fix(chart): preserve HAMi workload labels in ServiceMonitor

charts/hami-webui/README.md

@@ -58,7 +58,7 @@ The command removes all the Kubernetes components associated with the chart and
 | fullnameOverride | string | `""`                                                                               |  |
 | hamiServiceMonitor.additionalLabels.jobRelease | string | `"hami-webui-prometheus"`                                                          |  |
 | hamiServiceMonitor.enabled | bool | `true`                                                                             |  |
-| hamiServiceMonitor.honorLabels | bool | `false`                                                                            |  |
+| hamiServiceMonitor.honorLabels | bool | `true`                                                                            | Keep HAMi's workload namespace/pod/container labels when they collide with Prometheus scrape-target labels. |
 | hamiServiceMonitor.interval | string | `"15s"`                                                                            |  |
 | hamiServiceMonitor.relabelings | list | `[]`                                                                               |  |
 | hamiServiceMonitor.svcNamespace | string | "kube-system"                                                                      | Default is "kube-system", but it should be set according to the namespace where the HAMi components are installed. || image.backend.pullPolicy | string | `"IfNotPresent"` |  |

charts/hami-webui/values.yaml

@@ -114,7 +114,9 @@ serviceMonitor:
 hamiServiceMonitor:
   enabled: true
   interval: 15s
-  honorLabels: false
+  # Preserve HAMi's namespace/pod/container workload labels when they collide
+  # with Prometheus scrape-target labels; required by the WebUI queries.
+  honorLabels: true
   additionalLabels:
     jobRelease: hami-webui-prometheus
   svcNamespace: kube-system
@@ -202,4 +204,4 @@ metricsExporter:
 # evaluates process.env at build, not at runtime. To change it for a deployed
 # cluster you need to rebuild packages/web with a different VUE_APP_REQUEST_TIMEOUT
 # (set in packages/web/.env.production or as a build-arg). It cannot be tuned
-# through this values.yaml.
+# through this values.yaml.

docs/installation/helm/index.md

@@ -59,6 +59,33 @@ To set up the HAMi-WebUI Helm repository so that you download the correct HAMi-W
 
    You should get the expected both 'hami-webui' and 'hami-webui-dcgm-exporter' in running state if installation is successful.
 
+### Prometheus scrape labels
+
+HAMi's vgpu-monitor exposes workload identity as `namespace`, `pod`, and
+`container`. When Prometheus adds scrape-target labels with the same names and
+source labels are not honored, it renames the workload labels to `exported_*`.
+HAMi-WebUI's queries then return no series, and task utilization is reported as
+`0`.
+
+The included ServiceMonitor sets `hamiServiceMonitor.honorLabels: true`. If an
+external Prometheus discovers this ServiceMonitor, no additional scrape job is
+needed. For a separately managed scrape, set `honorLabels: true` on its
+ServiceMonitor or use the equivalent raw Prometheus setting:
+
+```yaml
+scrape_configs:
+  - job_name: hami-device-plugin-monitor
+    honor_labels: true
+    # ...
+```
+
+`Prometheus.spec.overrideHonorLabels: true` forces
+`honorLabels` off for every ServiceMonitor, including this one.
+
+If the same Prometheus selects both this chart's ServiceMonitor and HAMi's
+built-in device-plugin ServiceMonitor (`prometheus.enabled` in the HAMi chart),
+the endpoint is scraped twice. Configure that Prometheus to select only one.
+
 ### Access HAMi-WebUI
 
 1. Configure ~/.kube/config in your localhost to be able to connect your cluster.
@@ -110,4 +137,4 @@ If you want to delete the namespace `hami`, then run the command:
 
 ```bash
 kubectl delete namespace hami
-```
+```