Merged in MON-287_fix_service_check_monitors (pull request #127)

MON-287 update service check monitors configuration

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

middleware/apache/README.md

@@ -16,14 +16,15 @@ module "datadog-monitors-middleware-apache" {
 
 Creates DataDog monitors with the following checks:
 
-- Can't connect to apache vhost status
+- Apache vhost status does not respond
 
 ## Inputs
 
 | Name | Description | Type | Default | Required |
 |------|-------------|:----:|:-----:|:-----:|
-| apache_connect_message | Custom message for Apache process monitor | string | `` | no |
-| apache_connect_silenced | Groups to mute for Apache process monitor | map | `<map>` | no |
+| apache_connect_message | Custom message for Apache status monitor | string | `` | no |
+| apache_connect_silenced | Groups to mute for Apache status monitor | map | `<map>` | no |
+| apache_connect_threshold_critical | Apache status monitor (critical threshold) | string | `1.1754943508222875e-38` | no |
 | delay | Delay in seconds for the metric evaluation | string | `15` | no |
 | environment | Architecture Environment | string | - | yes |
 | filter_tags_custom | Tags used for custom filtering when filter_tags_use_defaults is false | string | `*` | no |

middleware/apache/inputs.tf

@@ -27,13 +27,19 @@ variable "filter_tags_custom" {
 # Apache Middleware specific
 
 variable "apache_connect_silenced" {
-  description = "Groups to mute for Apache process monitor"
+  description = "Groups to mute for Apache status monitor"
   type        = "map"
   default     = {}
 }
 
 variable "apache_connect_message" {
-  description = "Custom message for Apache process monitor"
+  description = "Custom message for Apache status monitor"
   type        = "string"
   default     = ""
 }
+
+variable "apache_connect_threshold_critical" {
+  description = "Apache status monitor (critical threshold)"
+  type        = "string"
+  default     = 1.1754943508222875e-38
+}

middleware/apache/monitors-apache.tf

@@ -1,21 +1,19 @@
 resource "datadog_monitor" "datadog_apache_process" {
-  name    = "[${var.environment}] Can't connect to apache vhost status"
+  name    = "[${var.environment}] Apache vhost status does not respond"
   message = "${coalesce(var.apache_connect_message, var.message)}"
 
   type = "service check"
 
   query = <<EOF
-    "apache.can_connect".over${module.filter-tags.service_check}.by("host","port").last(6).count_by_status()
+    "apache.can_connect".over${module.filter-tags.service_check}.by("host","port","server").last(1).pct_by_status()
   EOF
 
   thresholds = {
-    ok       = 1
-    warning  = 2
-    critical = 4
+    warning  = 0
+    critical = "${var.apache_connect_threshold_critical}"
   }
 
   notify_no_data      = true
-  evaluation_delay    = "${var.delay}"
   new_host_delay      = "${var.delay}"
   renotify_interval   = 0
   notify_audit        = false

middleware/nginx/README.md

@@ -16,7 +16,7 @@ module "datadog-monitors-middleware-nginx" {
 
 Creates DataDog monitors with the following checks:
 
-- Can't connect to nginx vhost status
+- Nginx vhost status does not respond
 
 ## Inputs
 
@@ -27,8 +27,9 @@ Creates DataDog monitors with the following checks:
 | filter_tags_custom | Tags used for custom filtering when filter_tags_use_defaults is false | string | `*` | no |
 | filter_tags_use_defaults | Use default filter tags convention | string | `true` | no |
 | message | Message sent when an alert is triggered | string | - | yes |
-| nginx_connect_message | Custom message for Nginx process monitor | string | `` | no |
-| nginx_connect_silenced | Groups to mute for Nginx process monitor | map | `<map>` | no |
+| nginx_connect_message | Custom message for Nginx status monitor | string | `` | no |
+| nginx_connect_silenced | Groups to mute for Nginx status monitor | map | `<map>` | no |
+| nginx_connect_threshold_critical | Nginx status monitor (critical threshold) | string | `1.1754943508222875e-38` | no |
 
 ## Outputs
 

middleware/nginx/inputs.tf

@@ -27,13 +27,19 @@ variable "filter_tags_custom" {
 # Nginx Middleware specific
 
 variable "nginx_connect_silenced" {
-  description = "Groups to mute for Nginx process monitor"
+  description = "Groups to mute for Nginx status monitor"
   type        = "map"
   default     = {}
 }
 
 variable "nginx_connect_message" {
-  description = "Custom message for Nginx process monitor"
+  description = "Custom message for Nginx status monitor"
   type        = "string"
   default     = ""
 }
+
+variable "nginx_connect_threshold_critical" {
+  description = "Nginx status monitor (critical threshold)"
+  type        = "string"
+  default     = 1.1754943508222875e-38
+}

middleware/nginx/monitors-nginx.tf

@@ -1,21 +1,19 @@
 resource "datadog_monitor" "datadog_nginx_process" {
-  name    = "[${var.environment}] Can't connect to nginx vhost status"
+  name    = "[${var.environment}] Nginx vhost status does not respond"
   message = "${coalesce(var.nginx_connect_message, var.message)}"
 
   type = "service check"
 
   query = <<EOF
-    "nginx.can_connect".over${module.filter-tags.service_check}.by("host","port").last(6).count_by_status()
+    "nginx.can_connect".over${module.filter-tags.service_check}.by("host","port","server").last(1).pct_by_status()
   EOF
 
   thresholds = {
-    ok       = 1
-    warning  = 2
-    critical = 4
+    warning  = 0
+    critical = "${var.nginx_connect_threshold_critical}"
   }
 
   notify_no_data      = true
-  evaluation_delay    = "${var.delay}"
   new_host_delay      = "${var.delay}"
   renotify_interval   = 0
   notify_audit        = false

middleware/php-fpm/README.md

@@ -16,8 +16,8 @@ module "datadog-monitors-middleware-php-fpm" {
 
 Creates DataDog monitors with the following checks:
 
-- php_fpm busy worker
-- Can't connect to php-fpm
+- Php-fpm busy worker
+- Php-fpm ping url does not respond
 
 ## Inputs
 

middleware/php-fpm/monitors-fpm.tf

@@ -1,5 +1,5 @@
 resource "datadog_monitor" "datadog_php_fpm_connect_idle" {
-  name    = "[${var.environment}] php_fpm busy worker {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
+  name    = "[${var.environment}] Php-fpm busy worker {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
   message = "${coalesce(var.php_fpm_busy_message, var.message)}"
 
   type = "metric alert"
@@ -32,7 +32,7 @@ resource "datadog_monitor" "datadog_php_fpm_connect_idle" {
 }
 
 resource "datadog_monitor" "datadog_fpm_process" {
-  name    = "[${var.environment}] Can't connect to php-fpm"
+  name    = "[${var.environment}] Php-fpm ping url does not respond"
   message = "${coalesce(var.php_fpm_connect_message, var.message)}"
 
   type = "service check"
@@ -44,11 +44,10 @@ resource "datadog_monitor" "datadog_fpm_process" {
   thresholds = {
     ok       = 1
     warning  = 2
-    critical = 4
+    critical = 5
   }
 
   notify_no_data      = true
-  evaluation_delay    = "${var.delay}"
   new_host_delay      = "${var.delay}"
   renotify_interval   = 0
   notify_audit        = false