chore(k8s): update priority classes to play nice with karpenter & preemption

infra/default-builds/dockerfiles/test-mm-lobby-echo/Dockerfile

@@ -1,4 +1,4 @@
-FROM clux/muslrust:1.77.2 AS build
+FROM clux/muslrust:1.77.2-stable AS build
 RUN cargo new --bin /app
 WORKDIR /app
 COPY Cargo.toml ./

infra/tf/cloudflare_tunnels/k8s.tf

@@ -47,6 +47,8 @@ resource "kubernetes_deployment" "cloudflared" {
 			}
 
 			spec {
+				priority_class_name = "service-priority"
+
 				container {
 					image = "cloudflare/cloudflared:2023.8.2"
 					name = "cloudflared"

infra/tf/grafana/grafana.tf

@@ -30,6 +30,7 @@ resource "helm_release" "grafana" {
 	chart = "grafana"
 	version = "7.3.9"
 	values = [yamlencode({
+		priorityClassName = "monitoring-priority"
 		"grafana.ini" = {
 			auth = {
 				disable_login_form = true

infra/tf/k8s_cluster_aws/overprovisioning.tf

@@ -0,0 +1,70 @@
+# Define low-priority deployments to maintain unallocated capacity on Karpenter
+# cluster to prevent premption.
+#
+# Higher replicas + lower resoruces = more smaller pods can preempt
+# Lower replicas + higher resources = larger pod can preempt, but will require bigger preempted "wedges"
+#
+# https://aws.amazon.com/blogs/containers/eliminate-kubernetes-node-scaling-lag-with-pod-priority-and-over-provisioning/
+
+resource "kubernetes_namespace" "karpenter_overprovision" {
+	metadata {
+		name = "karpenter-overprovision"
+	}
+}
+
+resource "kubernetes_priority_class" "overprovision_priority" {
+	metadata {
+		name = "overprovision-priority"
+	}
+	value = -1
+}
+
+resource "kubernetes_deployment" "overprovision" {
+	metadata {
+		name = "overprovision"
+		namespace = kubernetes_namespace.karpenter_overprovision.metadata.0.name
+		labels = {
+			app = "overprovisioning"
+		}
+	}
+
+	spec {
+		replicas = 2
+
+		selector {
+			match_labels = {
+				app = "overprovisioning"
+			}
+		}
+
+		template {
+			metadata {
+				labels = {
+					app = "overprovisioning"
+				}
+			}
+
+			spec {
+				container {
+					name = "pause"
+					image = "registry.k8s.io/pause"
+
+					resources {
+						requests = {
+							cpu	= "1"
+							memory = "500Mi"
+						}
+
+						limits = {
+							cpu = "1"
+							memory = "500Mi"
+						}
+					}
+				}
+
+				priority_class_name = kubernetes_priority_class.overprovision_priority.metadata.0.name
+			}
+		}
+	}
+}
+

infra/tf/k8s_infra/clickhouse.tf

@@ -27,16 +27,6 @@ resource "kubernetes_namespace" "clickhouse" {
 	}
 }
 
-resource "kubernetes_priority_class" "clickhouse_priority" {
-	count = local.clickhouse_enabled ? 1 : 0
-
-	metadata {
-		name = "clickhouse-priority"
-	}
-
-	value = 40
-}
-
 resource "helm_release" "clickhouse" {
 	count = local.clickhouse_enabled ? 1 : 0
 	depends_on = [null_resource.daemons]
@@ -57,7 +47,7 @@ resource "helm_release" "clickhouse" {
 			replicaCount = 1
 		}
 
-		priorityClassName = kubernetes_priority_class.clickhouse_priority.0.metadata.0.name
+		priorityClassName = kubernetes_priority_class.stateful_priority.metadata.0.name
 		resources = var.limit_resources ? {
 			limits = {
 				memory = "${local.service_clickhouse.resources.memory}Mi"

infra/tf/k8s_infra/cockroachdb.tf

@@ -24,14 +24,6 @@ resource "kubernetes_namespace" "cockroachdb" {
 	}
 }
 
-resource "kubernetes_priority_class" "cockroachdb_priority" {
-	metadata {
-		name = "cockroachdb-priority"
-	}
-
-	value = 40
-}
-
 # NOTE: Helm chart is no longer supported by CockroachDB. However, it's intended to be used only for development and it's the easiest to set up.
 resource "helm_release" "cockroachdb" {
 	depends_on = [null_resource.daemons]
@@ -46,7 +38,7 @@ resource "helm_release" "cockroachdb" {
 		statefulset = {
 			replicas = local.service_cockroachdb.count
 
