alert when instance count is low (#1472)

* alert when instance count is low * Fix non-pr env logic --------- Co-authored-by: Samuel Aquino <[email protected]> Co-authored-by: halprin <[email protected]> Co-authored-by: James Gilmore <[email protected]> Co-authored-by: jherrflexion <[email protected]>
CDCgov · Oct 23, 2024 · eca4729 · eca4729
1 parent 6ba8ac1
commit eca4729
Show file tree

Hide file tree

Showing 2 changed files with 44 additions and 1 deletion.
diff --git a/operations/template/alert.tf b/operations/template/alert.tf
@@ -161,6 +161,47 @@ resource "azurerm_monitor_metric_alert" "azure_4XX_alert" {
     ]
   }
 }
+resource "azurerm_monitor_metric_alert" "low_instance_count_alert" {
+  count               = local.non_pr_environment ? 1 : 0
+  name                = "cdcti-${var.environment}-azure-low-instance-count-alert"
+  resource_group_name = data.azurerm_resource_group.group.name
+  scopes              = [azurerm_monitor_autoscale_setting.api_autoscale.id]
+  description         = "The instance count in ${var.environment} is too low"
+  severity            = 2       // warning
+  frequency           = "PT1M"  // Checks every 1 minute
+  window_size         = "PT15M" // Every Check, looks back 15 minutes in history
+
+  criteria {
+    metric_namespace = "Microsoft.Insights/autoscalesettings"
+    metric_name      = "ObservedCapacity"
+    aggregation      = "Average"
+    operator         = "LessThanOrEqual"
+    threshold        = azurerm_monitor_autoscale_setting.api_autoscale.profile[0].capacity[0].default - 0.5
+  }
+
+  action {
+    action_group_id = azurerm_monitor_action_group.notify_slack_email[count.index].id
+  }
+
+  lifecycle {
+    # Ignore changes to tags because the CDC sets these automagically
+    ignore_changes = [
+      tags["business_steward"],
+      tags["center"],
+      tags["environment"],
+      tags["escid"],
+      tags["funding_source"],
+      tags["pii_data"],
+      tags["security_compliance"],
+      tags["security_steward"],
+      tags["support_group"],
+      tags["system"],
+      tags["technical_steward"],
+      tags["zone"]
+    ]
+  }
+}
+
 resource "azurerm_monitor_scheduled_query_rules_alert" "ti-log-errors-alert" {
   count               = local.non_pr_environment ? 1 : 0
   name                = "cdcti-${var.environment}-log-errors-alert"

diff --git a/operations/template/main.tf b/operations/template/main.tf
@@ -8,7 +8,9 @@ locals {
   rs_domain_prefix               = "${local.selected_rs_environment_prefix}${length(local.selected_rs_environment_prefix) == 0 ? "" : "."}"
   higher_environment_level       = var.environment == "stg" || var.environment == "prd"
   cdc_domain_environment         = var.environment == "dev" || var.environment == "stg" || var.environment == "prd"
-  non_pr_environment             = !strcontains(var.environment, "pr")
+
+  // If the environment looks like pr123, regexall will contain matches. If there are no matches, it's a non-pr env
+  non_pr_environment = length(regexall("^pr\\d+", var.environment)) == 0
 }
 
 data "azurerm_resource_group" "group" {