Merged in MON-226-gcp-pubsub (pull request #120)

MON-226 monitors for GCP Pub/Sub

Approved-by: Rafael Romero Carmona <rafael.romero.carmona@fr.clara.net>
Approved-by: Quentin Manfroi <quentin.manfroi@yahoo.fr>
Approved-by: Alexandre Gaillet <alexandre.gaillet@fr.clara.net>

README.md

@@ -96,6 +96,7 @@ The `//` is very important, it's a terraform specific syntax used to separate gi
 		- [cloud-sql](https://bitbucket.org/morea/terraform.feature.datadog/src/master/cloud/gcp/cloud-sql/)
 			- [common](https://bitbucket.org/morea/terraform.feature.datadog/src/master/cloud/gcp/cloud-sql/common/)
 			- [mysql](https://bitbucket.org/morea/terraform.feature.datadog/src/master/cloud/gcp/cloud-sql/mysql/)
+		- [pubsub](https://bitbucket.org/morea/terraform.feature.datadog/src/master/cloud/gcp/pubsub/)
 - [common](https://bitbucket.org/morea/terraform.feature.datadog/src/master/common/)
 	- [alerting-message](https://bitbucket.org/morea/terraform.feature.datadog/src/master/common/alerting-message/)
 	- [filter-tags](https://bitbucket.org/morea/terraform.feature.datadog/src/master/common/filter-tags/)

cloud/gcp/pubsub/README.md

@@ -0,0 +1,56 @@
+# CLOUD GCP PUBSUB DataDog monitors
+
+## How to use this module
+
+```
+module "datadog-monitors-cloud-gcp-pubsub" {
+  source = "git::ssh://git@bitbucket.org/morea/terraform.feature.datadog.git//cloud/gcp/pubsub?ref={revision}"
+
+  environment = "${var.environment}"
+  message     = "${module.datadog-message-alerting.alerting-message}"
+}
+
+```
+
+## Purpose
+
+Creates DataDog monitors with the following checks:
+
+- GCP pubsub sending messages operations
+- GCP pubsub sending messages with result unavailable
+
+## Inputs
+
+| Name | Description | Type | Default | Required |
+|------|-------------|:----:|:-----:|:-----:|
+| environment | Architecture environment | string | - | yes |
+| evaluation_delay | Delay in seconds for the metric evaluation | string | `900` | no |
+| filter_tags | Tags used for filtering | string | `*` | no |
+| message | Message sent when a monitor is triggered | string | - | yes |
+| new_host_delay | Delay in seconds for the new host evaluation | string | `300` | no |
+| sending_operations_count_extra_tags | Extra tags for GCP Pub/Sub Sending Operations Count monitor | list | `<list>` | no |
+| sending_operations_count_message | Custom message for the GCP Pub/Sub Sending Operations Count monitor | string | `` | no |
+| sending_operations_count_silenced | Groups to mute for GCP Pub/Sub Sending Operations Count monitor | map | `<map>` | no |
+| sending_operations_count_threshold_critical | Critical threshold for the number of sending operations. | string | `0` | no |
+| sending_operations_count_time_aggregator | Timeframe for the GCP Pub/Sub Sending Operations Count monitor | string | `sum` | no |
+| sending_operations_count_timeframe | Timeframe for the GCP Pub/Sub Sending Operations Count monitor | string | `last_30m` | no |
+| unavailable_sending_operations_count_extra_tags | Extra tags for GCP Pub/Sub Unavailable Sending Operations Count monitor | list | `<list>` | no |
+| unavailable_sending_operations_count_message | Custom message for the GCP Pub/Sub Unavailable Sending Operations Count monitor | string | `` | no |
+| unavailable_sending_operations_count_silenced | Groups to mute for GCP Pub/Sub Unavailable Sending Operations Count monitor | map | `<map>` | no |
+| unavailable_sending_operations_count_threshold_critical | Critical threshold for the number of unavailable sending operations | string | `4` | no |
+| unavailable_sending_operations_count_threshold_warning | Warning threshold for the number of unavailable sending operations | string | `2` | no |
+| unavailable_sending_operations_count_time_aggregator | Timeframe for the GCP Pub/Sub Unavailable Sending Operations Count monitor | string | `sum` | no |
+| unavailable_sending_operations_count_timeframe | Timeframe for the GCP Pub/Sub Unavailable Sending Operations Count monitor | string | `last_10m` | no |
+
+## Outputs
+
+| Name | Description |
+|------|-------------|
+| sending_operations_count_id | id for monitor sending_operations_count |
+| unavailable_sending_operations_count_id | id for monitor unavailable_sending_operations_count |
+
+Related documentation
+------------
+
+* [GCP Pub/Sub Metrics](https://cloud.google.com/monitoring/api/metrics_gcp#gcp-pubsub)
+* [Datadog GCP Pub/Sub integration](https://docs.datadoghq.com/integrations/google_cloud_pubsub/)

cloud/gcp/pubsub/inputs.tf

@@ -0,0 +1,110 @@
+#
+# Datadog global variables
+#
+variable "environment" {
+  description = "Architecture environment"
+  type        = "string"
+}
+
+variable "filter_tags" {
+  description = "Tags used for filtering"
+  default     = "*"
+}
+
+variable "message" {
+  description = "Message sent when a monitor is triggered"
+}
+
+variable "evaluation_delay" {
+  description = "Delay in seconds for the metric evaluation"
+  default     = 900
+}
+
+variable "new_host_delay" {
+  description = "Delay in seconds for the new host evaluation"
+  default     = 300
+}
+
+#
+# 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 for the number of sending operations."
+  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 for the number of unavailable sending operations"
+  type        = "string"
+  default     = 2
+}
+
+variable "unavailable_sending_operations_count_threshold_critical" {
+  description = "Critical threshold for the number of unavailable sending operations"
+  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 +1,70 @@
+#
+# 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{${var.filter_tags}} 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
+  require_full_window = false
+  notify_no_data      = true
+  renotify_interval   = 0
+
+  evaluation_delay = "${var.evaluation_delay}"
+  new_host_delay   = "${var.new_host_delay}"
+
+  silenced = "${var.sending_operations_count_silenced}"
+
+  tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:pubsub", "team:claranet", "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{${var.filter_tags},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 = "${var.unavailable_sending_operations_count_threshold_critical}"
+  }
+
+  notify_audit        = false
+  locked              = false
+  timeout_h           = 0
+  include_tags        = true
+  require_full_window = false
+  notify_no_data      = true
+  renotify_interval   = 0
+
+  evaluation_delay = "${var.evaluation_delay}"
+  new_host_delay   = "${var.new_host_delay}"
+
+  silenced = "${var.unavailable_sending_operations_count_silenced}"
+
+  tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:pubsub", "team:claranet", "created-by:terraform", "${var.unavailable_sending_operations_count_extra_tags}"]
+}

cloud/gcp/pubsub/outputs.tf

@@ -0,0 +1,9 @@
+output "sending_operations_count_id" {
+  description = "id for monitor sending_operations_count"
+  value       = "${datadog_monitor.sending_operations_count.*.id}"
+}
+
+output "unavailable_sending_operations_count_id" {
+  description = "id for monitor unavailable_sending_operations_count"
+  value       = "${datadog_monitor.unavailable_sending_operations_count.*.id}"
+}