MON-224 CloudSQL Instance Failover Unavailable monitor

cloud/gcp/cloud-sql/instance/README.md

@@ -23,6 +23,7 @@ Creates DataDog monitors with the following checks :
 * CloudSQL Instance Disk Utilization
 * CloudSQL Instance Memory Utilization
 * CloudSQL Instance Memory Utilization Forecast
+* CloudSQL Instance Failover Unavailable
 
 Useful links
 ------------
@@ -47,21 +48,18 @@ Inputs
 | disk_threshold_warning | Disk Utilization in fraction (warning threshold) | string | `0.8` | no |
 | disk_timeframe | Timeframe for the Disk Utilization monitor | string | `last_5m` | no |
 | environment | Architecture environment | string | - | yes |
+| failover_unavailable_message | Custom message for the Failover Unavailable monitor | string | `` | no |
+| failover_unavailable_silenced | Groups to mute for GCP Cloud SQL Failover Unavailable monitor | map | `<map>` | no |
+| failover_unavailable_threshold_critical | Failover Unavailable critical threshold | string | `0` | no |
+| failover_unavailable_timeframe | Timeframe for the Failover Unavailable monitor | string | `last_5m` | no |
 | filter_tags_custom | Tags used for custom filtering when filter_tags_use_defaults is false | string | `*` | no |
 | filter_tags_use_defaults | Use default filter tags convention | string | `true` | no |
 | memory_forecast_history | History for the Memory Utilization Forecast monitor | string | `12h` | no |
-| memory_forecast_history | History for the Memory Utilization Forecast monitor | string | `12h` | no |
-| memory_forecast_interval | Interval for the Memory Utilization Forecast monitor | string | `30m` | no |
 | memory_forecast_interval | Interval for the Memory Utilization Forecast monitor | string | `30m` | no |
 | memory_forecast_message | Custom message for the Memory Utilization Forecast monitor | string | `` | no |
-| memory_forecast_message | Custom message for the Memory Utilization Forecast monitor | string | `` | no |
-| memory_forecast_silenced | Groups to mute for GCP Cloud SQL Memory Utilization Forecast monitor | map | `<map>` | no |
 | memory_forecast_silenced | Groups to mute for GCP Cloud SQL Memory Utilization Forecast monitor | map | `<map>` | no |
 | memory_forecast_threshold_critical | Memory Utilization Forecast in fraction (critical threshold) | string | `0.9` | no |
-| memory_forecast_threshold_critical | Memory Utilization Forecast in fraction (critical threshold) | string | `0.9` | no |
 | memory_forecast_threshold_warning | Memory Utilization Forecast in fraction (warning threshold) | string | `0.8` | no |
-| memory_forecast_threshold_warning | Memory Utilization Forecast in fraction (warning threshold) | string | `0.8` | no |
-| memory_forecast_timeframe | Timeframe for the Memory Utilization Forecast monitor | string | `next_3d` | no |
 | memory_forecast_timeframe | Timeframe for the Memory Utilization Forecast monitor | string | `next_3d` | no |
 | memory_message | Custom message for the Memory Utilization monitor | string | `` | no |
 | memory_silenced | Groups to mute for GCP Cloud SQL Memory Utilization monitor | map | `<map>` | no |

cloud/gcp/cloud-sql/instance/inputs.tf

@@ -166,3 +166,31 @@ variable "memory_forecast_silenced" {
   type        = "map"
   default     = {}
 }
+
+#
+# Failover Unavailable
+#
+
+variable "failover_unavailable_message" {
+  description = "Custom message for the Failover Unavailable monitor"
+  type        = "string"
+  default     = ""
+}
+
+variable "failover_unavailable_timeframe" {
+  description = "Timeframe for the Failover Unavailable monitor"
+  type        = "string"
+  default     = "last_5m"
+}
+
+variable "failover_unavailable_threshold_critical" {
+  description = "Failover Unavailable critical threshold"
+  type        = "string"
+  default     = 0
+}
+
+variable "failover_unavailable_silenced" {
+  description = "Groups to mute for GCP Cloud SQL Failover Unavailable monitor"
+  type        = "map"
+  default     = {}
+}

cloud/gcp/cloud-sql/instance/monitors-cloud-sql-instance.tf

@@ -179,3 +179,43 @@ EOF
     "resource:cloud-sql",
   ]
 }
+
+#
+# Failover Unavailable
+#
+resource "datadog_monitor" "failover_unavailable" {
+  name    = "[${var.environment}] Cloud SQL MySQL Failover Unavailable {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
+  message = "${coalesce(var.failover_unavailable_message, var.message)}"
+
+  type = "metric alert"
+
+  query = <<EOF
+    max(${var.failover_unavailable_timeframe}):
+      avg:gcp.cloudsql.database.available_for_failover{${data.template_file.filter.rendered}}
+      by {database_id}
+      <= ${var.failover_unavailable_threshold_critical}
+EOF
+
+  thresholds {
+    critical = "${var.failover_unavailable_threshold_critical}"
+  }
+
+  include_tags        = true
+  notify_no_data      = true
+  require_full_window = false
+  renotify_interval   = 0
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  evaluation_delay    = "${var.delay}"
+  new_host_delay      = "${var.delay}"
+  silenced            = "${var.failover_unavailable_silenced}"
+
+  tags = [
+    "team:gcp",
+    "provider:gcp",
+    "env:${var.environment}",
+    "resource:cloud-sql",
+  ]
+}