From c4ff5090866f4739a6c73a3e47ac8ee759859e8a Mon Sep 17 00:00:00 2001
From: Garrett Delfosse <garrett@coder.com>
Date: Wed, 22 Jan 2025 19:44:39 +0000
Subject: [PATCH 1/5] feat: remote write prometheus data

---
 scaletest/terraform/action/coder_traffic.tf | 17 +----
 scaletest/terraform/action/gcp_clusters.tf  | 10 +--
 scaletest/terraform/action/gcp_db.tf        |  2 +-
 scaletest/terraform/action/gcp_vpc.tf       | 24 ++-----
 scaletest/terraform/action/prometheus.tf    | 78 +++++++++++++++++++++
 scaletest/terraform/action/vars.tf          |  5 ++
 scaletest/terraform/infra/gcp_db.tf         |  2 +-
 scaletest/terraform/infra/gcp_vpc.tf        |  4 +-
 8 files changed, 100 insertions(+), 42 deletions(-)
 create mode 100644 scaletest/terraform/action/prometheus.tf

diff --git a/scaletest/terraform/action/coder_traffic.tf b/scaletest/terraform/action/coder_traffic.tf
index bea829427af82..ebc809273c598 100644
--- a/scaletest/terraform/action/coder_traffic.tf
+++ b/scaletest/terraform/action/coder_traffic.tf
@@ -5,7 +5,6 @@ locals {
 
   traffic_types = {
     ssh = {
-      wait_duration = "0m"
       duration      = "30m"
       job_timeout   = "35m"
       flags = [
@@ -13,13 +12,11 @@ locals {
       ]
     }
     webterminal = {
-      wait_duration = "5m"
       duration      = "25m"
       job_timeout   = "30m"
       flags         = []
     }
     app = {
-      wait_duration = "10m"
       duration      = "20m"
       job_timeout   = "25m"
       flags = [
@@ -39,14 +36,6 @@ resource "time_sleep" "wait_baseline" {
   create_duration = local.wait_baseline_duration
 }
 
-resource "time_sleep" "wait_traffic" {
-  for_each = local.traffic_types
-
-  depends_on = [time_sleep.wait_baseline]
-
-  create_duration = local.traffic_types[each.key].wait_duration
-}
-
 resource "kubernetes_job" "workspace_traffic_primary" {
   provider = kubernetes.primary
 
@@ -106,7 +95,7 @@ resource "kubernetes_job" "workspace_traffic_primary" {
     create = local.traffic_types[each.key].job_timeout
   }
 
-  depends_on = [time_sleep.wait_baseline, time_sleep.wait_traffic[each.key]]
+  depends_on = [time_sleep.wait_baseline]
 }
 
 resource "kubernetes_job" "workspace_traffic_europe" {
@@ -169,7 +158,7 @@ resource "kubernetes_job" "workspace_traffic_europe" {
     create = local.traffic_types[each.key].job_timeout
   }
 
-  depends_on = [time_sleep.wait_baseline, time_sleep.wait_traffic[each.key]]
+  depends_on = [time_sleep.wait_baseline]
 }
 
 resource "kubernetes_job" "workspace_traffic_asia" {
@@ -232,5 +221,5 @@ resource "kubernetes_job" "workspace_traffic_asia" {
     create = local.traffic_types[each.key].job_timeout
   }
 
-  depends_on = [time_sleep.wait_baseline, time_sleep.wait_traffic[each.key]]
+  depends_on = [time_sleep.wait_baseline]
 }
diff --git a/scaletest/terraform/action/gcp_clusters.tf b/scaletest/terraform/action/gcp_clusters.tf
index c41d06c6c1c83..5681ff8b44ce5 100644
--- a/scaletest/terraform/action/gcp_clusters.tf
+++ b/scaletest/terraform/action/gcp_clusters.tf
@@ -10,21 +10,21 @@ locals {
       url       = "http://${var.name}-scaletest.${var.cloudflare_domain}"
       region    = "us-east1"
       zone      = "us-east1-c"
-      cidr      = "10.200.0.0/24"
+      subnet    = "scaletest"
     }
     europe = {
       subdomain = "${var.name}-europe-scaletest"
       url       = "http://${var.name}-europe-scaletest.${var.cloudflare_domain}"
       region    = "europe-west1"
       zone      = "europe-west1-b"
-      cidr      = "10.201.0.0/24"
+      subnet    = "scaletest"
     }
     asia = {
       subdomain = "${var.name}-asia-scaletest"
       url       = "http://${var.name}-asia-scaletest.${var.cloudflare_domain}"
       region    = "asia-southeast1"
       zone      = "asia-southeast1-a"
-      cidr      = "10.202.0.0/24"
+      subnet    = "scaletest"
     }
   }
   node_pools = {
@@ -72,8 +72,8 @@ resource "google_container_cluster" "cluster" {
   name                      = "${var.name}-${each.key}"
   location                  = each.value.zone
   project                   = var.project_id
-  network                   = google_compute_network.vpc.name
-  subnetwork                = google_compute_subnetwork.subnet[each.key].name
+  network                   = local.vpc_name
+  subnetwork                = local.subnet_name
   networking_mode           = "VPC_NATIVE"
   default_max_pods_per_node = 256
   ip_allocation_policy { # Required with networking_mode=VPC_NATIVE
diff --git a/scaletest/terraform/action/gcp_db.tf b/scaletest/terraform/action/gcp_db.tf
index 0443fa771fe65..318c1d027f64f 100644
--- a/scaletest/terraform/action/gcp_db.tf
+++ b/scaletest/terraform/action/gcp_db.tf
@@ -23,7 +23,7 @@ resource "google_sql_database_instance" "db" {
 
