Test Azure outage alert

operations/template/alert.tf

@@ -28,6 +28,48 @@ resource "azurerm_monitor_action_group" "notify_slack_email" {
   }
 }
 
+resource "azurerm_monitor_activity_log_alert" "azure_service_health_alert" {
+  count               = local.non_pr_environment ? 1 : 0
+  name                = "cdcti-${var.environment}-azure-status-alert"
+  location            = data.azurerm_resource_group.group.location
+  resource_group_name = data.azurerm_resource_group.group.name
+  scopes              = [azurerm_container_registry.registry.id]
+
+  criteria {
+    category = "ServiceHealth"
+    levels   = ["Error"]
+    service_health {
+      locations = var.service_health_locations
+      events    = ["Incident"]
+      services  = ["*"]
+    }
+  }
+
+  action {
+    action_group_id = azurerm_monitor_action_group.notify_slack_email[count.index].id
+  }
+
+  description = "Alert service(s) appear to be down"
+  enabled     = true
+
+  lifecycle {
+    ignore_changes = [
+      tags["business_steward"],
+      tags["center"],
+      tags["environment"],
+      tags["escid"],
+      tags["funding_source"],
+      tags["pii_data"],
+      tags["security_compliance"],
+      tags["security_steward"],
+      tags["support_group"],
+      tags["system"],
+      tags["technical_steward"],
+      tags["zone"]
+    ]
+  }
+}
+
 resource "azurerm_monitor_scheduled_query_rules_alert" "database_token_expired_alert" {
   count               = local.non_pr_environment ? 1 : 0
   name                = "cdcti-${var.environment}-api-log-token-alert"

operations/template/variables.tf

@@ -19,3 +19,8 @@ variable "alert_slack_email" {
   nullable  = false
   sensitive = true
 }
+
+variable "service_health_locations" {
+  type    = list(string)
+  default = ["East US"]
+}