feat(pyrra): add Parca Agent write-availability SLO (#443)

Author: metalmatze

pyrra/environments/scaleway-parca-demo/main.jsonnet

@@ -8,13 +8,51 @@ local pyrra = p.pyrra({
   ingressHosts: ['pyrra.parca.dev'],
 });
 
+// SLO for the Parca Agent writing profiles to the Parca server. This is the
+// core data path of the demo: agents continuously push profiles via the
+// ProfileStore Write gRPC method, so a high error rate there means the
+// user-facing pipeline is broken. The metric comes from the parca-agent
+// PodMonitors, which are the only thing this Prometheus currently scrapes.
+local writeAvailabilitySLO(name, job, description) = {
+  local writeMetric = 'grpc_client_handled_total{job="%s",grpc_service="parca.profilestore.v1alpha1.ProfileStoreService",grpc_method="Write"%s}',
+  apiVersion: 'pyrra.dev/v1alpha1',
+  kind: 'ServiceLevelObjective',
+  metadata: {
+    name: name,
+    namespace: 'monitoring',
+    labels: {
+      'app.kubernetes.io/name': 'pyrra',
+      'app.kubernetes.io/component': 'slo',
+    },
+  },
+  spec: {
+    target: '99.9',
+    window: '4w',
+    description: description,
+    indicator: {
+      ratio: {
+        errors: { metric: writeMetric % [job, ',grpc_code!="OK"'] },
+        total: { metric: writeMetric % [job, ''] },
+      },
+    },
+  },
+};
+
+local slos = [
+  writeAvailabilitySLO(
+    'parca-agent-write-availability',
+    'parca/parca-agent',
+    'Parca Agent writes profiles to the Parca server via ProfileStore Write.',
+  ),
+];
+
 {
   apiVersion: 'v1',
   kind: 'List',
   items: [
     pyrra[name]
     for name in std.objectFields(pyrra)
-    // SLO definitions come later; only ship the Pyrra components for now.
+    // The upstream example SLOs target cluster components we do not scrape.
     if !std.startsWith(name, 'slo-')
-  ],
+  ] + slos,
 }