     ip_configuration {
       ipv4_enabled    = false
-      private_network = google_compute_network.vpc.id
+      private_network = "projects/${var.project_id}/global/networks/${local.vpc_name}"
     }
 
     insights_config {
diff --git a/scaletest/terraform/action/gcp_vpc.tf b/scaletest/terraform/action/gcp_vpc.tf
index c9fd412aa3cb4..4bd29a14008e3 100644
--- a/scaletest/terraform/action/gcp_vpc.tf
+++ b/scaletest/terraform/action/gcp_vpc.tf
@@ -1,20 +1,6 @@
-
-resource "google_compute_network" "vpc" {
-  project                 = var.project_id
-  name                    = var.name
-  auto_create_subnetworks = "false"
-  depends_on = [
-    google_project_service.api["compute.googleapis.com"]
-  ]
-}
-
-resource "google_compute_subnetwork" "subnet" {
-  for_each      = local.deployments
-  name          = "${var.name}-${each.key}"
-  project       = var.project_id
-  region        = each.value.region
-  network       = google_compute_network.vpc.name
-  ip_cidr_range = each.value.cidr
+locals {
+  vpc_name = "scaletest"
+  subnet_name = "scaletest"
 }
 
 resource "google_compute_address" "coder" {
@@ -32,11 +18,11 @@ resource "google_compute_global_address" "sql_peering" {
   purpose       = "VPC_PEERING"
   address_type  = "INTERNAL"
   prefix_length = 16
-  network       = google_compute_network.vpc.id
+  network       = local.vpc_name
 }
 
 resource "google_service_networking_connection" "private_vpc_connection" {
-  network                 = google_compute_network.vpc.id
+  network                 = local.vpc_name
   service                 = "servicenetworking.googleapis.com"
   reserved_peering_ranges = [google_compute_global_address.sql_peering.name]
 }
diff --git a/scaletest/terraform/action/prometheus.tf b/scaletest/terraform/action/prometheus.tf
new file mode 100644
index 0000000000000..4b65230282626
--- /dev/null
+++ b/scaletest/terraform/action/prometheus.tf
@@ -0,0 +1,78 @@
+locals {
+  prometheus_helm_repo             = "https://charts.bitnami.com/bitnami"
+  prometheus_helm_chart            = "kube-prometheus"
+  prometheus_exporter_helm_repo    = "https://prometheus-community.github.io/helm-charts"
+  prometheus_exporter_helm_chart   = "prometheus-postgres-exporter"
+  prometheus_release_name          = "prometheus"
+  prometheus_exporter_release_name = "prometheus-postgres-exporter"
+  prometheus_namespace             = "prometheus"
+  prometheus_remote_write_send_interval = "15s"
+  prometheus_remote_write_metrics_regex = ".*"
+}
+
+resource "kubernetes_namespace" "prometheus_namespace_primary" {
+  provider = kubernetes.primary
+
+  metadata {
+    name = local.prometheus_namespace
+  }
+  lifecycle {
+    ignore_changes = [timeouts, wait_for_default_service_account]
+  }
+}
+
+resource "helm_release" "prometheus_chart_primary" {
+  provider = helm.primary
+
+  repository = local.prometheus_helm_repo
+  chart      = local.prometheus_helm_chart
+  name       = local.prometheus_release_name
+  namespace  = kubernetes_namespace.prometheus_namespace_primary.metadata.0.name
+  values = [<<EOF
+alertmanager:
+  enabled: false
+blackboxExporter:
+  affinity:
+    nodeAffinity:
+      requiredDuringSchedulingIgnoredDuringExecution:
+        nodeSelectorTerms:
+        - matchExpressions:
+          - key: "cloud.google.com/gke-nodepool"
+            operator: "In"
+            values: ["${google_container_node_pool.node_pool["primary_misc"].name}"]
+operator:
+  affinity:
+    nodeAffinity:
+      requiredDuringSchedulingIgnoredDuringExecution:
+        nodeSelectorTerms:
+        - matchExpressions:
+          - key: "cloud.google.com/gke-nodepool"
+            operator: "In"
+            values: ["${google_container_node_pool.node_pool["primary_misc"].name}"]
+prometheus:
+  affinity:
+    nodeAffinity:
+      requiredDuringSchedulingIgnoredDuringExecution:
+        nodeSelectorTerms:
+        - matchExpressions:
+          - key: "cloud.google.com/gke-nodepool"
+            operator: "In"
+            values: ["${google_container_node_pool.node_pool["primary_misc"].name}"]
+  externalLabels:
+    cluster: "primary"
+  persistence:
+    enabled: true
+    storageClass: standard
+  remoteWrite:
+    - url: "${var.prometheus_remote_write_url}"
+      tlsConfig:
+        insecureSkipVerify: true
+      writeRelabelConfigs:
+        - sourceLabels: [__name__]
+          regex: "${local.prometheus_remote_write_metrics_regex}"
+          action: keep
+      metadataConfig:
+        sendInterval: "${local.prometheus_remote_write_send_interval}"
+  EOF
+  ]
+}
diff --git a/scaletest/terraform/action/vars.tf b/scaletest/terraform/action/vars.tf
index 264110e239845..6788e843d8b6f 100644
--- a/scaletest/terraform/action/vars.tf
+++ b/scaletest/terraform/action/vars.tf
@@ -85,3 +85,8 @@ variable "provisionerd_image_tag" {
   description = "Tag to use for Provisionerd image."
   default     = "latest"
 }
+
+// Prometheus
+variable "prometheus_remote_write_url" {
+  description = "URL to push prometheus metrics to."
+}
diff --git a/scaletest/terraform/infra/gcp_db.tf b/scaletest/terraform/infra/gcp_db.tf
index 4d13b262c615f..2de7dd1ab2770 100644
--- a/scaletest/terraform/infra/gcp_db.tf
+++ b/scaletest/terraform/infra/gcp_db.tf
@@ -22,7 +22,7 @@ resource "google_sql_database_instance" "db" {
 