-			priorityClassName = kubernetes_priority_class.cockroachdb_priority.metadata.0.name
+			priorityClassName = kubernetes_priority_class.stateful_priority.metadata.0.name
 
 			resources = var.limit_resources ? {
 				limits = {

infra/tf/k8s_infra/imagor.tf

@@ -56,16 +56,6 @@ resource "kubernetes_namespace" "imagor" {
 	}
 }
 
-resource "kubernetes_priority_class" "imagor_priority" {
-	count = var.imagor_enabled ? 1 : 0
-
-	metadata {
-		name = "imagor-priority"
-	}
-
-	value = 35
-}
-
 resource "kubernetes_deployment" "imagor" {
 	count = var.imagor_enabled ? 1 : 0
 	depends_on = [null_resource.daemons, module.docker_auth]
@@ -92,7 +82,7 @@ resource "kubernetes_deployment" "imagor" {
 			}
 
 			spec {
-				priority_class_name = kubernetes_priority_class.imagor_priority.0.metadata.0.name
+				priority_class_name = kubernetes_priority_class.service_priority.metadata.0.name
 
 				# MARK: Docker auth
 				image_pull_secrets {

infra/tf/k8s_infra/loki.tf

@@ -16,16 +16,6 @@ resource "kubernetes_namespace" "loki" {
 	}
 }
 
-resource "kubernetes_priority_class" "loki_priority" {
-	count = var.prometheus_enabled ? 1 : 0
-
-	metadata {
-		name = "loki-priority"
-	}
-
-	value = 40
-}
-
 resource "helm_release" "loki" {
 	count = var.prometheus_enabled ? 1 : 0
 
@@ -36,7 +26,7 @@ resource "helm_release" "loki" {
 	version = "5.36.0"
 	values = [yamlencode({
 		global = {
-			priorityClassName = kubernetes_priority_class.loki_priority.0.metadata.0.name
+			priorityClassName = kubernetes_priority_class.monitoring_priority.metadata.0.name
 		}
 		loki = {
 			auth_enabled = false

infra/tf/k8s_infra/minio.tf

@@ -26,14 +26,6 @@ module "minio_secrets" {
 	optional = true
 }
 
-resource "kubernetes_priority_class" "minio_priority" {
-	metadata {
-		name = "minio-priority"
-	}
-
-	value = 40
-}
-
 resource "helm_release" "minio" {
 	depends_on = [null_resource.daemons]
 	count = local.has_minio ? 1 : 0
@@ -48,7 +40,7 @@ resource "helm_release" "minio" {
 			storageClass = var.k8s_storage_class
 		}
 		replicaCount = local.service_minio.count
-		priorityClassName = kubernetes_priority_class.minio_priority.metadata.0.name
+		priorityClassName = kubernetes_priority_class.service_priority.metadata.0.name
 		resources = var.limit_resources ? {
 			limits = {
 				memory = "${local.service_minio.resources.memory}Mi"

infra/tf/k8s_infra/nats.tf

@@ -14,14 +14,7 @@ resource "kubernetes_namespace" "nats" {
 	}
 }
 
-resource "kubernetes_priority_class" "nats_priority" {
-	metadata {
-		name = "nats-priority"
-	}
-
-	value = 40
-}
-
+# TODO(RVTEE-105): Fix priority class
 resource "helm_release" "nats" {
 	depends_on = [null_resource.daemons]
 
@@ -37,11 +30,6 @@ resource "helm_release" "nats" {
 				replicas = local.service_nats.count
 			}
 		}
-		podTemplate = {
-			merge = {
-				priorityClassName = kubernetes_priority_class.nats_priority.metadata.0.name
-			}
-		}
 		container = {
 			env = {
 				# See https://artifacthub.io/packages/helm/grafana/grafana#nats-container-resources

infra/tf/k8s_infra/nomad.tf

@@ -190,16 +190,6 @@ resource "kubectl_manifest" "nomad_server_monitor" {
 	})
 }
 
