Fix demo readme and grafana dashboard for autoscaling demo

api/v1alpha1/workerresourcetemplate_types.go

@@ -34,10 +34,10 @@ type WorkerResourceTemplateSpec struct {
 	//     PodDisruptionBudgets and other resources that select pods.
 	//
 	//   spec.metrics[*].external.metric.selector.matchLabels: {} (or with user labels)
-	//     The controller appends temporal_temporal_worker_deployment_name, temporal_worker_build_id, and
+	//     The controller appends temporal_worker_deployment_name, temporal_worker_build_id, and
 	//     temporal_namespace to any External metric selector where matchLabels is present.
 	//     User labels (e.g. task_type: "Activity") coexist alongside the injected keys.
-	//     Do not set temporal_temporal_worker_deployment_name, temporal_worker_build_id, or
+	//     Do not set temporal_worker_deployment_name, temporal_worker_build_id, or
 	//     temporal_namespace manually — the webhook will reject them.
 	// +kubebuilder:validation:Required
 	// +kubebuilder:pruning:PreserveUnknownFields

internal/demo/README.md

@@ -263,11 +263,12 @@ Stop the load generator (`Ctrl-C`) and watch the HPA scale back down as in-fligh
 `approximate_backlog_count` measures tasks queued in Temporal but not yet started on a worker. Adding it as a second HPA metric means the HPA scales up on *arriving* work even before slots are full — important for bursty traffic.
 
 > **Note:** Temporal Cloud emits `temporal_approximate_backlog_count` with a combined
-> `version="namespace/twd-name:build-id"` label that contains characters invalid in
-> Kubernetes label values (`/` and `:`). The recording rule in
-> `prometheus-stack-values.yaml` uses `label_replace` to extract `twd_name` and
-> `build_id` as separate k8s-compatible labels, producing `temporal_backlog_count_by_version`.
-> The HPA then selects on those labels — the same pair used by Phase 1.
+> `worker_version="<worker-deployment-name>_<build-id>"` label that easily exceeds Kubernetes max label
+> length of 63 characters. The recording rule in `prometheus-stack-values.yaml` uses `label_replace` 
+> to extract `temporal_worker_deployment_name` and `temporal_worker_build_id` as separate k8s-compatible 
+> labels, producing `temporal_backlog_count_by_version`. The HPA then selects on those labels — the same 
+> pair used by Phase 1. Temporal Cloud is in the process of rolling out the new separate labels, so this
+> workaround is required until then.
 
 **Step 1 — Create the Temporal Cloud credentials secret.**
 

internal/demo/k8s/grafana-dashboard.json

@@ -95,8 +95,8 @@
       },
       "targets": [
         {
-          "expr": "temporal_slot_utilization",
-          "legendFormat": "{{temporal_worker_deployment_name}} / {{temporal_worker_build_id}}",
+          "expr": "temporal_slot_utilization{temporal_worker_deployment_name=\"default_helloworld\"}",
+          "legendFormat": "{{worker_type}} - {{temporal_worker_deployment_name}} / {{temporal_worker_build_id}}",
           "refId": "A"
         }
       ],
@@ -127,7 +127,7 @@
       },
       "targets": [
         {
-          "expr": "temporal_backlog_count_by_version{task_type=\"Workflow\"}",
+          "expr": "temporal_backlog_count_by_version{task_type=\"Workflow\", temporal_worker_deployment_name=\"default_helloworld\"}",
           "legendFormat": "{{temporal_worker_deployment_name}} / {{temporal_worker_build_id}}",
           "refId": "A"
         }
@@ -152,7 +152,7 @@
       },
       "targets": [
         {
-          "expr": "temporal_backlog_count_by_version{task_type=\"Activity\"}",
+          "expr": "temporal_backlog_count_by_version{task_type=\"Activity\", temporal_worker_deployment_name=\"default_helloworld\"}",
           "legendFormat": "{{temporal_worker_deployment_name}} / {{temporal_worker_build_id}}",
           "refId": "A"
         }

