Repository navigation
Conversation
Add a commented draft of release notes for the next GPU Operator release, with the metric rename table, upgrade notes, and the OpenShift dashboard note. Add a reminder to update the component table at release time. Signed-off-by: Nick Curry <ncurry@nvidia.com>
Use the default-counters.csv path, the DCGM 4.7 metric names in sample output, and the new utilization metric in the Prometheus check. Add a custom metrics files section and a note about the Grafana dashboard. Signed-off-by: Nick Curry <ncurry@nvidia.com>
Documentation preview |
Use the real dcgmExporter.config.name value, add the ConfigMap procedure, explain that old field names still load but are deprecated, refresh the container image examples, and link the DCGM Exporter reference pages. Signed-off-by: Nick Curry <ncurry@nvidia.com>
List the remaining dcgmExporter Helm values, describe how to enable KUBERNETES_VIRTUAL_GPUS for GPU sharing, and use a current exporter tag in the air-gapped example. Signed-off-by: Nick Curry <ncurry@nvidia.com>
Signed-off-by: Nick Curry <ncurry@nvidia.com>
The DCGM Exporter page now covers only the GPU Operator: the resources it creates, the environment it sets, the dcgmExporter Helm values, custom metrics files through a ConfigMap, Prometheus scraping through the ServiceMonitor, shared GPUs and MIG, and a separate DCGM host engine. Standalone docker run steps, the exporter's own options, the metrics sample output, and the MIG output examples are replaced by links to the DCGM Exporter documentation on docs.nvidia.com. The custom metrics section now says that the Operator mounts the file with subPath, so pods must be restarted after a ConfigMap change, and that DCGM_EXPORTER_CONFIGMAP_DATA is deprecated. The host engine section says that DCGM Exporter 4.9.0 needs DCGM 4.7 or later and that DCGM on the host needs dcgmExporter.hostNetwork. The dcgmExporter rows added to the getting-started table move to the DCGM Exporter page, with a link from the table. The time-slicing metrics text gets its own section in the GPU sharing page. Signed-off-by: Nick Curry <ncurry@nvidia.com>
Remove "Integrating GPU Telemetry into Kubernetes", which set up drivers, Docker, the toolkit, and the device plugin by hand, and "Setting up Prometheus", which installed kube-prometheus-stack, Grafana, and the exporter's own Helm chart. The GPU Operator install guide and the DCGM Exporter documentation cover these tasks. Redirect integrating-telemetry-kubernetes.html to the GPU Operator getting-started page, and kube-prometheus.html to the Prometheus section of the DCGM Exporter page. Remove the screenshots and diagrams that only these pages used. Signed-off-by: Nick Curry <ncurry@nvidia.com>
Replace the metric rename table with an example and a link to the DCGM Exporter release notes. Keep the upgrade notes to what a GPU Operator user must do: update queries and dashboards, upgrade the dcgm and dcgmExporter images together when a separate DCGM host engine is used, because DCGM Exporter 4.9.0 needs DCGM 4.7 or later, check pod logs for settings that now stop startup, and roll back both components together. Restore the air-gapped values sample, which shows GPU Operator v1.9.0, and say in the life-cycle reminder that the DCGM and DCGM Exporter rows change together. Signed-off-by: Nick Curry <ncurry@nvidia.com>
…otes Signed-off-by: Nick Curry <ncurry@nvidia.com>
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Draft: merge with the GPU Operator release that ships DCGM Exporter 4.9.0.
The GPU telemetry pages now cover only what the NVIDIA GPU Operator does for DCGM Exporter.
Everything about the exporter itself (installs outside the Operator, options, config file, metrics, Prometheus setup, release notes) links to the DCGM Exporter documentation on docs.nvidia.com.
Changes
gpu-telemetry/dcgm-exporter.rst: rewritten as "DCGM Exporter with the NVIDIA GPU Operator".dcgmExporterHelm values (anddcgm.enabled), checked against the chartvalues.yamland theClusterPolicytypes.dcgmExporter.config.name, orconfig.createandconfig.data). Pods must be restarted after a ConfigMap change, because the Operator mounts the file withsubPath.DCGM_EXPORTER_CONFIGMAP_DATAis deprecated.dcgm.enabled, or DCGM on the host (needsdcgmExporter.hostNetwork). DCGM Exporter 4.9.0 needs DCGM 4.7 or later.docker runexamples, the exporter option table, sample output, and MIG output. These now link to the DCGM Exporter docs.gpu-telemetry/integrating-telemetry-kubernetes.rst: removed. It set up drivers, Docker, the toolkit, and the device plugin by hand. Redirects to the GPU Operator getting-started page.gpu-telemetry/kube-prometheus.rst: removed. It installed kube-prometheus-stack, Grafana, and the exporter's own Helm chart. Redirects to the Prometheus section of the DCGM Exporter page.gpu-telemetry/about-telemetry.rst,index.rst: a short overview, fixed quick links, and the toctree without the removed pages. Unused screenshots and diagrams are removed.repo.toml: redirects for the two removed pages.gpu-operator/release-notes.rst: a commented draft for the next release, in the usual New Features, Improvements, Known Issues, and Removals and Deprecations sections. It gives one rename example and links to the DCGM Exporter release notes for the full list. It says what an Operator user must do: update queries and dashboards, upgrade thedcgmanddcgmExporterimages together when a separate host engine is used, and roll back both together. It stays a comment because the version number and date are not known yet.gpu-operator/life-cycle-policy.rst: a reminder to set the DCGM Exporter and DCGM rows for the new release.gpu-operator/getting-started.rst: thedcgmExporter.enabledrow links to the full list of DCGM Exporter values.gpu-operator/gpu-sharing.rst: a "GPU Metrics with Time-Slicing" section that setsKUBERNETES_VIRTUAL_GPUSthroughdcgmExporter.env. It replaces the old limitation.openshift/enable-gpu-monitoring-dashboard.rst: a note about which dashboard file to use and theaccelerator_*rules.Links to the DCGM documentation are plain external links. The repository's DCGM intersphinx inventory (
work/dcgm-offline.inv) is from DCGM 3.3 and has no DCGM Exporter labels.Testing
./repo docsin the container fromdocker/Dockerfile, as CI does. The build has no warnings, the same as before the change.gpu-telemetry: all links are OK, including the DCGM Exporter pages on docs.nvidia.com.git diff --check.gpu-operator, because theopenshiftlinkcheck stops on old broken links in pages this change does not touch. The new#dcgm-exporter-helm-valuesand#dcgm-exporter-custom-metricsanchors on the gpu-telemetry page are not published yet.