     ip_configuration {
       ipv4_enabled    = false
-      private_network = google_compute_network.vpc.id
+      private_network = "scaletest"
     }
 
     insights_config {
diff --git a/scaletest/terraform/infra/gcp_vpc.tf b/scaletest/terraform/infra/gcp_vpc.tf
index b125c60cfd25a..63d63403b3ce5 100644
--- a/scaletest/terraform/infra/gcp_vpc.tf
+++ b/scaletest/terraform/infra/gcp_vpc.tf
@@ -21,7 +21,7 @@ resource "google_compute_global_address" "sql_peering" {
   purpose       = "VPC_PEERING"
   address_type  = "INTERNAL"
   prefix_length = 16
-  network       = google_compute_network.vpc.id
+  network       = "scaletest"
 }
 
 resource "google_compute_address" "coder" {
@@ -33,7 +33,7 @@ resource "google_compute_address" "coder" {
 }
 
 resource "google_service_networking_connection" "private_vpc_connection" {
-  network                 = google_compute_network.vpc.id
+  network                 = "scaletest"
   service                 = "servicenetworking.googleapis.com"
   reserved_peering_ranges = [google_compute_global_address.sql_peering.name]
 }

From 9bfd90cc90e8bfbc7e41ce5e3fba454501082626 Mon Sep 17 00:00:00 2001
From: Garrett Delfosse <garrett@coder.com>
Date: Wed, 22 Jan 2025 23:51:42 +0000
Subject: [PATCH 2/5] use bitnami chart

---
 scaletest/terraform/action/coder_traffic.tf   |   3 +
 scaletest/terraform/action/gcp_db.tf          |   2 +-
 scaletest/terraform/action/gcp_vpc.tf         |   3 +-
 scaletest/terraform/action/prometheus.tf      | 176 +++++++++++++-----
 .../action/prometheus_helm_values.tftpl       |  44 +++++
 5 files changed, 175 insertions(+), 53 deletions(-)
 create mode 100644 scaletest/terraform/action/prometheus_helm_values.tftpl

diff --git a/scaletest/terraform/action/coder_traffic.tf b/scaletest/terraform/action/coder_traffic.tf
index ebc809273c598..f51661b7edfde 100644
--- a/scaletest/terraform/action/coder_traffic.tf
+++ b/scaletest/terraform/action/coder_traffic.tf
@@ -31,6 +31,9 @@ resource "time_sleep" "wait_baseline" {
     kubernetes_job.create_workspaces_primary,
     kubernetes_job.create_workspaces_europe,
     kubernetes_job.create_workspaces_asia,
+    helm_release.prometheus_chart_primary,
+    helm_release.prometheus_chart_europe,
+    helm_release.prometheus_chart_asia,
   ]
 
   create_duration = local.wait_baseline_duration
diff --git a/scaletest/terraform/action/gcp_db.tf b/scaletest/terraform/action/gcp_db.tf
index 318c1d027f64f..9eb17464e1ce9 100644
--- a/scaletest/terraform/action/gcp_db.tf
+++ b/scaletest/terraform/action/gcp_db.tf
@@ -23,7 +23,7 @@ resource "google_sql_database_instance" "db" {
 
     ip_configuration {
       ipv4_enabled    = false
-      private_network = "projects/${var.project_id}/global/networks/${local.vpc_name}"
+      private_network = local.vpc_id
     }
 
     insights_config {
diff --git a/scaletest/terraform/action/gcp_vpc.tf b/scaletest/terraform/action/gcp_vpc.tf
index 4bd29a14008e3..d0b7c36760c18 100644
--- a/scaletest/terraform/action/gcp_vpc.tf
+++ b/scaletest/terraform/action/gcp_vpc.tf
@@ -1,5 +1,6 @@
 locals {
   vpc_name = "scaletest"
+  vpc_id = "projects/${var.project_id}/global/networks/${local.vpc_name}"
   subnet_name = "scaletest"
 }
 
