Merge pull request #1 from rhythmictech/fix

bug fixes and tuning
rhythmictech · Mar 13, 2024 · 2f9e033 · 2f9e033
2 parents 3fe573c + 1d73043
commit 2f9e033
Show file tree

Hide file tree

Showing 32 changed files with 632 additions and 177 deletions.
diff --git a/.github/workflows/tfsec.yaml b/.github/workflows/tfsec.yaml
diff --git a/README.md b/README.md
@@ -1,10 +1,12 @@
-# terraform-datadog-monitors
+# terraform-datadog-monitor
 
-[![tflint](https://github.com/rhythmictech/terraform-datadog-monitors/workflows/tflint/badge.svg?branch=master&event=push)](https://github.com/rhythmictech/terraform-datadog-monitors/actions?query=workflow%3Atflint+event%3Apush+branch%3Amaster)
-[![tfsec](https://github.com/rhythmictech/terraform-datadog-monitors/workflows/tfsec/badge.svg?branch=master&event=push)](https://github.com/rhythmictech/terraform-datadog-monitors/actions?query=workflow%3Atfsec+event%3Apush+branch%3Amaster)
-[![yamllint](https://github.com/rhythmictech/terraform-datadog-monitors/workflows/yamllint/badge.svg?branch=master&event=push)](https://github.com/rhythmictech/terraform-datadog-monitors/actions?query=workflow%3Ayamllint+event%3Apush+branch%3Amaster)
-[![misspell](https://github.com/rhythmictech/terraform-datadog-monitors/workflows/misspell/badge.svg?branch=master&event=push)](https://github.com/rhythmictech/terraform-datadog-monitors/actions?query=workflow%3Amisspell+event%3Apush+branch%3Amaster)
-[![pre-commit-check](https://github.com/rhythmictech/terraform-datadog-monitors/workflows/pre-commit-check/badge.svg?branch=master&event=push)](https://github.com/rhythmictech/terraform-datadog-monitors/actions?query=workflow%3Apre-commit-check+event%3Apush+branch%3Amaster)
+A collection of semi-opinionated Datadog monitors for a variety of use cases.
+
+[![tflint](https://github.com/rhythmictech/terraform-datadog-monitor/workflows/tflint/badge.svg?branch=master&event=push)](https://github.com/rhythmictech/terraform-datadog-monitor/actions?query=workflow%3Atflint+event%3Apush+branch%3Amaster)
+[![tfsec](https://github.com/rhythmictech/terraform-datadog-monitor/workflows/tfsec/badge.svg?branch=master&event=push)](https://github.com/rhythmictech/terraform-datadog-monitor/actions?query=workflow%3Atfsec+event%3Apush+branch%3Amaster)
+[![yamllint](https://github.com/rhythmictech/terraform-datadog-monitor/workflows/yamllint/badge.svg?branch=master&event=push)](https://github.com/rhythmictech/terraform-datadog-monitor/actions?query=workflow%3Ayamllint+event%3Apush+branch%3Amaster)
+[![misspell](https://github.com/rhythmictech/terraform-datadog-monitor/workflows/misspell/badge.svg?branch=master&event=push)](https://github.com/rhythmictech/terraform-datadog-monitor/actions?query=workflow%3Amisspell+event%3Apush+branch%3Amaster)
+[![pre-commit-check](https://github.com/rhythmictech/terraform-datadog-monitor/workflows/pre-commit-check/badge.svg?branch=master&event=push)](https://github.com/rhythmictech/terraform-datadog-monitor/actions?query=workflow%3Apre-commit-check+event%3Apush+branch%3Amaster)
 <a href="https://twitter.com/intent/follow?screen_name=RhythmicTech"><img src="https://img.shields.io/twitter/follow/RhythmicTech?style=social&logo=twitter" alt="follow on Twitter"></a>
 
 ## Requirements
@@ -19,7 +21,7 @@ provider "datadog" {
 }
 
 module "monitor" {
-  source = "rhythmictech/monitors/datadog/module"
+  source = "rhythmictech/monitors/datadog/aws/ec2"
 
