Merged in MON-286-host-unreachable-monitor-on-syst (pull request #134)

MON-286 add host unreachable monitor to system generic

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

system/generic/README.md

@@ -16,6 +16,7 @@ module "datadog-monitors-system-generic" {
 
 Creates DataDog monitors with the following checks:
 
+- Host unreachable
 - CPU usage
 - CPU load 5
 - Free disk space
@@ -81,6 +82,10 @@ Creates DataDog monitors with the following checks:
 | free_memory_timeframe | Monitor timeframe for Free memory [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
 | message | Message sent when an alert is triggered | string | - | yes |
 | new_host_delay | Delay in seconds before monitor new resource | string | `300` | no |
+| unreachable_extra_tags | Extra tags for Host unreachable monitor | list | `<list>` | no |
+| unreachable_message | Custom message for Host unreachable monitor | string | `` | no |
+| unreachable_no_data_timeframe | Timeframe for Host unreachable monitor to alert on no data | string | `2` | no |
+| unreachable_silenced | Groups to mute for Host unreachable monitor | map | `<map>` | no |
 
 ## Outputs
 
@@ -91,6 +96,7 @@ Creates DataDog monitors with the following checks:
 | datadog_free_disk_space_inodes_too_low_id | id for monitor datadog_free_disk_space_inodes_too_low |
 | datadog_free_disk_space_too_low_id | id for monitor datadog_free_disk_space_too_low |
 | datadog_free_memory_id | id for monitor datadog_free_memory |
+| datadog_host_unreachable_id | id for monitor datadog_host_unreachable |
 | datadog_load_too_high_id | id for monitor datadog_load_too_high |
 
 ## Related documentation

system/generic/inputs.tf

@@ -29,7 +29,31 @@ variable "filter_tags_custom" {
   default     = "*"
 }
 
-# Custom CPU instance specific
+# System generic specific
+
+variable "unreachable_silenced" {
+  description = "Groups to mute for Host unreachable monitor"
+  type        = "map"
+  default     = {}
+}
+
+variable "unreachable_extra_tags" {
+  description = "Extra tags for Host unreachable monitor"
+  type        = "list"
+  default     = []
+}
+
+variable "unreachable_message" {
+  description = "Custom message for Host unreachable monitor"
+  type        = "string"
+  default     = ""
+}
+
+variable "unreachable_no_data_timeframe" {
+  description = "Timeframe for Host unreachable monitor to alert on no data"
+  type        = "string"
+  default     = 2
+}
 
 variable "cpu_high_silenced" {
   description = "Groups to mute for CPU high monitor"

system/generic/modules.tf

@@ -2,7 +2,7 @@ module "filter-tags" {
   source = "../../common/filter-tags"
 
   environment              = "${var.environment}"
-  resource                 = "generic"
+  resource                 = "system"
   filter_tags_use_defaults = "${var.filter_tags_use_defaults}"
   filter_tags_custom       = "${var.filter_tags_custom}"
 }
@@ -11,7 +11,7 @@ module "filter-tags-disk" {
   source = "../../common/filter-tags"
 
   environment              = "${var.environment}"
-  resource                 = "generic"
+  resource                 = "system"
   filter_tags_use_defaults = "${var.filter_tags_use_defaults}"
   filter_tags_custom       = "${var.filter_tags_custom}"
   extra_tags               = ["dd_disk:enabled"]

system/generic/monitors-system.tf

@@ -1,3 +1,31 @@
+resource "datadog_monitor" "datadog_host_unreachable" {
+  name    = "[${var.environment}] Host unreachable"
+  message = "${coalesce(var.unreachable_message, var.message)}"
+
+  query = "\"datadog.agent.up\".over${module.filter-tags.service_check}.last(6).count_by_status()"
+
+  type = "service check"
+
+  thresholds {
+    ok       = 1
+    warning  = 1
+    critical = 5
+  }
+
+  notify_no_data      = true
+  new_host_delay      = "${var.new_host_delay}"
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = true
+  no_data_timeframe   = "${var.unreachable_no_data_timeframe}"
+
+  silenced = "${var.unreachable_silenced}"
+
+  tags = ["env:${var.environment}", "type:system", "provider:host", "resource:generic", "team:claranet", "created-by:terraform", "${var.unreachable_extra_tags}"]
+}
+
 resource "datadog_monitor" "datadog_cpu_too_high" {
   name    = "[${var.environment}] CPU usage {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
   message = "${coalesce(var.cpu_high_message, var.message)}"

system/generic/outputs.tf

@@ -1,3 +1,8 @@
+output "datadog_host_unreachable_id" {
+  description = "id for monitor datadog_host_unreachable"
+  value       = "${datadog_monitor.datadog_host_unreachable.id}"
+}
+
 output "datadog_cpu_too_high_id" {
   description = "id for monitor datadog_cpu_too_high"
   value       = "${datadog_monitor.datadog_cpu_too_high.id}"