-resource "kubernetes_priority_class" "nomad_priority" {
-	count = var.edge_enabled ? 1 : 0
-
-	metadata {
-		name = "nomad-priority"
-	}
-
-	value = 40
-}
-
 resource "kubernetes_stateful_set" "nomad_server" {
 	count = var.edge_enabled ? 1 : 0
 	depends_on = [null_resource.daemons]
@@ -234,7 +224,7 @@ resource "kubernetes_stateful_set" "nomad_server" {
 			}
 
 			spec {
-				priority_class_name = kubernetes_priority_class.nomad_priority.0.metadata.0.name
+				priority_class_name = kubernetes_priority_class.stateful_priority.metadata.0.name
 
 				security_context {
 					run_as_user = 0

infra/tf/k8s_infra/nsfw_api.tf

@@ -16,16 +16,6 @@ resource "kubernetes_namespace" "nsfw_api" {
 	}
 }
 
-resource "kubernetes_priority_class" "nsfw_api_priority" {
-	count = var.nsfw_api_enabled ? 1 : 0
-
-	metadata {
-		name = "nsfw-api-priority"
-	}
-
-	value = 40
-}
-
 resource "kubernetes_deployment" "nsfw_api" {
 	count = var.nsfw_api_enabled ? 1 : 0
 	depends_on = [null_resource.daemons, module.docker_auth]
@@ -52,7 +42,7 @@ resource "kubernetes_deployment" "nsfw_api" {
 			}
 
 			spec {
-				priority_class_name = kubernetes_priority_class.nsfw_api_priority.0.metadata.0.name
+				priority_class_name = kubernetes_priority_class.service_priority.metadata.0.name
 
 				# MARK: Docker auth
 				image_pull_secrets {

infra/tf/k8s_infra/priority_class.tf

@@ -0,0 +1,41 @@
+# Used for services that can be preempted often.
+#
+# We set almost everything to this class (even crucial infrastrcuture) because
+# we need to wait for Karpenter to boot a new node instead of shutting down
+# existing services.
+resource "kubernetes_priority_class" "service_priority" {
+
+	metadata {
+		name = "service-priority"
+	}
+	value = 50
+}
+
+# Used for anything required to monitor the other services. These should take
+# priority no matter what in order to ensure we have visibility on what's going
+# on.
+resource "kubernetes_priority_class" "monitoring_priority" {
+
+	metadata {
+		name = "monitoring-priority"
+	}
+	value = 50
+}
+
+# Used for anything stateful that should not be frequently preempted.
+resource "kubernetes_priority_class" "stateful_priority" {
+
+	metadata {
+		name = "stateful-priority"
+	}
+	value = 60
+}
+
+# Used for daemons that run on the machines and need to be scheduled no matter what.
+resource "kubernetes_priority_class" "daemon_priority" {
+
+	metadata {
+		name = "daemon-priority"
+	}
+	value = 90
+}

infra/tf/k8s_infra/prometheus.tf

@@ -86,25 +86,6 @@ resource "kubernetes_namespace" "prometheus" {
 	}
 }
 
-# Set a high priority for Node Exporter so it can run on all nodes
-resource "kubernetes_priority_class" "node_exporter_priority" {
-	count = var.prometheus_enabled ? 1 : 0
-
-	metadata {
-		name = "node-exporter-priority"
-	}
-	value = 90
-}
-
-resource "kubernetes_priority_class" "prometheus_priority" {
-	count = var.prometheus_enabled ? 1 : 0
-
-	metadata {
-		name = "prometheus-priority"
-	}
-	value = 40
-}
-
 resource "helm_release" "prometheus" {
 	count = var.prometheus_enabled ? 1 : 0
 	depends_on = [helm_release.vpa]
@@ -122,7 +103,7 @@ resource "helm_release" "prometheus" {
 					cpu = "${local.service_node_exporter.resources.cpu}m"
 				}
 			} : null
-			priorityClassName = kubernetes_priority_class.node_exporter_priority.0.metadata.0.name
+			priorityClassName = kubernetes_priority_class.daemon_priority.metadata.0.name
 			affinity = {
 				nodeAffinity = {
 					requiredDuringSchedulingIgnoredDuringExecution = {
@@ -270,7 +251,7 @@ resource "helm_release" "prometheus" {
 					}
 				}
 
-				priorityClassName = kubernetes_priority_class.prometheus_priority.0.metadata.0.name
+				priorityClassName = kubernetes_priority_class.monitoring_priority.metadata.0.name
 				resources = var.limit_resources ? {
 					limits = {
 						memory = "${local.service_prometheus.resources.memory}Mi"