reset color for the Starts label when selection changes.

Miscellaneous enhancements

• Make the QluMan GUI work flawlessly under Wayland desktops.
• Misc fixes for Redfish/BIOS management to reliably allow the check and auto-correction of BIOS settings.
• Added new node types ‘OTHER’ (for non-Qlustar machines) and ‘VIRTUAL’ (for HA virtual IPs) to provide special integration functionality e.g. for monitoring external nodes via prometheus.
• Improvements of the BootConfig dialog:
  • Added Clone/Edit/Delete functionality/buttons.
  • Added Clone context menu.

Revamped Grafana dashboards and metrics collection improvements

• The Qlustar Grafana Home dashboard now shows:
  • Nice overviews of Slurm and Kubernetes stats.
  • More detailed summary stats for the different type of node groups in a cluster.
• The Slurm dashboard has been fully redesigned with:
  • New global and per partition overview pie charts displaying node states and core/memory allocation.
  • More concise cluster node state graphs both cluster global and per partition.
  • New job data graphs with many details again for the whole cluster and per partition.

Lustre 2.17

• Lustre 2.17 server and client support has been fully validatet and was added to Qlustar with ready to use packages and image modules. It adds many new great features as compared to the previously provided 2.15, e.g.
  • DNE3: Metadata overstriping
  • DNE3: directory migration in non-recursive mode
  • New lljobstats command
  • Dynamically configurable nodemap
  • Support for multiple filesets per nodemap
  • Add permanent Token Bucket Filter rules
  • Hundreds of bug fixes and small improvements

All Q-Leap managed Lustre systems are now running Lustre 2.17!