internal/demo/k8s/prometheus-adapter-values.yaml

@@ -21,21 +21,22 @@ prometheus:
 rules:
   external:
     # Phase 1: slot utilization per worker version.
-    # HPA selector: metric.name=temporal_slot_utilization, matchLabels: twd_name + build_id.
-    # The worker emits twd_name and build_id as separate Prometheus labels (both valid
-    # Kubernetes label values), and the recording rule in prometheus-stack-values.yaml
-    # aggregates them into temporal_slot_utilization.
+    # HPA selector: metric.name=temporal_slot_utilization,
+    # matchLabels: temporal_worker_deployment_name + temporal_worker_build_id + temporal_namespace.
+    # The worker emits those as separate Prometheus labels (all valid Kubernetes label values), and
+    # the recording rule in prometheus-stack-values.yaml aggregates them into temporal_slot_utilization.
     - seriesQuery: 'temporal_slot_utilization{}'
       metricsQuery: 'avg(<<.Series>>{<<.LabelMatchers>>})'
       name:
         as: "temporal_slot_utilization"
       resources:
-        namespaced: false  # cluster-scoped: HPAs in any namespace can consume this metric
+        namespaced: false  # cluster-scoped: HPAs in any k8s namespace can consume this metric
 
     # Phase 2: approximate backlog count per worker version (from Temporal Cloud).
     # Uses the temporal_backlog_count_by_version recording rule.
     # cluster-scoped so HPAs in any namespace can consume it; temporal_worker_deployment_name
-    # + build_id matchLabels in the HPA are sufficient to select the right series.
+    # + temporal_worker_build_id + temporal_namespace matchLabels in the HPA are sufficient to
+    # select the right series.
     - seriesQuery: 'temporal_backlog_count_by_version{}'
       metricsQuery: 'sum(<<.Series>>{<<.LabelMatchers>>})'
       name:

internal/demo/k8s/prometheus-stack-values.yaml

@@ -61,8 +61,8 @@ additionalPrometheusRulesMap:
           # Slot utilization ratio per worker version, filtered to activity workers only.
           # Range: 0.0 (idle) to 1.0 (fully saturated).
           #
-          # The worker emits twd_name and build_id as separate labels (both valid
-          # Kubernetes label values), so no label_replace is needed here.
+          # The worker emits temporal_worker_deployment_name, temporal_worker_build_id, temporal_namespace, and worker_type
+          # as separate labels (both valid Kubernetes label values), so no label_replace is needed here.
           - record: temporal_slot_utilization
             expr: |
               sum by (temporal_worker_deployment_name, temporal_worker_build_id, temporal_namespace, worker_type) (
@@ -86,7 +86,8 @@ additionalPrometheusRulesMap:
           # Backlog count per worker version, shaped to match the label format that
           # Temporal Cloud will emit natively in a future release. This recording rule
           # is a temporary shim: once Temporal Cloud emits temporal_worker_deployment_name and
-          # build_id as separate labels, this rule can be deleted with no other changes.
+          # temporal_worker_build_id as separate labels, this rule can be deleted with no
+          # other changes. Note: this rule only works with Build IDs that don't have underscores.
           #
           # Current Temporal Cloud label:
           #   worker_version="{k8s-namespace}_{twd-name}_{build-id}"

internal/demo/util/observability.go

@@ -37,7 +37,7 @@ func configureObservability(deploymentName, buildID, temporalNamespace string, m
 	m = opentelemetry.NewMetricsHandler(opentelemetry.MetricsHandlerOptions{
 		Meter: metric.NewMeterProvider(metric.WithReader(exporter)).Meter("worker"),
 		InitialAttributes: attribute.NewSet(
-			attribute.String("temporal_temporal_worker_deployment_name", deploymentNameCleanForLabel),
+			attribute.String("temporal_worker_deployment_name", deploymentNameCleanForLabel),
 			attribute.String("temporal_worker_build_id", buildID),
 			attribute.String("temporal_namespace", temporalNamespace),
 		),