MON-160 - ALB monitors updated

cloud/aws/alb/README.md

@@ -32,6 +32,7 @@ Inputs
 |------|-------------|:----:|:-----:|:-----:|
 | alb_no_healthy_instances_message | Custom message for ALB no healthy instances monitor | string | `` | no |
 | alb_no_healthy_instances_silenced | Groups to mute for ALB no healthy instances monitor | map | `<map>` | no |
+| alb_no_healthy_instances_timeframe | Monitor timeframe for ALB no healthy instances [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_1m` | no |
 | artificial_requests_count | Number of false requests used to mitigate false positive in case of low trafic | string | `5` | no |
 | delay | Delay in seconds for the metric evaluation | string | `900` | no |
 | environment | Architecture environment | string | - | yes |
@@ -41,22 +42,27 @@ Inputs
 | httpcode_elb_4xx_silenced | Groups to mute for ALB httpcode 4xx monitor | map | `<map>` | no |
 | httpcode_elb_4xx_threshold_critical | loadbalancer 4xx critical threshold in percentage | string | `80` | no |
 | httpcode_elb_4xx_threshold_warning | loadbalancer 4xx warning threshold in percentage | string | `60` | no |
+| httpcode_elb_4xx_timeframe | Monitor timeframe for ALB httpcode 4xx [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
 | httpcode_elb_5xx_message | Custom message for ALB httpcode 5xx monitor | string | `` | no |
 | httpcode_elb_5xx_silenced | Groups to mute for ALB httpcode 5xx monitor | map | `<map>` | no |
 | httpcode_elb_5xx_threshold_critical | loadbalancer 5xxcritical threshold in percentage | string | `80` | no |
 | httpcode_elb_5xx_threshold_warning | loadbalancer 5xx warning threshold in percentage | string | `60` | no |
+| httpcode_elb_5xx_timeframe | Monitor timeframe for ALB httpcode 5xx [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
 | httpcode_target_4xx_message | Custom message for ALB target httpcode 4xx monitor | string | `` | no |
 | httpcode_target_4xx_silenced | Groups to mute for ALB target httpcode 4xx monitor | map | `<map>` | no |
 | httpcode_target_4xx_threshold_critical | target 4xx critical threshold in percentage | string | `80` | no |
 | httpcode_target_4xx_threshold_warning | target 4xx warning threshold in percentage | string | `60` | no |
+| httpcode_target_4xx_timeframe | Monitor timeframe for ALB target httpcode 4xx [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
 | httpcode_target_5xx_message | Custom message for ALB target httpcode 5xx monitor | string | `` | no |
 | httpcode_target_5xx_silenced | Groups to mute for ALB target httpcode 5xx monitor | map | `<map>` | no |
 | httpcode_target_5xx_threshold_critical | target 5xx critical threshold in percentage | string | `80` | no |
 | httpcode_target_5xx_threshold_warning | target 5xx warning threshold in percentage | string | `60` | no |
+| httpcode_target_5xx_timeframe | Monitor timeframe for ALB target httpcode 5xx [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
 | latency_message | Custom message for ALB latency monitor | string | `` | no |
 | latency_silenced | Groups to mute for ALB latency monitor | map | `<map>` | no |
 | latency_threshold_critical | latency critical threshold in milliseconds | string | `1000` | no |
 | latency_threshold_warning | latency warning threshold in milliseconds | string | `500` | no |
+| latency_timeframe | Monitor timeframe for ALB latency [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 a monitor is triggered | string | - | yes |
 
 Related documentation

cloud/aws/alb/inputs.tf

@@ -38,6 +38,12 @@ variable "alb_no_healthy_instances_message" {
   default     = ""
 }
 
+variable "alb_no_healthy_instances_timeframe" {
+  description = "Monitor timeframe for ALB no healthy instances [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  type        = "string"
+  default     = "last_1m"
+}
+
 variable "latency_silenced" {
   description = "Groups to mute for ALB latency monitor"
   type        = "map"
@@ -50,6 +56,12 @@ variable "latency_message" {
   default     = ""
 }
 
+variable "latency_timeframe" {
+  description = "Monitor timeframe for ALB latency [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  type        = "string"
+  default     = "last_5m"
+}
+
 variable "latency_threshold_critical" {
   default     = 1000
   description = "latency critical threshold in milliseconds"
@@ -72,6 +84,12 @@ variable "httpcode_elb_4xx_message" {
   default     = ""
 }
 
+variable "httpcode_elb_4xx_timeframe" {
+  description = "Monitor timeframe for ALB httpcode 4xx [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  type        = "string"
+  default     = "last_5m"
+}
+
 variable "httpcode_elb_4xx_threshold_critical" {
   default     = 80
   description = "loadbalancer 4xx critical threshold in percentage"
@@ -94,6 +112,12 @@ variable "httpcode_target_4xx_message" {
   default     = ""
 }
 
