Merge pull request #111 from DFE-Digital/1927-allow-metric-frequency-…

…override

Added alert frequency monitor metric

aks/postgres/resources.tf

@@ -10,6 +10,16 @@ locals {
   azure_enable_monitoring     = var.use_azure && var.azure_enable_monitoring
 
   kubernetes_name = "${var.service_name}-${var.environment}-postgres${local.name_suffix}"
+
+  alert_frequency_map = {
+    PT5M  = "PT1M"
+    PT15M = "PT1M"
+    PT30M = "PT1M"
+    PT1H  = "PT1M"
+    PT6H  = "PT5M"
+    PT12H = "PT5M"
+  }
+  alert_frequency = local.alert_frequency_map[var.alert_window_size]
 }
 
 # Username & password
@@ -172,6 +182,7 @@ resource "azurerm_monitor_metric_alert" "memory" {
   scopes              = [azurerm_postgresql_flexible_server.main[0].id]
   description         = "Action will be triggered when memory use is greater than 75%"
   window_size         = var.alert_window_size
+  frequency           = local.alert_frequency
 
   criteria {
     metric_namespace = "Microsoft.DBforPostgreSQL/flexibleServers"
@@ -200,6 +211,7 @@ resource "azurerm_monitor_metric_alert" "cpu" {
   scopes              = [azurerm_postgresql_flexible_server.main[0].id]
   description         = "Action will be triggered when cpu use is greater than ${var.azure_cpu_threshold}%"
   window_size         = var.alert_window_size
+  frequency           = local.alert_frequency
 
   criteria {
     metric_namespace = "Microsoft.DBforPostgreSQL/flexibleServers"
@@ -228,6 +240,7 @@ resource "azurerm_monitor_metric_alert" "storage" {
   scopes              = [azurerm_postgresql_flexible_server.main[0].id]
   description         = "Action will be triggered when storage use is greater than ${var.azure_storage_threshold}%"
   window_size         = var.alert_window_size
+  frequency           = local.alert_frequency
 
   criteria {
     metric_namespace = "Microsoft.DBforPostgreSQL/flexibleServers"

aks/postgres/tfdocs.md

@@ -48,7 +48,7 @@ No modules.
 |------|-------------|------|---------|:--------:|
 | <a name="input_admin_password"></a> [admin\_password](#input\_admin\_password) | Password of the admin user | `string` | `null` | no |
 | <a name="input_admin_username"></a> [admin\_username](#input\_admin\_username) | Username of the admin user | `string` | `null` | no |
-| <a name="input_alert_window_size"></a> [alert\_window\_size](#input\_alert\_window\_size) | The period of time that is used to monitor alert activity e.g PT1M, PT5M, PT15M, PT30M, PT1H, PT6H or PT12H | `string` | `"PT5M"` | no |
+| <a name="input_alert_window_size"></a> [alert\_window\_size](#input\_alert\_window\_size) | The period of time that is used to monitor alert activity e.g. PT1M, PT5M, PT15M, PT30M, PT1H, PT6H, PT12H. The interval between checks is adjusted accordingly. | `string` | `"PT5M"` | no |
 | <a name="input_azure_cpu_threshold"></a> [azure\_cpu\_threshold](#input\_azure\_cpu\_threshold) | n/a | `number` | `80` | no |
 | <a name="input_azure_enable_backup_storage"></a> [azure\_enable\_backup\_storage](#input\_azure\_enable\_backup\_storage) | n/a | `bool` | `true` | no |
 | <a name="input_azure_enable_high_availability"></a> [azure\_enable\_high\_availability](#input\_azure\_enable\_high\_availability) | n/a | `bool` | `false` | no |