@@ -22,7 +23,7 @@ resource "google_compute_global_address" "sql_peering" {
 }
 
 resource "google_service_networking_connection" "private_vpc_connection" {
-  network                 = local.vpc_name
+  network                 = local.vpc_id
   service                 = "servicenetworking.googleapis.com"
   reserved_peering_ranges = [google_compute_global_address.sql_peering.name]
 }
diff --git a/scaletest/terraform/action/prometheus.tf b/scaletest/terraform/action/prometheus.tf
index 4b65230282626..d7c94f9fb384c 100644
--- a/scaletest/terraform/action/prometheus.tf
+++ b/scaletest/terraform/action/prometheus.tf
@@ -1,10 +1,7 @@
 locals {
-  prometheus_helm_repo             = "https://charts.bitnami.com/bitnami"
+  prometheus_helm_repo             = "oci://registry-1.docker.io/bitnamicharts"
   prometheus_helm_chart            = "kube-prometheus"
-  prometheus_exporter_helm_repo    = "https://prometheus-community.github.io/helm-charts"
-  prometheus_exporter_helm_chart   = "prometheus-postgres-exporter"
   prometheus_release_name          = "prometheus"
-  prometheus_exporter_release_name = "prometheus-postgres-exporter"
   prometheus_namespace             = "prometheus"
   prometheus_remote_write_send_interval = "15s"
   prometheus_remote_write_metrics_regex = ".*"
@@ -28,51 +25,128 @@ resource "helm_release" "prometheus_chart_primary" {
   chart      = local.prometheus_helm_chart
   name       = local.prometheus_release_name
   namespace  = kubernetes_namespace.prometheus_namespace_primary.metadata.0.name
-  values = [<<EOF
-alertmanager:
-  enabled: false
-blackboxExporter:
-  affinity:
-    nodeAffinity:
-      requiredDuringSchedulingIgnoredDuringExecution:
-        nodeSelectorTerms:
-        - matchExpressions:
-          - key: "cloud.google.com/gke-nodepool"
-            operator: "In"
-            values: ["${google_container_node_pool.node_pool["primary_misc"].name}"]
-operator:
-  affinity:
-    nodeAffinity:
-      requiredDuringSchedulingIgnoredDuringExecution:
-        nodeSelectorTerms:
-        - matchExpressions:
-          - key: "cloud.google.com/gke-nodepool"
-            operator: "In"
-            values: ["${google_container_node_pool.node_pool["primary_misc"].name}"]
-prometheus:
-  affinity:
-    nodeAffinity:
-      requiredDuringSchedulingIgnoredDuringExecution:
-        nodeSelectorTerms:
-        - matchExpressions:
-          - key: "cloud.google.com/gke-nodepool"
-            operator: "In"
-            values: ["${google_container_node_pool.node_pool["primary_misc"].name}"]
-  externalLabels:
-    cluster: "primary"
-  persistence:
-    enabled: true
-    storageClass: standard
-  remoteWrite:
-    - url: "${var.prometheus_remote_write_url}"
-      tlsConfig:
-        insecureSkipVerify: true
-      writeRelabelConfigs:
-        - sourceLabels: [__name__]
-          regex: "${local.prometheus_remote_write_metrics_regex}"
-          action: keep
-      metadataConfig:
-        sendInterval: "${local.prometheus_remote_write_send_interval}"
-  EOF
-  ]
+  values = [templatefile("${path.module}/prometheus_helm_values.tftpl", {
+    nodepool = google_container_node_pool.node_pool["primary_misc"].name,
+    cluster = "primary",
+    prometheus_remote_write_url = var.prometheus_remote_write_url,
+    prometheus_remote_write_metrics_regex = local.prometheus_remote_write_metrics_regex,
+    prometheus_remote_write_send_interval = local.prometheus_remote_write_send_interval,
+  })]
+}
+
+resource "kubectl_manifest" "pod_monitor_primary" {
+  provider = kubectl.primary
+
+  yaml_body = <<YAML
+apiVersion: monitoring.coreos.com/v1
+kind: PodMonitor
+metadata:
+  namespace: ${kubernetes_namespace.coder_primary.metadata.0.name}
+  name: coder-monitoring
+spec:
+  selector:
+    matchLabels:
+      "app.kubernetes.io/name": coder
+  podMetricsEndpoints:
+    - port: prometheus-http
+      interval: 30s
+YAML
+
+  depends_on = [ helm_release.prometheus_chart_primary ]
+}
+
+resource "kubernetes_namespace" "prometheus_namespace_europe" {
+  provider = kubernetes.europe
+
+  metadata {
+    name = local.prometheus_namespace
+  }
+  lifecycle {
+    ignore_changes = [timeouts, wait_for_default_service_account]
+  }
+}
+
