MON-226 Basic monitors

cloud/gcp/pubsub/inputs.tf

@@ -0,0 +1,118 @@
+#
+# Datadog global variables
+#
+variable "environment" {
+  description = "Architecture environment"
+  type        = "string"
+}
+
+variable "filter_tags_use_defaults" {
+  description = "Use default filter tags convention"
+  default     = "true"
+}
+
+variable "filter_tags_custom" {
+  description = "Tags used for custom filtering when filter_tags_use_defaults is false"
+  default     = "*"
+}
+
+variable "message" {
+  description = "Message sent when a monitor is triggered"
+}
+
+variable "delay" {
+  description = "Delay in seconds for the metric evaluation"
+  default     = 900
+}
+
+#
+# Filter variables
+#
+variable "project_id" {
+  type        = "string"
+  description = "ID of the GCP Project"
+}
+
+#
+# Sending Operations Count
+#
+variable "sending_operations_count_message" {
+  description = "Custom message for the GCP Pub/Sub Sending Operations Count monitor"
+  type        = "string"
+  default     = ""
+}
+
+variable "sending_operations_count_time_aggregator" {
+  description = "Timeframe for the GCP Pub/Sub Sending Operations Count monitor"
+  type        = "string"
+  default     = "sum"
+}
+
+variable "sending_operations_count_timeframe" {
+  description = "Timeframe for the GCP Pub/Sub Sending Operations Count monitor"
+  type        = "string"
+  default     = "last_30m"
+}
+
+variable "sending_operations_count_threshold_critical" {
+  description = "Critical threshold"
+  type        = "string"
+  default     = 0
+}
+
+variable "sending_operations_count_silenced" {
+  description = "Groups to mute for GCP Pub/Sub Sending Operations Count monitor"
+  type        = "map"
+  default     = {}
+}
+
+variable "sending_operations_count_extra_tags" {
+  description = "Extra tags for GCP Pub/Sub Sending Operations Count monitor"
+  type        = "list"
+  default     = []
+}
+
+#
+# Unavailable Sending Operations Count
+#
+variable "unavailable_sending_operations_count_message" {
+  description = "Custom message for the GCP Pub/Sub Unavailable Sending Operations Count monitor"
+  type        = "string"
+  default     = ""
+}
+
+variable "unavailable_sending_operations_count_time_aggregator" {
+  description = "Timeframe for the GCP Pub/Sub Unavailable Sending Operations Count monitor"
+  type        = "string"
+  default     = "sum"
+}
+
+variable "unavailable_sending_operations_count_timeframe" {
+  description = "Timeframe for the GCP Pub/Sub Unavailable Sending Operations Count monitor"
+  type        = "string"
+  default     = "last_10m"
+}
+
+variable "unavailable_sending_operations_count_threshold_warning" {
+  description = "Warning threshold"
+  type        = "string"
+  default     = 2
+}
+
+variable "unavailable_sending_operations_count_threshold_critical" {
+  description = "Critical threshold"
+  type        = "string"
+  default     = 4
+}
+
+variable "unavailable_sending_operations_count_silenced" {
+  description = "Groups to mute for GCP Pub/Sub Unavailable Sending Operations Count monitor"
+  type        = "map"
+  default     = {}
+}
+
+variable "unavailable_sending_operations_count_extra_tags" {
+  description = "Extra tags for GCP Pub/Sub Unavailable Sending Operations Count monitor"
+  type        = "list"
+  default     = []
+}

cloud/gcp/pubsub/monitors-pubsub.tf

@@ -0,0 +1,99 @@
+#
+# FILTER
+#
+data "template_file" "filter" {
+  template = "$${filter}"
+
+  vars {
+    filter = "${var.filter_tags_use_defaults == "true" ?
+              format("project_id:%s", var.project_id) :
+             "${var.filter_tags_custom}"}"
+  }
+}
+
+#
+# Sending Operations Count
+#
+resource "datadog_monitor" "sending_operations_count" {
+  name    = "[${var.environment}] GCP pubsub sending messages operations {{#is_alert}}{{{comparator}}} {{threshold}} ({{value}}){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}} ({{value}}){{/is_warning}}"
+  message = "${coalesce(var.sending_operations_count_message, var.message)}"
+
+  type = "query alert"
+
+  query = <<EOF
+  ${var.sending_operations_count_time_aggregator}(${var.sending_operations_count_timeframe}):
+    default(avg:gcp.pubsub.topic.send_message_operation_count{${data.template_file.filter.rendered}} by {topic_id}.as_count(), 0)
+    <= ${var.sending_operations_count_threshold_critical}
+EOF
+
+  thresholds {
+    critical = "${var.sending_operations_count_threshold_critical}"
+  }
+
+  notify_audit        = false
+  locked              = false
+  timeout_h           = 0
+  include_tags        = true
+  no_data_timeframe   = 45
+  require_full_window = false
+  notify_no_data      = true
+  renotify_interval   = 0
+
+  evaluation_delay = "${var.delay}"
+  new_host_delay   = "${var.delay}"
+
+  silenced = "${var.sending_operations_count_silenced}"
+
+  tags = [
+    "team:gcp",
+    "provider:gcp",
+    "resource:pubsub",
+    "env:${var.environment}",
+    "created_by:terraform",
+    "${var.sending_operations_count_extra_tags}",
+  ]
+}
+
+#
+# Unavailable Sending Operations Count
+#
+resource "datadog_monitor" "unavailable_sending_operations_count" {
+  name    = "[${var.environment}] GCP pubsub sending messages with result unavailable {{#is_alert}}{{{comparator}}} {{threshold}} ({{value}}){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}} ({{value}}){{/is_warning}}"
+  message = "${coalesce(var.unavailable_sending_operations_count_message, var.message)}"
+
+  type = "query alert"
+
+  query = <<EOF
+  ${var.unavailable_sending_operations_count_time_aggregator}(${var.unavailable_sending_operations_count_timeframe}):
+    default(avg:gcp.pubsub.topic.send_message_operation_count{${data.template_file.filter.rendered},response_code:unavailable} by {topic_id}.as_count(), 0) 
+    >= ${var.unavailable_sending_operations_count_threshold_critical}
+EOF
+
+  thresholds {
+    warning  = "${var.unavailable_sending_operations_count_threshold_warning}"
+    critical = "${var.unavailable_sending_operations_count_threshold_critical}"
+  }
+
+  notify_audit        = false
+  locked              = false
+  timeout_h           = 0
+  include_tags        = true
+  no_data_timeframe   = 25
+  require_full_window = false
+  notify_no_data      = true
+  renotify_interval   = 0
+
+  evaluation_delay = "${var.delay}"
+  new_host_delay   = "${var.delay}"
+
+  silenced = "${var.unavailable_sending_operations_count_silenced}"
+
+  tags = [
+    "team:gcp",
+    "provider:gcp",
+    "resource:pubsub",
+    "env:${var.environment}",
+    "created_by:terraform",
+    "${var.unavailable_sending_operations_count_extra_tags}",
+  ]
+}