aks/postgres/variables.tf

@@ -124,11 +124,17 @@ variable "azure_enable_monitoring" {
   default  = true
 }
 
+
+
 variable "alert_window_size" {
-  type        = string
-  nullable    = false
-  default     = "PT5M"
-  description = "The period of time that is used to monitor alert activity e.g PT1M, PT5M, PT15M, PT30M, PT1H, PT6H or PT12H"
+  type     = string
+  nullable = false
+  default  = "PT5M"
+  validation {
+    condition     = contains(["PT1M", "PT5M", "PT15M", "PT30M", "PT1H", "PT6H", "PT12H"], var.alert_window_size)
+    error_message = "The alert_window_size must be one of: PT1M, PT5M, PT15M, PT30M, PT1H, PT6H, PT12H"
+  }
+  description = "The period of time that is used to monitor alert activity e.g. PT1M, PT5M, PT15M, PT30M, PT1H, PT6H, PT12H. The interval between checks is adjusted accordingly."
 }
 
 variable "azure_maintenance_window" {

aks/redis/resources.tf

@@ -6,6 +6,16 @@ locals {
   azure_enable_monitoring     = var.use_azure && var.azure_enable_monitoring
 
   kubernetes_name = "${var.service_name}-${var.environment}-redis${local.name_suffix}"
+
+  alert_frequency_map = {
+    PT5M  = "PT1M"
+    PT15M = "PT1M"
+    PT30M = "PT1M"
+    PT1H  = "PT1M"
+    PT6H  = "PT5M"
+    PT12H = "PT5M"
+  }
+  alert_frequency = local.alert_frequency_map[var.alert_window_size]
 }
 
 # Azure
@@ -86,6 +96,7 @@ resource "azurerm_monitor_metric_alert" "memory" {
   scopes              = [azurerm_redis_cache.main[0].id]
   description         = "Action will be triggered when memory use is greater than ${var.azure_memory_threshold}%"
   window_size         = var.alert_window_size
+  frequency           = local.alert_frequency
 
   criteria {
     metric_namespace = "Microsoft.Cache/redis"

aks/redis/tfdocs.md

@@ -32,7 +32,7 @@ No modules.
 
 | Name | Description | Type | Default | Required |
 |------|-------------|------|---------|:--------:|
-| <a name="input_alert_window_size"></a> [alert\_window\_size](#input\_alert\_window\_size) | The period of time that is used to monitor alert activity e.g PT1M, PT5M, PT15M, PT30M, PT1H, PT6H or PT12H | `string` | `"PT5M"` | no |
+| <a name="input_alert_window_size"></a> [alert\_window\_size](#input\_alert\_window\_size) | The period of time that is used to monitor alert activity e,g, PT1M, PT5M, PT15M, PT30M, PT1H, PT6H, PT12H. The interval between checks is adjusted accordingly. | `string` | `"PT5M"` | no |
 | <a name="input_azure_capacity"></a> [azure\_capacity](#input\_azure\_capacity) | n/a | `number` | `1` | no |
 | <a name="input_azure_enable_monitoring"></a> [azure\_enable\_monitoring](#input\_azure\_enable\_monitoring) | n/a | `bool` | `true` | no |
 | <a name="input_azure_family"></a> [azure\_family](#input\_azure\_family) | n/a | `string` | `"C"` | no |

aks/redis/variables.tf

@@ -111,8 +111,12 @@ variable "azure_patch_schedule" {
 }
 
 variable "alert_window_size" {
-  type        = string
-  default     = "PT5M"
-  nullable    = false
-  description = "The period of time that is used to monitor alert activity e.g PT1M, PT5M, PT15M, PT30M, PT1H, PT6H or PT12H"
+  type     = string
+  default  = "PT5M"
+  nullable = false
+  validation {
+    condition     = contains(["PT1M", "PT5M", "PT15M", "PT30M", "PT1H", "PT6H", "PT12H"], var.alert_window_size)
+    error_message = "The alert_window_size must be one of: PT1M, PT5M, PT15M, PT30M, PT1H, PT6H, PT12H"
+  }
+  description = "The period of time that is used to monitor alert activity e,g, PT1M, PT5M, PT15M, PT30M, PT1H, PT6H, PT12H. The interval between checks is adjusted accordingly."
 }