+resource "helm_release" "prometheus_chart_europe" {
+  provider = helm.europe
+
+  repository = local.prometheus_helm_repo
+  chart      = local.prometheus_helm_chart
+  name       = local.prometheus_release_name
+  namespace  = kubernetes_namespace.prometheus_namespace_europe.metadata.0.name
+  values = [templatefile("${path.module}/prometheus_helm_values.tftpl", {
+    nodepool = google_container_node_pool.node_pool["europe_misc"].name,
+    cluster = "europe",
+    prometheus_remote_write_url = var.prometheus_remote_write_url,
+    prometheus_remote_write_metrics_regex = local.prometheus_remote_write_metrics_regex,
+    prometheus_remote_write_send_interval = local.prometheus_remote_write_send_interval,
+  })]
+}
+
+resource "kubectl_manifest" "pod_monitor_europe" {
+  provider = kubectl.europe
+
+  yaml_body = <<YAML
+apiVersion: monitoring.coreos.com/v1
+kind: PodMonitor
+metadata:
+  namespace: ${kubernetes_namespace.coder_europe.metadata.0.name}
+  name: coder-monitoring
+spec:
+  selector:
+    matchLabels:
+      "app.kubernetes.io/name": coder
+  podMetricsEndpoints:
+    - port: prometheus-http
+      interval: 30s
+YAML
+
+  depends_on = [ helm_release.prometheus_chart_europe ]
+}
+
+resource "kubernetes_namespace" "prometheus_namespace_asia" {
+  provider = kubernetes.asia
+
+  metadata {
+    name = local.prometheus_namespace
+  }
+  lifecycle {
+    ignore_changes = [timeouts, wait_for_default_service_account]
+  }
+}
+
+resource "helm_release" "prometheus_chart_asia" {
+  provider = helm.asia
+
+  repository = local.prometheus_helm_repo
+  chart      = local.prometheus_helm_chart
+  name       = local.prometheus_release_name
+  namespace  = kubernetes_namespace.prometheus_namespace_asia.metadata.0.name
+  values = [templatefile("${path.module}/prometheus_helm_values.tftpl", {
+    nodepool = google_container_node_pool.node_pool["asia_misc"].name,
+    cluster = "asia",
+    prometheus_remote_write_url = var.prometheus_remote_write_url,
+    prometheus_remote_write_metrics_regex = local.prometheus_remote_write_metrics_regex,
+    prometheus_remote_write_send_interval = local.prometheus_remote_write_send_interval,
+  })]
+}
+
+resource "kubectl_manifest" "pod_monitor_asia" {
+  provider = kubectl.asia
+
+  yaml_body = <<YAML
+apiVersion: monitoring.coreos.com/v1
+kind: PodMonitor
+metadata:
+  namespace: ${kubernetes_namespace.coder_asia.metadata.0.name}
+  name: coder-monitoring
+spec:
+  selector:
+    matchLabels:
+      "app.kubernetes.io/name": coder
+  podMetricsEndpoints:
+    - port: prometheus-http
+      interval: 30s
+YAML
+
+  depends_on = [ helm_release.prometheus_chart_asia ]
 }
diff --git a/scaletest/terraform/action/prometheus_helm_values.tftpl b/scaletest/terraform/action/prometheus_helm_values.tftpl
new file mode 100644
index 0000000000000..4729aa5d7103d
--- /dev/null
+++ b/scaletest/terraform/action/prometheus_helm_values.tftpl
@@ -0,0 +1,44 @@
+alertmanager:
+  enabled: false
+blackboxExporter:
+  affinity:
+    nodeAffinity:
+      requiredDuringSchedulingIgnoredDuringExecution:
+        nodeSelectorTerms:
+        - matchExpressions:
+          - key: "cloud.google.com/gke-nodepool"
+            operator: "In"
+            values: ["${nodepool}"]
+operator:
+  affinity:
+    nodeAffinity:
+      requiredDuringSchedulingIgnoredDuringExecution:
+        nodeSelectorTerms:
+        - matchExpressions:
+          - key: "cloud.google.com/gke-nodepool"
+            operator: "In"
+            values: ["${nodepool}"]
+prometheus:
+  affinity:
+    nodeAffinity:
+      requiredDuringSchedulingIgnoredDuringExecution:
+        nodeSelectorTerms:
+        - matchExpressions:
+          - key: "cloud.google.com/gke-nodepool"
+            operator: "In"
+            values: ["${nodepool}"]
+  externalLabels:
+    cluster: "${cluster}"
+  persistence:
+    enabled: true
+    storageClass: standard
+  remoteWrite:
+    - url: "${prometheus_remote_write_url}"
+      tlsConfig:
+        insecureSkipVerify: true
+      writeRelabelConfigs:
+        - sourceLabels: [__name__]
+          regex: "${prometheus_remote_write_metrics_regex}"
+          action: keep
+      metadataConfig:
+        sendInterval: "${prometheus_remote_write_send_interval}"

