MON-497 Warning threshold for memory limit increased from 80 to 85%. Removed the inputs for the critical threshold and the last of the service check.

caas/docker/README.md

@@ -32,17 +32,15 @@ Creates DataDog monitors with the following checks:
 | memory\_used\_extra\_tags | Extra tags for Container Memory Usage monitor | list(string) | `[]` | no |
 | memory\_used\_message | Custom message for the Container Memory Usage monitor | string | `""` | no |
 | memory\_used\_threshold\_critical | Container Memory Usage critical threshold | string | `"90"` | no |
-| memory\_used\_threshold\_warning | Container Memory Usage warning threshold | string | `"80"` | no |
+| memory\_used\_threshold\_warning | Container Memory Usage warning threshold | string | `"85"` | no |
 | memory\_used\_time\_aggregator | Time aggregator for the Container Memory Usage monitor | string | `"min"` | no |
 | memory\_used\_timeframe | Timeframe for the Container Memory Usage monitor | string | `"last_5m"` | no |
 | message | Message sent when an alert is triggered | string | n/a | yes |
 | new\_host\_delay | Delay in seconds before monitor new resource | string | `"300"` | no |
 | not\_responding\_enabled | Flag to enable Docker does not respond monitor | string | `"true"` | no |
 | not\_responding\_extra\_tags | Extra tags for Docker does not respond monitor | list(string) | `[]` | no |
-| not\_responding\_last | Parameter 'last' for the service check | string | `"6"` | no |
 | not\_responding\_message | Custom message for Docker does not respond monitor | string | `""` | no |
 | not\_responding\_no\_data\_timeframe | Docker does not respond monitor no data timeframe | string | `"10"` | no |
-| not\_responding\_threshold\_critical | Docker does not respond monitor (warning threshold) | string | `"5"` | no |
 | not\_responding\_threshold\_warning | Docker does not respond monitor (warning threshold) | string | `"3"` | no |
 | prefix\_slug | Prefix string to prepend between brackets on every monitors names | string | `""` | no |
 

caas/docker/inputs.tf

@@ -54,24 +54,12 @@ variable "not_responding_message" {
   default     = ""
 }
 
-variable "not_responding_last" {
-  description = "Parameter 'last' for the service check"
-  type        = string
-  default     = 6
-}
-
 variable "not_responding_threshold_warning" {
   description = "Docker does not respond monitor (warning threshold)"
   type        = string
   default     = 3
 }
 
-variable "not_responding_threshold_critical" {
-  description = "Docker does not respond monitor (warning threshold)"
-  type        = string
-  default     = 5
-}
-
 variable "not_responding_no_data_timeframe" {
   description = "Docker does not respond monitor no data timeframe"
   type        = string
@@ -114,7 +102,7 @@ variable "memory_used_timeframe" {
 variable "memory_used_threshold_warning" {
   description = "Container Memory Usage warning threshold"
   type        = string
-  default     = 80
+  default     = 85
 }
 
 variable "memory_used_threshold_critical" {

caas/docker/monitors-docker.tf

@@ -8,12 +8,12 @@ resource "datadog_monitor" "not_responding" {
   type    = "service check"
 
   query = <<EOQ
-    "docker.service_up"${module.filter-tags.service_check}.by("host").last(${var.not_responding_last}).count_by_status()
+    "docker.service_up"${module.filter-tags.service_check}.by("host").last(6).count_by_status()
 EOQ
 
   thresholds = {
     warning  = var.not_responding_threshold_warning
-    critical = var.not_responding_threshold_critical
+    critical = 5
   }
 
   new_host_delay      = var.new_host_delay

caas/docker/outputs.tf

@@ -1,10 +1,10 @@
-output "not_responding_id" {
-  description = "id for monitor not_responding"
-  value       = datadog_monitor.not_responding.*.id
-}
-
 output "memory_used_id" {
   description = "id for monitor memory_used"
   value       = datadog_monitor.memory_used.*.id
 }
 
+output "not_responding_id" {
+  description = "id for monitor not_responding"
+  value       = datadog_monitor.not_responding.*.id
+}
+