 }
 ```

diff --git a/aws/alb/.terraform.lock.hcl b/aws/alb/.terraform.lock.hcl
diff --git a/aws/alb/main.tf b/aws/alb/main.tf
@@ -27,8 +27,8 @@ resource "datadog_monitor" "http_5xx_responses" {
 
   query = <<END
     min(${var.http_5xx_responses_evaluation_window}):
-      default(avg:aws.applicationelb.httpcode_elb_5xx${local.query_filter} by {region,loadbalancer}.as_rate(), 0) / (
-      default(avg:aws.applicationelb.request_count${local.query_filter} by {region,loadbalancer}.as_rate(), 1)
+      default(avg:aws.applicationelb.httpcode_elb_5xx${local.query_filter} by {loadbalancer,region,aws_account}.as_rate(), 0) / (
+      default(avg:aws.applicationelb.request_count${local.query_filter} by {loadbalancer,region,aws_account}.as_rate(), 1)
     ) * 100 > ${var.http_5xx_responses_threshold_critical}
 END
 
@@ -57,8 +57,8 @@ resource "datadog_monitor" "http_5xx_tg_responses" {
 
   query = <<END
     min(${var.http_5xx_tg_responses_evaluation_window}):
-      default(avg:aws.applicationelb.httpcode_elb_5xx${local.query_filter} by {region,loadbalancer,targetgroup}.as_rate(), 0) / (
-      default(avg:aws.applicationelb.request_count${local.query_filter} by {region,loadbalancer,targetgroup}.as_rate(), 1)
+      default(avg:aws.applicationelb.httpcode_elb_5xx${local.query_filter} by {loadbalancer,region,aws_account,targetgroup}.as_rate(), 0) / (
+      default(avg:aws.applicationelb.request_count${local.query_filter} by {loadbalancer,region,aws_account,targetgroup}.as_rate(), 1)
     ) * 100 > ${var.http_5xx_tg_responses_threshold_critical}
 END
 
@@ -88,7 +88,7 @@ resource "datadog_monitor" "latency" {
 
   query = <<END
     avg(${var.latency_evaluation_window}):
-      default(avg:aws.applicationelb.target_response_time.average${local.query_filter} by {region,loadbalancer}, 0
+      default(avg:aws.applicationelb.target_response_time.average${local.query_filter} by {loadbalancer,region,aws_account}, 0
     ) > ${var.latency_threshold_critical}
 END
 
@@ -117,9 +117,9 @@ resource "datadog_monitor" "no_healthy_instances" {
 
   query = <<END
     min(${var.no_healthy_instances_evaluation_window}): (
-      sum:aws.applicationelb.healthy_host_count.minimum${local.query_filter} by {region,loadbalancer} / (
-      sum:aws.applicationelb.healthy_host_count.minimum${local.query_filter} by {region,loadbalancer} +
-      sum:aws.applicationelb.un_healthy_host_count.maximum${local.query_filter} by {region,loadbalancer} )
+      sum:aws.applicationelb.healthy_host_count.minimum${local.query_filter} by {loadbalancer,region,aws_account} / (
+      sum:aws.applicationelb.healthy_host_count.minimum${local.query_filter} by {loadbalancer,region,aws_account} +
+      sum:aws.applicationelb.un_healthy_host_count.maximum${local.query_filter} by {loadbalancer,region,aws_account} )
     ) <= ${var.no_healthy_instances_threshold_critical}
 END
 

diff --git a/aws/apigateway/.terraform.lock.hcl b/aws/apigateway/.terraform.lock.hcl
diff --git a/aws/apigateway/main.tf b/aws/apigateway/main.tf
@@ -27,8 +27,8 @@ resource "datadog_monitor" "http_5xx_responses" {
 
   query = <<END
     min(${var.http_5xx_responses_evaluation_window}):
-      default(avg:aws.apigateway.5xxerror{${local.query_filter}} by {region,apiname,stage}.as_rate(), 0) / (
-      default(avg:aws.apigateway.count{${local.query_filter}} by {region,apiname,stage}.as_rate(), 1)
+      default(avg:aws.apigateway.5xxerror{${local.query_filter}} by {stage,apiname,region,aws_account}.as_rate(), 0) / (
+      default(avg:aws.apigateway.count{${local.query_filter}} by {stage,apiname,region,aws_account}.as_rate(), 1)
     ) * 100 > ${var.http_5xx_responses_threshold_critical}
 END
 
@@ -57,7 +57,7 @@ resource "datadog_monitor" "latency" {
 
   query = <<END
     avg(${var.latency_evaluation_window}):
-      default(avg:aws.apigateway.latency{${local.query_filter}} by {region,apiname,stage}, 0)
+      default(avg:aws.apigateway.latency{${local.query_filter}} by {stage,apiname,region,aws_account}, 0)
     ) > ${var.latency_threshold_critical}
 END
 

diff --git a/aws/beanstalk/.terraform.lock.hcl b/aws/beanstalk/.terraform.lock.hcl
diff --git a/aws/beanstalk/main.tf b/aws/beanstalk/main.tf
@@ -40,7 +40,7 @@ resource "datadog_monitor" "health" {
 
   query = <<END
     min(${var.health_evaluation_window}):
-      min:aws.elasticbeanstalk.environment_health${local.query_filter} by {region,environmentname}
+      min:aws.elasticbeanstalk.environment_health${local.query_filter} by {environmentname,region,aws_account}
     >= ${var.health_threshold_critical}
 END
 
@@ -69,8 +69,8 @@ resource "datadog_monitor" "http_5xx_responses" {
 
   query = <<END
     min(${var.http_5xx_responses_evaluation_window}):(
-      default(sum:aws.elasticbeanstalk.application_requests_5xx${local.query_filter} by {region,environmentname}.as_rate(), 0) /
-      default(sum:aws.elasticbeanstalk.application_requests_total${local.query_filter} by {region,environmentname}.as_rate(), 1)
+      default(sum:aws.elasticbeanstalk.application_requests_5xx${local.query_filter} by {environmentname,region,aws_account}.as_rate(), 0) /
+      default(sum:aws.elasticbeanstalk.application_requests_total${local.query_filter} by {environmentname,region,aws_account}.as_rate(), 1)
     ) * 100 > ${var.http_5xx_responses_threshold_critical}
 END
 
@@ -98,7 +98,7 @@ resource "datadog_monitor" "latency" {
   timeout_h           = var.timeout_h
 
   query = <<END
-    min:${var.latency_evaluation_window}):min:aws.elasticbeanstalk.${local.latency_metric}${local.query_filter} by {region,environmentname}
+    min:${var.latency_evaluation_window}):min:aws.elasticbeanstalk.${local.latency_metric}${local.query_filter} by {environmentname,region,aws_account}
      >= ${var.latency_threshold_critical}
 END
 
@@ -127,7 +127,7 @@ resource "datadog_monitor" "root_disk_usage" {
 
   query = <<END
     max:${var.latency_evaluation_window}):
-      min:aws.elasticbeanstalk.root_filesystem_util${local.query_filter} by {region,environmentname,host}
+      min:aws.elasticbeanstalk.root_filesystem_util${local.query_filter} by {host,environmentname,region,aws_account}
     >= ${var.root_disk_usage_threshold_critical}
 END
 

diff --git a/aws/ec2/.terraform.lock.hcl b/aws/ec2/.terraform.lock.hcl
diff --git a/aws/ec2/README.md b/aws/ec2/README.md
@@ -2,6 +2,8 @@
 
 Configures monitors for EC2 instance health, including host issues, instance status check, ec2 status checks, and attached volume status check.
 
+All checks are enabled by default.
+
 <!-- BEGIN_TF_DOCS -->
 ## Requirements