Target Porter application nodes by default

charts/superlog-otel/Chart.yaml

@@ -2,7 +2,7 @@ apiVersion: v2
 name: superlog-otel
 description: Collect Kubernetes telemetry and export it to Superlog.
 type: application
-version: 0.1.0
+version: 0.1.1
 appVersion: "0.156.0"
 home: https://superlog.sh
 sources:

charts/superlog-otel/README.md

@@ -12,7 +12,7 @@ In Porter, open **Add-ons → Create add-on → Helm Chart** and enter:
 | Add-on name | `superlog-otel` |
 | Helm Repository URL | `https://superloglabs.github.io/helm-charts` |
 | Chart Name | `superlog-otel` |
-| Chart Version | `0.1.0` |
+| Chart Version | `0.1.1` |
 
 Paste this into **Values YAML**, replacing the placeholder with the project
 ingest key shown in Superlog:
@@ -31,11 +31,12 @@ Kubernetes cluster UID as its stable identity.
 
 The chart installs two coordinated collector roles in one Helm release:
 
-- A node agent (`DaemonSet`) collects stdout/stderr from every pod, host CPU,
-  memory, disk and network metrics, and kubelet pod/container metrics.
+- A node agent (`DaemonSet`) collects stdout/stderr from every pod on Porter's
+  Application nodes, host CPU, memory, disk and network metrics, and kubelet
+  pod/container metrics.
 - One cluster collector (`Deployment`) collects Kubernetes events and workload
-  state metrics and accepts application OTLP traces, logs, and metrics on ports
-  `4317` and `4318`.
+  state metrics across the cluster and accepts application OTLP traces, logs,
+  and metrics on ports `4317` and `4318`.
 - Both roles attach Kubernetes workload metadata and export directly to the
   selected Superlog project.
 
@@ -48,11 +49,32 @@ objects may contain sensitive environment values and create unnecessary
 telemetry volume. Kubernetes events, resource identity, and operational metrics
 are collected instead.
 
+## Porter node groups
+
+By default, the node agent targets Porter's **Application** node group. This
+keeps a one-click install schedulable on a fresh Porter cluster, where the fixed
+System nodes may already be at their pod limit. Kubernetes events and workload
+state metrics remain cluster-wide because the cluster collector is not a
+DaemonSet.
+
+To collect container logs and node metrics from every Linux node, replace the
+default in Porter's **Values YAML** with the following. Use this only when every
+node has capacity for one additional pod; otherwise Porter can remain in the
+**Deploying** state while an agent waits to schedule.
+
+```yaml
+collectors:
+  agent:
+    porterNodeGroups: []
+```
+
 ## Existing monitoring
 
 [Porter's built-in application metrics](https://docs.porter.run/applications/observability/monitoring)
-use Prometheus, so a standard Porter cluster does not have a competing
-OpenTelemetry Collector. Install the default chart as shown above.
+use a separate monitoring pipeline. It may collect some of the same Kubernetes
+and kubelet metrics for Porter's dashboards, but it does not send application
+container logs, metrics, or traces to Superlog. The default chart can therefore
+be installed alongside Porter's built-in monitoring.
 
 If the cluster's **Add-ons** page shows an OpenTelemetry, Datadog, New Relic,
 Fluent, Promtail, or Vector add-on that already exports the same data to

charts/superlog-otel/templates/daemonset.yaml

@@ -34,6 +34,17 @@ spec:
       nodeSelector:
         {{- toYaml . | nindent 8 }}
       {{- end }}
+      {{- with .Values.collectors.agent.porterNodeGroups }}
+      affinity:
+        nodeAffinity:
+          requiredDuringSchedulingIgnoredDuringExecution:
+            nodeSelectorTerms:
+              - matchExpressions:
+                  - key: porter.run/workload-kind
+                    operator: In
+                    values:
+                      {{- toYaml . | nindent 22 }}
+      {{- end }}
       containers:
         - name: collector
           image: "{{ .Values.image.repository }}:{{ .Values.image.tag }}"

charts/superlog-otel/tests/render-contract.sh

@@ -52,6 +52,9 @@ render() {
 helm lint "$chart_dir" \
   --set global.superlog.apiKey=sl_public_test >/dev/null
 
+grep -Fxq 'version: 0.1.1' "$chart_dir/Chart.yaml" || \
+  fail "expected chart version 0.1.1"
+
 # Default install owns node collection and one cluster-wide OTLP gateway.
 render
 assert_line "kind: DaemonSet"
@@ -61,6 +64,9 @@ assert_line_count 1 "kind: DaemonSet"
 assert_line_count 1 "kind: Deployment"
 assert_line_count 1 "kind: Service"
 assert_line_count 2 "        kubernetes.io/os: linux"
+assert_contains "key: porter.run/workload-kind"
+assert_contains "operator: In"
+assert_contains "- application"
 assert_contains "name: otlp"
 assert_contains "port: 4317"
 assert_contains "name: otlp-http"
@@ -85,6 +91,11 @@ assert_contains "Authorization: Bearer \${env:SUPERLOG_API_KEY}"
 assert_not_contains "x-api-key:"
 assert_contains 'value: "porter-monitoring"'
 
+# Clusters with capacity on every node can opt into full-node collection.
+render --set-json 'collectors.agent.porterNodeGroups=[]'
+assert_not_contains "key: porter.run/workload-kind"
+assert_line_count 2 "        kubernetes.io/os: linux"
+
 # Component suffixes remain distinct even at Kubernetes' 63-character limit.
 long_fullname="$(printf 'a%.0s' {1..63})"
 expected_agent_name="${long_fullname:0:57}-agent"

charts/superlog-otel/values.yaml

@@ -18,6 +18,12 @@ collectors:
   # host metrics, and kubelet metrics.
   agent:
     enabled: true
+    # Porter keeps system, monitoring, and application workloads in separate
+    # node groups. Application-only is the safe zero-config default because a
+    # fresh Porter's fixed system nodes may already be at their pod limit.
+    # Set this to [] only after ensuring every Linux node has a free pod slot.
+    porterNodeGroups:
+      - application
     nodeSelector:
       kubernetes.io/os: linux
     kubelet: