NVIDIA / gpu-operator / 30657870311
38%
main: 38%

LAST BUILD BRANCH: pull-request/2754
DEFAULT BRANCH: main
Ran
Jobs 1
Files 73
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 38.464% (-0.004%) from 38.468%
30657870311

push

github

rajathagasthya 
Attribute DCGM exporter metrics to pods on DRA nodes by default

Under GPUCluster, dcgm-exporter metrics carry only hardware labels: no
pod, namespace, container or dra_* label appears unless the user sets
dcgmExporter.enablePodLabels. Under ClusterPolicy the same three pod
labels are present out of the box, so migrating to GPUCluster silently
loses metric attribution.

The two modes reach attribution differently. The kubelet reports
device-plugin allocations as GPU UUIDs, which the exporter can match
against DCGM directly from the pod-resources socket with no API access.
DRA allocations are reported as (pool, device) names instead, so the
exporter must read the ResourceSlice published for the node to resolve
them back to a UUID. That needs a ServiceAccount token and read access
to resourceslices, both of which were gated behind the pod-labels flag
along with KUBERNETES_ENABLE_DRA itself.

Render the DRA env var, the token mount and the read ClusterRole and
binding unconditionally. Attribution is not an opt-in feature in either
mode now; enablePodLabels and enablePodUID go back to controlling only
the pod-label and pod-UID enrichment they are named for. Neither is
defaulted on: with no allowlist configured the exporter copies every pod
label onto every GPU metric series, which is left to the operator to opt
into deliberately.

PodMetadataEnabled had no remaining template consumers and is removed.

Signed-off-by: Rajath Agasthya <ragasthya@nvidia.com>

5575 of 14494 relevant lines covered (38.46%)

0.44 hits per line

Jobs
ID Job ID Ran Files Coverage
1 30657870311.1 73
38.46
 GitHub Action Run
Source Files on build 30657870311