From 371e79c4054922f89d12ef4b9c1021f9ae645698 Mon Sep 17 00:00:00 2001
From: Garrett Delfosse <garrett@coder.com>
Date: Wed, 22 Jan 2025 23:53:55 +0000
Subject: [PATCH 3/5] fix replace

---
 scaletest/terraform/infra/gcp_db.tf  | 2 +-
 scaletest/terraform/infra/gcp_vpc.tf | 4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/scaletest/terraform/infra/gcp_db.tf b/scaletest/terraform/infra/gcp_db.tf
index 2de7dd1ab2770..4d13b262c615f 100644
--- a/scaletest/terraform/infra/gcp_db.tf
+++ b/scaletest/terraform/infra/gcp_db.tf
@@ -22,7 +22,7 @@ resource "google_sql_database_instance" "db" {
 
     ip_configuration {
       ipv4_enabled    = false
-      private_network = "scaletest"
+      private_network = google_compute_network.vpc.id
     }
 
     insights_config {
diff --git a/scaletest/terraform/infra/gcp_vpc.tf b/scaletest/terraform/infra/gcp_vpc.tf
index 63d63403b3ce5..b125c60cfd25a 100644
--- a/scaletest/terraform/infra/gcp_vpc.tf
+++ b/scaletest/terraform/infra/gcp_vpc.tf
@@ -21,7 +21,7 @@ resource "google_compute_global_address" "sql_peering" {
   purpose       = "VPC_PEERING"
   address_type  = "INTERNAL"
   prefix_length = 16
-  network       = "scaletest"
+  network       = google_compute_network.vpc.id
 }
 
 resource "google_compute_address" "coder" {
@@ -33,7 +33,7 @@ resource "google_compute_address" "coder" {
 }
 
 resource "google_service_networking_connection" "private_vpc_connection" {
-  network                 = "scaletest"
+  network                 = google_compute_network.vpc.id
   service                 = "servicenetworking.googleapis.com"
   reserved_peering_ranges = [google_compute_global_address.sql_peering.name]
 }

From c4937806d9a29a65ee6572a41654e35046ba568d Mon Sep 17 00:00:00 2001
From: Garrett Delfosse <garrett@coder.com>
Date: Wed, 22 Jan 2025 23:54:11 +0000
Subject: [PATCH 4/5] fmt

---
 scaletest/terraform/action/coder_traffic.tf | 14 ++++-----
 scaletest/terraform/action/gcp_vpc.tf       |  4 +--
 scaletest/terraform/action/prometheus.tf    | 32 ++++++++++-----------
 3 files changed, 25 insertions(+), 25 deletions(-)

diff --git a/scaletest/terraform/action/coder_traffic.tf b/scaletest/terraform/action/coder_traffic.tf
index f51661b7edfde..b477f3847a6d6 100644
--- a/scaletest/terraform/action/coder_traffic.tf
+++ b/scaletest/terraform/action/coder_traffic.tf
@@ -5,20 +5,20 @@ locals {
 
   traffic_types = {
     ssh = {
-      duration      = "30m"
-      job_timeout   = "35m"
+      duration    = "30m"
+      job_timeout = "35m"
       flags = [
         "--ssh",
       ]
     }
     webterminal = {
-      duration      = "25m"
-      job_timeout   = "30m"
-      flags         = []
+      duration    = "25m"
+      job_timeout = "30m"
+      flags       = []
     }
     app = {
-      duration      = "20m"
-      job_timeout   = "25m"
+      duration    = "20m"
+      job_timeout = "25m"
       flags = [
         "--app=wsec",
       ]
diff --git a/scaletest/terraform/action/gcp_vpc.tf b/scaletest/terraform/action/gcp_vpc.tf
index d0b7c36760c18..10624edaddf91 100644
--- a/scaletest/terraform/action/gcp_vpc.tf
+++ b/scaletest/terraform/action/gcp_vpc.tf
@@ -1,6 +1,6 @@
 locals {
-  vpc_name = "scaletest"
-  vpc_id = "projects/${var.project_id}/global/networks/${local.vpc_name}"
+  vpc_name    = "scaletest"
+  vpc_id      = "projects/${var.project_id}/global/networks/${local.vpc_name}"
   subnet_name = "scaletest"
 }
 
