Version
1.1.0
Which installation method(s) does this occur on?
Helm / Kubernetes manifests
Describe the bug.
Setting vendor.nvidia.dcgm.enabled: false (the default) does not prevent the DCGM prolog/epilog scripts from being added to prologScriptRefs/epilogScriptRefs in the Controller CR. As a result, the Slinky operator always injects Prolog=prolog-90-dcgm.sh and PrologFlags=Contain into slurm.conf, even on CPU-only clusters where DCGM is not intended to be used. This prevents workers from starting successfully on some CPU environments (minikube on macOS here).
Minimum reproducible example
# Render the chart with dcgm explicitly disabled (the default)
helm template my-cluster oci://ghcr.io/slinkyproject/charts/slurm \
--version 1.1.0 \
--set vendor.nvidia.dcgm.enabled=false \
| grep -E "prologScriptRefs|epilogScriptRefs|dcgm"
# Observed output — DCGM refs are present despite enabled=false:
# prologScriptRefs:
# - name: my-cluster-slurm-prolog-dcgm
# epilogScriptRefs:
# - name: my-cluster-slurm-epilog-dcgm
# Also confirm with enabled= (empty string, the only working workaround):
helm template my-cluster oci://ghcr.io/slinkyproject/charts/slurm \
--version 1.1.0 \
--set "vendor.nvidia.dcgm.enabled=" \
| grep -E "prologScriptRefs|epilogScriptRefs|dcgm"
# With empty string the refs are correctly absent.
Relevant log output
[2026-06-15T23:55:53] error: prolog failed: rc:127 output:[2026-06-15T23:55:53] error: _run_command_child_exec: execv(/usr/sbin/slurmd): No such file or directory
Additional environment details
Helm: 3.x
Kubernetes: minikube v1.35 (single-node, cgroup v2, IgnoreSystemd=yes)
slurm chart: 1.1.0
Slinky operator: 1.1.0
Slurm: 26.05.0
Other/Misc.
No response
Code of Conduct
Version
1.1.0
Which installation method(s) does this occur on?
Helm / Kubernetes manifests
Describe the bug.
Setting
vendor.nvidia.dcgm.enabled: false(the default) does not prevent the DCGM prolog/epilog scripts from being added toprologScriptRefs/epilogScriptRefsin the Controller CR. As a result, the Slinky operator always injectsProlog=prolog-90-dcgm.shandPrologFlags=Containintoslurm.conf, even on CPU-only clusters where DCGM is not intended to be used. This prevents workers from starting successfully on some CPU environments (minikube on macOS here).Minimum reproducible example
Relevant log output
Additional environment details
Other/Misc.
No response
Code of Conduct