MON-487: Add Datadog monitor VM RAM used

MON-487: Fix variable

MON-487: Fix variable 2

MON-487: Fix inputs

MON-487: Fix query invalid

MON-487: Change naming of variables and update README

MON-487: Fix CI 2

MON-487: Increase timeframe from 5 min to 15min

MON-487: Update README for timeframe increase

cloud/azure/virtual-machine/README.md

@@ -19,6 +19,7 @@ Creates DataDog monitors with the following checks:
 - Virtual Machine CPU usage
 - Virtual Machine credit CPU
 - Virtual Machine is unreachable
+- Virtual Machine RAM reserved
 
 ## Inputs
 
@@ -46,6 +47,13 @@ Creates DataDog monitors with the following checks:
 | message | Message sent when a Redis monitor is triggered | string | n/a | yes |
 | new\_host\_delay | Delay in seconds before monitor new resource | string | `"300"` | no |
 | prefix\_slug | Prefix string to prepend between brackets on every monitors names | string | `""` | no |
+| ram\_reserved\_enabled | Flag to enable Virtual Machine RAM reserved monitor | string | `"true"` | no |
+| ram\_reserved\_extra\_tags | Extra tags for Virtual Machine RAM reserved monitor | list(string) | `[]` | no |
+| ram\_reserved\_message | Custom message for Virtual Machine RAM reserved monitor | string | `""` | no |
+| ram\_reserved\_threshold\_critical | Virtual Machine RAM reserved limit (critical threshold) | string | `"95"` | no |
+| ram\_reserved\_threshold\_warning | Virtual Machine RAM reserved limit (warning threshold) | string | `"90"` | no |
+| ram\_reserved\_time\_aggregator | Monitor aggregator for Virtual Machine RAM reserved [available values: min, max, sum or avg] | string | `"min"` | no |
+| ram\_reserved\_timeframe | Monitor timeframe for Virtual Machine RAM reserved [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `"last_15m"` | no |
 | status\_enabled | Flag to enable Virtual Machine status monitor | string | `"true"` | no |
 | status\_extra\_tags | Extra tags for Virtual Machine status monitor | list(string) | `[]` | no |
 | status\_message | Custom message for Virtual Machine status monitor | string | `""` | no |
@@ -58,6 +66,7 @@ Creates DataDog monitors with the following checks:
 |------|-------------|
 | virtualmachine\_cpu\_usage\_id | id for monitor virtualmachine_cpu_usage |
 | virtualmachine\_credit\_cpu\_remaining\_too\_low\_id | id for monitor virtualmachine_credit_cpu_remaining_too_low |
+| virtualmachine\_ram\_reserved\_id | id for monitor virtualmachine_ram_reserved |
 | virtualmachine\_status\_id | id for monitor virtualmachine_status |
 
 ## Related documentation

cloud/azure/virtual-machine/inputs.tf

@@ -151,3 +151,44 @@ variable "cpu_remaining_rate_threshold_critical" {
   default     = 15
 }
 
+variable "ram_reserved_enabled" {
+  description = "Flag to enable Virtual Machine RAM reserved monitor"
+  type        = string
+  default     = "true"
+}
+
+variable "ram_reserved_message" {
+  description = "Custom message for Virtual Machine RAM reserved monitor"
+  type        = string
+  default     = ""
+}
+
+variable "ram_reserved_extra_tags" {
+  description = "Extra tags for Virtual Machine RAM reserved monitor"
+  type        = list(string)
+  default     = []
+}
+
+variable "ram_reserved_time_aggregator" {
+  description = "Monitor aggregator for Virtual Machine RAM reserved [available values: min, max, sum or avg]"
+  type        = string
+  default     = "min"
+}
+
+variable "ram_reserved_timeframe" {
+  description = "Monitor timeframe for Virtual Machine RAM reserved [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  type        = string
+  default     = "last_15m"
+}
+
+variable "ram_reserved_threshold_warning" {
+  description = "Virtual Machine RAM reserved limit (warning threshold)"
+  default     = 90
+}
+
+variable "ram_reserved_threshold_critical" {
+  description = "Virtual Machine RAM reserved limit (critical threshold)"
+  default     = 95
+}
+
+

cloud/azure/virtual-machine/monitors-virtual-machine.tf

@@ -98,3 +98,39 @@ EOQ
   }
 }
 
+resource "datadog_monitor" "virtualmachine_ram_reserved" {
+  count   = var.ram_reserved_enabled == "true" ? 1 : 0
+  name    = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Virtual Machine RAM reserved {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
+  message = coalesce(var.ram_reserved_message, var.message)
+  type    = "query alert"
+
+  query = <<EOQ
+    ${var.ram_reserved_time_aggregator}(${var.ram_reserved_timeframe}): 
+      avg:azure.vm.memory_committed_bytes${module.filter-tags.query_alert} by {resource_group,region,name} / (
+      avg:azure.vm.memory_committed_bytes${module.filter-tags.query_alert} by {resource_group,region,name} +
+      avg:azure.vm.memory_available_bytes${module.filter-tags.query_alert} by {resource_group,region,name}) * 100
+      > ${var.ram_reserved_threshold_critical}
+EOQ
+
+  thresholds = {
+    critical = var.ram_reserved_threshold_critical
+    warning  = var.ram_reserved_threshold_warning
+  }
+
+  evaluation_delay    = var.evaluation_delay
+  new_host_delay      = var.new_host_delay
+  notify_no_data      = false
+  renotify_interval   = 0
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = false
+
+  tags = concat(["env:${var.environment}", "type:cloud", "provider:azure", "resource:virtualmachine", "team:claranet", "created-by:terraform"], var.ram_reserved_extra_tags)
+
+  lifecycle {
+    ignore_changes = ["silenced"]
+  }
+}
+

cloud/azure/virtual-machine/outputs.tf

@@ -13,3 +13,8 @@ output "virtualmachine_credit_cpu_remaining_too_low_id" {
   value       = datadog_monitor.virtualmachine_credit_cpu_remaining_too_low.*.id
 }
 
+output "virtualmachine_ram_reserved_id" {
+  description = "id for monitor virtualmachine_ram_reserved"
+  value       = datadog_monitor.virtualmachine_ram_reserved.*.id
+}
+