Prometheus
Prometheus
- GPU Reaper
built
Find wasted GPU allocations on a Slurm cluster and, only if enabled, drain and cancel them, with guardrails that refuse to act on missing, stale or unreadable telemetry.
- IB Slurm Exporter
built
A Prometheus exporter that attributes InfiniBand and RoCE counters to the Slurm job using the HCA, refuses to attribute a port it can see is shared, and handles Slurm 26.05's SLUID cgroup layout.
- Planning a Slurm 25.11 Upgrade? 26.05 Has Already Shipped.
Three versions, two renamed config options that now log deprecation warnings, and one cgroup change that quietly breaks job-attribution tools. Corrected September 2026 against Slurm's source.