Open jakubkrzykowski opened 3 weeks ago
@jakubkrzykowski this sounds like a bug. Looking at the helm helper, we definitely don't check for the null
value.
Would you be able to submit a PR with a fix?
As a workaround, please set the value to none
for the time being.
Hi @elezar !
I was able to workaround it by setting it explicitly to none
as you suggested. Just wanted to be sure if this is not intentional.
Happy to raise a PR with a fix.
I've checked the _helpers.tpl file thoroughly and found that the problem lies with {{- else if ne (include "nvidia-device-plugin.configMapName" .) "true" -}} in the allPossibleMigStrategiesAreNone definition. With the default values, it will never be true because configMapName will always be an empty string, not the string "true", according to this:
{{- define "nvidia-device-plugin.configMapName" -}}
{{- $result := "" -}}
{{- if .Values.config.name -}}
{{- $result = .Values.config.name -}}
{{- else if not (empty .Values.config.map) -}}
{{- $result = printf "%s-%s" (include "nvidia-device-plugin.fullname" .) "configs" -}}
{{- end -}}
{{- $result -}}
{{- end -}}
@elezar here is the PR to address that - https://github.com/NVIDIA/k8s-device-plugin/pull/675
1. Quick Debug Information
N/A
2. Issue or feature description
When trying to deploy the latest (also tested on v0.13.0) version of the helm chart with the default values file I can see that
NVIDIA_MIG_MONITOR_DEVICES
env var are being set toall
andsecurityContext.capabilities.add
are being set toSYS_ADMIN
.allPossibleMigStrategiesAreNone
checks ifmigStrategy
is set to "none" but doesn't cover the scenario when this value is null (not set at all), which is the default behavior of the helm chart.Is this intentional?