diff --git a/scaletest/terraform/action/prometheus.tf b/scaletest/terraform/action/prometheus.tf
index d7c94f9fb384c..8de2b7e90fda4 100644
--- a/scaletest/terraform/action/prometheus.tf
+++ b/scaletest/terraform/action/prometheus.tf
@@ -1,8 +1,8 @@
 locals {
-  prometheus_helm_repo             = "oci://registry-1.docker.io/bitnamicharts"
-  prometheus_helm_chart            = "kube-prometheus"
-  prometheus_release_name          = "prometheus"
-  prometheus_namespace             = "prometheus"
+  prometheus_helm_repo                  = "oci://registry-1.docker.io/bitnamicharts"
+  prometheus_helm_chart                 = "kube-prometheus"
+  prometheus_release_name               = "prometheus"
+  prometheus_namespace                  = "prometheus"
   prometheus_remote_write_send_interval = "15s"
   prometheus_remote_write_metrics_regex = ".*"
 }
@@ -26,9 +26,9 @@ resource "helm_release" "prometheus_chart_primary" {
   name       = local.prometheus_release_name
   namespace  = kubernetes_namespace.prometheus_namespace_primary.metadata.0.name
   values = [templatefile("${path.module}/prometheus_helm_values.tftpl", {
-    nodepool = google_container_node_pool.node_pool["primary_misc"].name,
-    cluster = "primary",
-    prometheus_remote_write_url = var.prometheus_remote_write_url,
+    nodepool                              = google_container_node_pool.node_pool["primary_misc"].name,
+    cluster                               = "primary",
+    prometheus_remote_write_url           = var.prometheus_remote_write_url,
     prometheus_remote_write_metrics_regex = local.prometheus_remote_write_metrics_regex,
     prometheus_remote_write_send_interval = local.prometheus_remote_write_send_interval,
   })]
@@ -52,7 +52,7 @@ spec:
       interval: 30s
 YAML
 
-  depends_on = [ helm_release.prometheus_chart_primary ]
+  depends_on = [helm_release.prometheus_chart_primary]
 }
 
 resource "kubernetes_namespace" "prometheus_namespace_europe" {
@@ -74,9 +74,9 @@ resource "helm_release" "prometheus_chart_europe" {
   name       = local.prometheus_release_name
   namespace  = kubernetes_namespace.prometheus_namespace_europe.metadata.0.name
   values = [templatefile("${path.module}/prometheus_helm_values.tftpl", {
-    nodepool = google_container_node_pool.node_pool["europe_misc"].name,
-    cluster = "europe",
-    prometheus_remote_write_url = var.prometheus_remote_write_url,
+    nodepool                              = google_container_node_pool.node_pool["europe_misc"].name,
+    cluster                               = "europe",
+    prometheus_remote_write_url           = var.prometheus_remote_write_url,
     prometheus_remote_write_metrics_regex = local.prometheus_remote_write_metrics_regex,
     prometheus_remote_write_send_interval = local.prometheus_remote_write_send_interval,
   })]
@@ -100,7 +100,7 @@ spec:
       interval: 30s
 YAML
 
-  depends_on = [ helm_release.prometheus_chart_europe ]
+  depends_on = [helm_release.prometheus_chart_europe]
 }
 
 resource "kubernetes_namespace" "prometheus_namespace_asia" {
@@ -122,9 +122,9 @@ resource "helm_release" "prometheus_chart_asia" {
   name       = local.prometheus_release_name
   namespace  = kubernetes_namespace.prometheus_namespace_asia.metadata.0.name
   values = [templatefile("${path.module}/prometheus_helm_values.tftpl", {
-    nodepool = google_container_node_pool.node_pool["asia_misc"].name,
-    cluster = "asia",
-    prometheus_remote_write_url = var.prometheus_remote_write_url,
+    nodepool                              = google_container_node_pool.node_pool["asia_misc"].name,
+    cluster                               = "asia",
+    prometheus_remote_write_url           = var.prometheus_remote_write_url,
     prometheus_remote_write_metrics_regex = local.prometheus_remote_write_metrics_regex,
     prometheus_remote_write_send_interval = local.prometheus_remote_write_send_interval,
   })]
@@ -148,5 +148,5 @@ spec:
       interval: 30s
 YAML
 
-  depends_on = [ helm_release.prometheus_chart_asia ]
+  depends_on = [helm_release.prometheus_chart_asia]
 }

From c43d65f7803602e566333fd0546dc13830727417 Mon Sep 17 00:00:00 2001
From: Garrett Delfosse <garrett@coder.com>
Date: Thu, 23 Jan 2025 16:51:52 +0000
Subject: [PATCH 5/5] use community chart

---
 scaletest/terraform/action/prometheus.tf      | 44 +++----------------
 .../action/prometheus_helm_values.tftpl       | 42 +++++++-----------
 2 files changed, 22 insertions(+), 64 deletions(-)

diff --git a/scaletest/terraform/action/prometheus.tf b/scaletest/terraform/action/prometheus.tf
index 8de2b7e90fda4..de22a5c949684 100644
--- a/scaletest/terraform/action/prometheus.tf
+++ b/scaletest/terraform/action/prometheus.tf
@@ -1,30 +1,18 @@
 locals {
-  prometheus_helm_repo                  = "oci://registry-1.docker.io/bitnamicharts"
-  prometheus_helm_chart                 = "kube-prometheus"
+  prometheus_helm_repo                  = "https://prometheus-community.github.io/helm-charts"
+  prometheus_helm_chart                 = "kube-prometheus-stack"
   prometheus_release_name               = "prometheus"
-  prometheus_namespace                  = "prometheus"
   prometheus_remote_write_send_interval = "15s"
   prometheus_remote_write_metrics_regex = ".*"
 }
 
