MON-185 - Monitor updated with last best practices

cloud/azure/README.md

@@ -152,6 +152,8 @@ Inputs
 | redis_status_silenced | Groups to mute for Redis status monitor | map | `<map>` | no |
 | servicebus_status_message | Custom message for Service Bus status monitor | string | `` | no |
 | servicebus_status_silenced | Groups to mute for Service Bus status monitor | map | `<map>` | no |
+| servicebus_status_timeframe | Monitor timeframe for Service Bus status [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_15m` | no |
+| servicebus_status_aggregator | Monitor timeframe aggregator for Service Bus status [available values: min, max, sum or avg] | string | `min` | no |
 | sqldatabase_cpu_message | Custom message for SQL CPU monitor | string | `` | no |
 | sqldatabase_cpu_silenced | Groups to mute for SQL CPU monitor | map | `<map>` | no |
 | sqldatabase_cpu_threshold_critical | CPU usage in percent (critical threshold) | string | `90` | no |

cloud/azure/inputs.tf

@@ -668,6 +668,17 @@ variable "servicebus_status_message" {
   default     = ""
 }
 
+variable "servicebus_status_aggregator" {
+  description = "Monitor aggregator for Service Bus status [available values: min, max, sum or avg]"
+  type        = "string"
+  default     = "min"
+}
+
+variable "servicebus_status_timeframe" {
+  description = "Monitor timeframe for Service Bus status [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  default     = "last_15m"
+}
+
 # Azure SQL Database specific variables
 variable "sqldatabase_cpu_silenced" {
   description = "Groups to mute for SQL CPU monitor"

cloud/azure/monitors.tf

@@ -178,8 +178,10 @@ module "servicebus" {
   filter_tags_use_defaults = "${var.filter_tags_use_defaults}"
   filter_tags_custom       = "${var.filter_tags_custom}"
 
-  status_silenced = "${var.servicebus_status_silenced}"
+  status_silenced             = "${var.servicebus_status_silenced}"
-  status_message  = "${var.servicebus_status_message}"
+  status_message              = "${var.servicebus_status_message}"
+  status_timeframe            = "${var.servicebus_status_timeframe}"
+  status_timeframe_aggregator = "${var.servicebus_status_aggregator}"
 }
 
 module "sqldatabase" {

cloud/azure/servicebus/README.md

@@ -31,6 +31,8 @@ Inputs
 | message | Message sent when an alert is triggered | string | - | yes |
 | status_message | Custom message for Service Bus status monitor | string | `` | no |
 | status_silenced | Groups to mute for Service Bus status monitor | map | `<map>` | no |
+| status_timeframe | Monitor timeframe for Service Bus status [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_15m` | no |
+| status_aggregator | Monitor aggregator for Service Bus status [available values: min, max, sum or avg] | string | `min` | no |
 
 Related documentation
 ---------------------

cloud/azure/servicebus/inputs.tf

@@ -36,3 +36,14 @@ variable "status_message" {
   type        = "string"
   default     = ""
 }
+
+variable "status_aggregator" {
+  description = "Monitor aggregator for Service Bus status [available values: min, max, sum or avg]"
+  type        = "string"
+  default     = "min"
+}
+
+variable "status_timeframe" {
+  description = "Monitor timeframe for Service Bus status [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  default     = "last_15m"
+}

cloud/azure/servicebus/monitors-service-bus.tf

@@ -11,7 +11,9 @@ resource "datadog_monitor" "servicebus_status" {
   message = "${coalesce(var.status_message, var.message)}"
 
   query = <<EOF
-      min(last_5m): avg:azure.servicebus_namespaces.status{${data.template_file.filter.rendered}} by {resource_group,region,name} != 1
+      ${var.status_aggregator}(${var.status_timeframe}): (
+        ${var.status_aggregator}:azure.servicebus_namespaces.status{${data.template_file.filter.rendered}} by {resource_group,region,name}
+      ) != 1
 EOF
 
   type = "metric alert"