Aws ecs transformation fix (#546)

* fix: replacing min() with lasting() * fix: removing dangerous usage of mean() * fix: terraform lint * fix: remove mean() on heartbeat monitor --------- Co-authored-by: Florent DELAHAYE <[email protected]>
claranet · Jan 29, 2024 · 200d100 · 200d100
1 parent 45f46e4
commit 200d100
Show file tree

Hide file tree

Showing 2 changed files with 28 additions and 9 deletions.
diff --git a/modules/integration_aws-ecs-service/detectors-ecs-service.tf b/modules/integration_aws-ecs-service/detectors-ecs-service.tf
@@ -35,9 +35,9 @@ resource "signalfx_detector" "cpu_utilization" {
   tags                    = compact(concat(local.common_tags, local.tags, var.extra_tags))
 
   program_text = <<-EOF
-    signal = data('CPUUtilization', filter=filter('namespace', 'AWS/ECS') and filter('stat', 'mean') and filter('ServiceName', '*') and ${module.filtering.signalflow}).mean(by=['ServiceName'])${var.cpu_utilization_aggregation_function}${var.cpu_utilization_transformation_function}.publish('signal')
-    detect(when(signal > ${var.cpu_utilization_threshold_critical})).publish('CRIT')
-    detect(when(signal > ${var.cpu_utilization_threshold_major}) and (not when(signal > ${var.cpu_utilization_threshold_critical}))).publish('MAJOR')
+    signal = data('CPUUtilization', filter=filter('namespace', 'AWS/ECS') and filter('stat', 'mean') and filter('ServiceName', '*') and ${module.filtering.signalflow})${var.cpu_utilization_aggregation_function}${var.cpu_utilization_transformation_function}.publish('signal')
+    detect(when(signal > ${var.cpu_utilization_threshold_critical}, lasting=%{if var.cpu_utilization_lasting_duration_critical == null}None%{else}'${var.cpu_utilization_lasting_duration_critical}'%{endif})).publish('CRIT')
+    detect(when(signal > ${var.cpu_utilization_threshold_major}, lasting=%{if var.cpu_utilization_lasting_duration_major == null}None%{else}'${var.cpu_utilization_lasting_duration_major}'%{endif}) and (not when(signal > ${var.cpu_utilization_threshold_critical}, lasting=%{if var.cpu_utilization_lasting_duration_critical == null}None%{else}'${var.cpu_utilization_lasting_duration_critical}'%{endif}))).publish('MAJOR')
 EOF
 
   rule {
@@ -75,9 +75,9 @@ resource "signalfx_detector" "memory_utilization" {
   tags                    = compact(concat(local.common_tags, local.tags, var.extra_tags))
 
   program_text = <<-EOF
-    signal = data('MemoryUtilization', filter=filter('namespace', 'AWS/ECS') and filter('stat', 'mean') and filter('ServiceName', '*') and ${module.filtering.signalflow}).mean(by=['ServiceName'])${var.memory_utilization_aggregation_function}${var.memory_utilization_transformation_function}.publish('signal')
-    detect(when(signal > ${var.memory_utilization_threshold_critical})).publish('CRIT')
-    detect(when(signal > ${var.memory_utilization_threshold_major}) and (not when(signal > ${var.memory_utilization_threshold_critical}))).publish('MAJOR')
+    signal = data('MemoryUtilization', filter=filter('namespace', 'AWS/ECS') and filter('stat', 'mean') and filter('ServiceName', '*') and ${module.filtering.signalflow})${var.memory_utilization_aggregation_function}${var.memory_utilization_transformation_function}.publish('signal')
+    detect(when(signal > ${var.memory_utilization_threshold_critical}, lasting=%{if var.memory_utilization_lasting_duration_critical == null}None%{else}'${var.memory_utilization_lasting_duration_critical}'%{endif})).publish('CRIT')
+    detect(when(signal > ${var.memory_utilization_threshold_major}, lasting=%{if var.memory_utilization_lasting_duration_major == null}None%{else}'${var.memory_utilization_lasting_duration_major}'%{endif}) and (not when(signal > ${var.memory_utilization_threshold_critical}, lasting=%{if var.memory_utilization_lasting_duration_critical == null}None%{else}'${var.memory_utilization_lasting_duration_critical}'%{endif}))).publish('MAJOR')
 EOF
 
   rule {

diff --git a/modules/integration_aws-ecs-service/variables.tf b/modules/integration_aws-ecs-service/variables.tf
@@ -41,7 +41,7 @@ variable "heartbeat_timeframe" {
 variable "heartbeat_aggregation_function" {
   description = "Aggregation function and group by for heartbeat detector (i.e. \".mean(by=['host'])\")"
   type        = string
-  default     = ".mean(by=['ServiceName'])"
+  default     = ""
 }
 
 # CPU_utilization detector
@@ -97,7 +97,7 @@ variable "cpu_utilization_aggregation_function" {
 variable "cpu_utilization_transformation_function" {
   description = "Transformation function for cpu_utilization detector (i.e. \".mean(over='5m')\")"
   type        = string
-  default     = ".min(over='5m')"
+  default     = ""
 }
 
 variable "cpu_utilization_threshold_critical" {
@@ -106,12 +106,22 @@ variable "cpu_utilization_threshold_critical" {
   default     = 90
 }
 
+variable "cpu_utilization_lasting_duration_critical" {
+  type    = string
+  default = "30m"
+}
+
 variable "cpu_utilization_threshold_major" {
   description = "Major threshold for cpu_utilization detector"
   type        = number
   default     = 80
 }
 
+variable "cpu_utilization_lasting_duration_major" {
+  type    = string
+  default = "5m"
+}
+
 # Memory_utilization detector
 
 variable "memory_utilization_max_delay" {
@@ -165,7 +175,7 @@ variable "memory_utilization_aggregation_function" {
 variable "memory_utilization_transformation_function" {
   description = "Transformation function for memory_utilization detector (i.e. \".mean(over='5m')\")"
   type        = string
-  default     = ".min(over='5m')"
+  default     = ""
 }
 
 variable "memory_utilization_threshold_critical" {
@@ -174,9 +184,18 @@ variable "memory_utilization_threshold_critical" {
   default     = 90
 }
 
+variable "memory_utilization_lasting_duration_critical" {
+  type    = string
+  default = "30m"
+}
+
 variable "memory_utilization_threshold_major" {
   description = "Major threshold for memory_utilization detector"
   type        = number
   default     = 85
 }
 
+variable "memory_utilization_lasting_duration_major" {
+  type    = string
+  default = "5m"
+}