-resource "kubernetes_namespace" "prometheus_namespace_primary" {
-  provider = kubernetes.primary
-
-  metadata {
-    name = local.prometheus_namespace
-  }
-  lifecycle {
-    ignore_changes = [timeouts, wait_for_default_service_account]
-  }
-}
-
 resource "helm_release" "prometheus_chart_primary" {
   provider = helm.primary
 
   repository = local.prometheus_helm_repo
   chart      = local.prometheus_helm_chart
   name       = local.prometheus_release_name
-  namespace  = kubernetes_namespace.prometheus_namespace_primary.metadata.0.name
+  namespace  = kubernetes_namespace.coder_primary.metadata.0.name
   values = [templatefile("${path.module}/prometheus_helm_values.tftpl", {
     nodepool                              = google_container_node_pool.node_pool["primary_misc"].name,
     cluster                               = "primary",
@@ -55,24 +43,13 @@ YAML
   depends_on = [helm_release.prometheus_chart_primary]
 }
 
-resource "kubernetes_namespace" "prometheus_namespace_europe" {
-  provider = kubernetes.europe
-
-  metadata {
-    name = local.prometheus_namespace
-  }
-  lifecycle {
-    ignore_changes = [timeouts, wait_for_default_service_account]
-  }
-}
-
 resource "helm_release" "prometheus_chart_europe" {
   provider = helm.europe
 
   repository = local.prometheus_helm_repo
   chart      = local.prometheus_helm_chart
   name       = local.prometheus_release_name
-  namespace  = kubernetes_namespace.prometheus_namespace_europe.metadata.0.name
+  namespace  = kubernetes_namespace.coder_europe.metadata.0.name
   values = [templatefile("${path.module}/prometheus_helm_values.tftpl", {
     nodepool                              = google_container_node_pool.node_pool["europe_misc"].name,
     cluster                               = "europe",
@@ -103,24 +80,13 @@ YAML
   depends_on = [helm_release.prometheus_chart_europe]
 }
 
-resource "kubernetes_namespace" "prometheus_namespace_asia" {
-  provider = kubernetes.asia
-
-  metadata {
-    name = local.prometheus_namespace
-  }
-  lifecycle {
-    ignore_changes = [timeouts, wait_for_default_service_account]
-  }
-}
-
 resource "helm_release" "prometheus_chart_asia" {
   provider = helm.asia
 
   repository = local.prometheus_helm_repo
   chart      = local.prometheus_helm_chart
   name       = local.prometheus_release_name
-  namespace  = kubernetes_namespace.prometheus_namespace_asia.metadata.0.name
+  namespace  = kubernetes_namespace.coder_asia.metadata.0.name
   values = [templatefile("${path.module}/prometheus_helm_values.tftpl", {
     nodepool                              = google_container_node_pool.node_pool["asia_misc"].name,
     cluster                               = "asia",
diff --git a/scaletest/terraform/action/prometheus_helm_values.tftpl b/scaletest/terraform/action/prometheus_helm_values.tftpl
index 4729aa5d7103d..1db6d1e2c9bef 100644
--- a/scaletest/terraform/action/prometheus_helm_values.tftpl
+++ b/scaletest/terraform/action/prometheus_helm_values.tftpl
@@ -1,15 +1,8 @@
 alertmanager:
   enabled: false
-blackboxExporter:
-  affinity:
-    nodeAffinity:
-      requiredDuringSchedulingIgnoredDuringExecution:
-        nodeSelectorTerms:
-        - matchExpressions:
-          - key: "cloud.google.com/gke-nodepool"
-            operator: "In"
-            values: ["${nodepool}"]
-operator:
+grafana:
+  enabled: false
+prometheusOperator:
   affinity:
     nodeAffinity:
       requiredDuringSchedulingIgnoredDuringExecution:
@@ -27,18 +20,17 @@ prometheus:
           - key: "cloud.google.com/gke-nodepool"
             operator: "In"
             values: ["${nodepool}"]
-  externalLabels:
-    cluster: "${cluster}"
-  persistence:
-    enabled: true
-    storageClass: standard
-  remoteWrite:
-    - url: "${prometheus_remote_write_url}"
-      tlsConfig:
-        insecureSkipVerify: true
-      writeRelabelConfigs:
-        - sourceLabels: [__name__]
-          regex: "${prometheus_remote_write_metrics_regex}"
-          action: keep
-      metadataConfig:
-        sendInterval: "${prometheus_remote_write_send_interval}"
+  prometheusSpec:
+    externalLabels:
+      cluster: "${cluster}"
+    podMonitorSelectorNilUsesHelmValues: false
+    remoteWrite:
+      - url: "${prometheus_remote_write_url}"
+        tlsConfig:
+          insecureSkipVerify: true
+        writeRelabelConfigs:
+          - sourceLabels: [__name__]
+            regex: "${prometheus_remote_write_metrics_regex}"
+            action: keep
+        metadataConfig:
+          sendInterval: "${prometheus_remote_write_send_interval}"