chore: Bootstrap alarms (#75)

bootstrap/main.tf

@@ -23,6 +23,7 @@ module "ogmios_v1_feature" {
   api_key_salt       = var.api_key_salt
   dcu_per_frame      = var.dcu_per_frame
   dns_zone           = var.dns_zone
+  resources          = var.operator_resources
 }
 
 module "ogmios_v1_proxy" {
@@ -76,8 +77,26 @@ module "ogmios_instances" {
   ogmios_image     = each.value.ogmios_image
   node_private_dns = each.value.node_private_dns
   ogmios_version   = each.value.ogmios_version
-  tolerations      = each.value.tolerations
   replicas         = each.value.replicas
+  tolerations = coalesce(each.value.tolerations, [
+    {
+      effect   = "NoSchedule"
+      key      = "demeter.run/compute-profile"
+      operator = "Exists"
+    },
+    {
+      effect   = "NoSchedule"
+      key      = "demeter.run/compute-arch"
+      operator = "Equal"
+      value    = "arm64"
+    },
+    {
+      effect   = "NoSchedule"
+      key      = "demeter.run/availability-sla"
+      operator = "Equal"
+      value    = "consistent"
+    }
+  ])
 }
 
 module "ogmios_services" {
@@ -90,4 +109,8 @@ module "ogmios_services" {
   network        = each.value.network
 }
 
+module "ogmios_monitoring" {
+  source = "./monitoring"
 
+  o11y_datasource_uid = var.o11y_datasource_uid
+}

bootstrap/monitoring/main.tf

@@ -0,0 +1,100 @@
+terraform {
+  required_providers {
+    grafana = {
+      source  = "grafana/grafana"
+      version = ">= 1.28.2"
+    }
+  }
+}
+
+variable "o11y_datasource_uid" {
+  type = string
+}
+
+resource "grafana_folder" "folder" {
+  title = "Ogmios"
+}
+
+resource "grafana_rule_group" "instance_is_down" {
+  name             = "Ogmios is down"
+  folder_uid       = grafana_folder.folder.uid
+  interval_seconds = 60
+  org_id           = 1
+
+  rule {
+    name           = "Ogmios is down"
+    condition      = "B"
+    for            = "5m"
+    no_data_state  = "OK"
+    exec_err_state = "OK"
+    annotations = {
+      description = "We are not receiving more metrics from a particular Ogmios instance.",
+      summary     = "{{ range $k, $v := $values -}}\n{{ if (match \"A[0-9]+\" $k) -}}\nPod: {{ $v.Labels.pod }}\n{{ end }}\n{{ end }}"
+    }
+
+    data {
+      ref_id         = "A"
+      datasource_uid = var.o11y_datasource_uid
+
+      relative_time_range {
+        from = 3600
+        to   = 0
+      }
+
+      model = jsonencode({
+        editorMode    = "code",
+        expr          = "count(avg_over_time(ogmios_connected[10m] offset 1h)) by (pod) unless count(avg_over_time(ogmios_connected[10m])) by (pod)",
+        hide          = false,
+        intervalMs    = 1000,
+        legendFormat  = "__auto",
+        maxDataPoints = 43200,
+        range         = true,
+        refId         = "A"
+      })
+    }
+
+    data {
+      ref_id         = "B"
+      datasource_uid = "-100"
+
+      relative_time_range {
+        from = 3600
+        to   = 0
+      }
+
+      model = jsonencode({
+        conditions = [
+          {
+            evaluator = {
+              params = [0]
+              type   = "gt"
+            },
+            operator = {
+              type = "and"
+            },
+            query = {
+              params : [
+                "A"
+              ]
+            },
+            reducer = {
+              params = [],
+              type   = "count_non_null"
+            },
+            type = "query"
+          }
+        ],
+        datasource = {
+          type = "__expr__",
+          uid  = "-100"
+        },
+        expression    = "A",
+        hide          = false,
+        intervalMs    = 1000,
+        maxDataPoints = 43200,
+        refId         = "B",
+        type          = "classic_conditions"
+      })
+    }
+  }
+}

bootstrap/variables.tf

@@ -32,6 +32,10 @@ variable "versions" {
   default = ["5", "6"]
 }
 
+variable "o11y_datasource_uid" {
+  type = string
+}
+
 // operator settings
 
 variable "operator_image_tag" {
@@ -145,7 +149,6 @@ variable "instances" {
     ogmios_image     = string
     node_private_dns = string
     ogmios_version   = string
-    compute_arch     = string
     replicas         = number
     resources = optional(object({
       limits = object({