+variable "httpcode_target_4xx_timeframe" {
+  description = "Monitor timeframe for ALB target httpcode 4xx [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  type        = "string"
+  default     = "last_5m"
+}
+
 variable "httpcode_target_4xx_threshold_critical" {
   default     = 80
   description = "target 4xx critical threshold in percentage"
@@ -116,6 +140,12 @@ variable "httpcode_elb_5xx_message" {
   default     = ""
 }
 
+variable "httpcode_elb_5xx_timeframe" {
+  description = "Monitor timeframe for ALB httpcode 5xx [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  type        = "string"
+  default     = "last_5m"
+}
+
 variable "httpcode_elb_5xx_threshold_critical" {
   default     = 80
   description = "loadbalancer 5xxcritical threshold in percentage"
@@ -138,6 +168,12 @@ variable "httpcode_target_5xx_message" {
   default     = ""
 }
 
+variable "httpcode_target_5xx_timeframe" {
+  description = "Monitor timeframe for ALB target httpcode 5xx [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  type        = "string"
+  default     = "last_5m"
+}
+
 variable "httpcode_target_5xx_threshold_critical" {
   default     = 80
   description = "target 5xx critical threshold in percentage"

cloud/aws/alb/monitors-alb.tf

@@ -14,7 +14,7 @@ resource "datadog_monitor" "ALB_no_healthy_instances" {
   message = "${coalesce(var.alb_no_healthy_instances_message, var.message)}"
 
   query = <<EOF
-    min(last_1m): (
+    min(${var.alb_no_healthy_instances_timeframe}): (
       min:aws.applicationelb.healthy_host_count{${data.template_file.filter.rendered}} by {region,loadbalancer}
     ) <= 0
   EOF
@@ -43,7 +43,7 @@ resource "datadog_monitor" "ALB_latency" {
   message = "${coalesce(var.latency_message, var.message)}"
 
   query = <<EOF
-    min(last_5m): (
+    min(${var.latency_timeframe}): (
       min:aws.applicationelb.target_response_time.average{${data.template_file.filter.rendered}} by {region,loadbalancer}
     ) > ${var.latency_threshold_critical}
   EOF
@@ -73,7 +73,7 @@ resource "datadog_monitor" "ALB_httpcode_elb_5xx" {
   message = "${coalesce(var.httpcode_elb_5xx_message, var.message)}"
 
   query = <<EOF
-    min(last_5m): (
+    min(${var.httpcode_elb_5xx_timeframe}): (
       default(
         min:aws.applicationelb.httpcode_elb_5xx{${data.template_file.filter.rendered}} by {region,loadbalancer}.as_count() /
         (min:aws.applicationelb.request_count{${data.template_file.filter.rendered}} by {region,loadbalancer}.as_count() + ${var.artificial_requests_count}),
@@ -106,7 +106,7 @@ resource "datadog_monitor" "ALB_httpcode_elb_4xx" {
   message = "${coalesce(var.httpcode_elb_4xx_message, var.message)}"
 
   query = <<EOF
-    min(last_5m): (
+    min(${var.httpcode_elb_4xx_timeframe}): (
       default(
         min:aws.applicationelb.httpcode_elb_4xx{${data.template_file.filter.rendered}} by {region,loadbalancer}.as_count() /
         (min:aws.applicationelb.request_count{${data.template_file.filter.rendered}} by {region,loadbalancer}.as_count() + ${var.artificial_requests_count}),
@@ -139,7 +139,7 @@ resource "datadog_monitor" "ALB_httpcode_target_5xx" {
   message = "${coalesce(var.httpcode_target_5xx_message, var.message)}"
 
   query = <<EOF
-    min(last_5m): (
+    min(${var.httpcode_target_5xx_timeframe}): (
       default(
         min:aws.applicationelb.httpcode_target_5xx{${data.template_file.filter.rendered}} by {region,loadbalancer}.as_count() /
         (min:aws.applicationelb.request_count{${data.template_file.filter.rendered}} by {region,loadbalancer}.as_count() + ${var.artificial_requests_count}),
@@ -172,7 +172,7 @@ resource "datadog_monitor" "ALB_httpcode_target_4xx" {
   message = "${coalesce(var.httpcode_target_4xx_message, var.message)}"
 
   query = <<EOF
-    min(last_5m): (
+    min(${var.httpcode_target_4xx_timeframe}): (
       default(
         min:aws.applicationelb.httpcode_target_4xx{${data.template_file.filter.rendered}} by {region,loadbalancer}.as_count() /
         (min:aws.applicationelb.request_count{${data.template_file.filter.rendered}} by {region,loadbalancer}.as_count() + ${var.artificial_requests_count}),