heheHEADmain

3 files changed, 194 insertions, 0 deletions

diff --git a/terraform/modules/observability/main.tf b/terraform/modules/observability/main.tf
new file mode 100644
index 0000000..2f88f2e
--- /dev/null
+++ b/terraform/modules/observability/main.tf
@@ -0,0 +1,156 @@
+resource "kubernetes_namespace_v1" "monitoring" {
+  metadata {
+    name = var.namespace
+  }
+}
+
+resource "kubernetes_namespace_v1" "ingress" {
+  metadata {
+    name = "ingress-nginx"
+  }
+}
+
+resource "helm_release" "ingress_nginx" {
+  name       = "ingress-nginx"
+  repository = "https://kubernetes.github.io/ingress-nginx"
+  chart      = "ingress-nginx"
+  version    = var.ingress_nginx_version
+  namespace  = kubernetes_namespace_v1.ingress.metadata[0].name
+  wait       = true
+  timeout    = 300
+
+  values = [
+    yamlencode({
+      controller = {
+        hostPort = { enabled = true, ports = { http = 80, https = 443 } }
+        service  = { type = "NodePort" }
+        nodeSelector = {
+          "ingress-ready" = "true"
+        }
+        tolerations = [
+          { key = "node-role.kubernetes.io/control-plane", operator = "Equal", effect = "NoSchedule" },
+          { key = "node-role.kubernetes.io/master", operator = "Equal", effect = "NoSchedule" },
+        ]
+        publishService = { enabled = false }
+        admissionWebhooks = { enabled = false } # speeds up kind cluster installs
+        # Cap worker_processes so nginx doesn't try to spawn 14 threads under
+        # CPU pressure from vLLM cold-starts. With auto (= one per CPU) it
+        # sometimes hits pthread EAGAIN and workers die without respawn.
+        config = {
+          "worker-processes" = "4"
+        }
+      }
+    }),
+  ]
+}
+
+resource "helm_release" "kps" {
+  name       = "kube-prometheus-stack"
+  repository = "https://prometheus-community.github.io/helm-charts"
+  chart      = "kube-prometheus-stack"
+  version    = var.kps_version
+  namespace  = kubernetes_namespace_v1.monitoring.metadata[0].name
+  wait       = true
+  timeout    = 600
+
+  values = [
+    yamlencode({
+      fullnameOverride = "kps"
+      prometheus = {
+        prometheusSpec = {
+          # Let Prometheus pick up ServiceMonitors from any namespace matching
+          # the release=kube-prometheus-stack label (the chart's default).
+          serviceMonitorSelectorNilUsesHelmValues = false
+          podMonitorSelectorNilUsesHelmValues     = false
+          ruleSelectorNilUsesHelmValues           = false
+          retention                               = "2d"
+          resources = {
+            requests = { cpu = "100m", memory = "400Mi" }
+            limits   = { memory = "1Gi" }
+          }
+        }
+        ingress = {
+          enabled          = true
+          ingressClassName = "nginx"
+          hosts            = ["prom.localtest.me"]
+        }
+      }
+      alertmanager = { enabled = false }
+      grafana = {
+        adminPassword = var.grafana_admin_password
+        sidecar = {
+          dashboards = {
+            enabled        = true
+            label          = "grafana_dashboard"
+            labelValue     = "1"
+            searchNamespace = "ALL"
+          }
+        }
+        service = { type = "ClusterIP" }
+        ingress = {
+          enabled          = true
+          ingressClassName = "nginx"
+          hosts            = ["grafana.localtest.me"]
+        }
+      }
+    }),
+  ]
+}
+
+resource "helm_release" "prometheus_adapter" {
+  name       = "prometheus-adapter"
+  repository = "https://prometheus-community.github.io/helm-charts"
+  chart      = "prometheus-adapter"
+  version    = var.prometheus_adapter_version
+  namespace  = kubernetes_namespace_v1.monitoring.metadata[0].name
+  wait       = true
+  timeout    = 300
+
+  values = [
+    yamlencode({
+      prometheus = {
+        url  = "http://kps-prometheus.${kubernetes_namespace_v1.monitoring.metadata[0].name}.svc"
+        port = 9090
+      }
+      rules = {
+        default = false
+        custom = [
+          {
+            # In-flight request count per pod; basis for autoscaling.
+            # vLLM exposes this as a gauge per model-engine.
+            seriesQuery = "vllm:num_requests_running{namespace!=\"\",pod!=\"\"}"
+            resources = {
+              overrides = {
+                namespace = { resource = "namespace" }
+                pod       = { resource = "pod" }
+              }
+            }
+            name = {
+              matches = "^vllm:num_requests_running$"
+              as      = "vllm:num_requests_running"
+            }
+            metricsQuery = "avg(<<.Series>>{<<.LabelMatchers>>}) by (<<.GroupBy>>)"
+          },
+          {
+            # Waiting (queued) requests per pod — an alternative scale signal.
+            seriesQuery = "vllm:num_requests_waiting{namespace!=\"\",pod!=\"\"}"
+            resources = {
+              overrides = {
+                namespace = { resource = "namespace" }
+                pod       = { resource = "pod" }
+              }
+            }
+            name = {
+              matches = "^vllm:num_requests_waiting$"
+              as      = "vllm:num_requests_waiting"
+            }
+            metricsQuery = "avg(<<.Series>>{<<.LabelMatchers>>}) by (<<.GroupBy>>)"
+          },
+        ]
+      }
+    }),
+  ]
+
+  depends_on = [helm_release.kps]
+}
+
diff --git a/terraform/modules/observability/outputs.tf b/terraform/modules/observability/outputs.tf
new file mode 100644
index 0000000..06a507d
--- /dev/null
+++ b/terraform/modules/observability/outputs.tf
@@ -0,0 +1,11 @@
+output "namespace" {
+  value = kubernetes_namespace_v1.monitoring.metadata[0].name
+}
+
+output "grafana_service" {
+  value = "kube-prometheus-stack-grafana.${kubernetes_namespace_v1.monitoring.metadata[0].name}.svc.cluster.local"
+}
+
+output "prometheus_service" {
+  value = "kps-prometheus.${kubernetes_namespace_v1.monitoring.metadata[0].name}.svc.cluster.local"
+}
diff --git a/terraform/modules/observability/variables.tf b/terraform/modules/observability/variables.tf
new file mode 100644
index 0000000..6aeaca3
--- /dev/null
+++ b/terraform/modules/observability/variables.tf
@@ -0,0 +1,27 @@
+variable "namespace" {
+  type    = string
+  default = "monitoring"
+}
+
+variable "kps_version" {
+  type        = string
+  default     = "65.5.1"
+  description = "kube-prometheus-stack chart version."
+}
+
+variable "ingress_nginx_version" {
+  type        = string
+  default     = "4.11.3"
+  description = "ingress-nginx chart version."
+}
+
+variable "grafana_admin_password" {
+  type      = string
+  default   = "admin"
+  sensitive = true
+}
+
+variable "prometheus_adapter_version" {
+  type    = string
+  default = "4.11.0"
+}