feat: create monitoring in cloudwatch for aws batch failures (#367)

* feat: add eventbridge rule and sns topic

* fix: tflint and add data source

* fix: event pattern

* docs: update Terraform docs

---------

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

infra/terraform/modules/service/README.md

@@ -22,9 +22,11 @@
 | <a name="module_ecs_cluster"></a> [ecs\_cluster](#module\_ecs\_cluster) | terraform-aws-modules/ecs/aws//modules/cluster | ~> 5.10 |
 | <a name="module_ecs_service"></a> [ecs\_service](#module\_ecs\_service) | terraform-aws-modules/ecs/aws//modules/service | ~> 5.10 |
 | <a name="module_eventbridge"></a> [eventbridge](#module\_eventbridge) | terraform-aws-modules/eventbridge/aws | ~> 3.7 |
+| <a name="module_eventbridge_sns"></a> [eventbridge\_sns](#module\_eventbridge\_sns) | terraform-aws-modules/eventbridge/aws | ~> 3.7 |
 | <a name="module_log_bucket"></a> [log\_bucket](#module\_log\_bucket) | terraform-aws-modules/s3-bucket/aws | ~> 4.0 |
 | <a name="module_records"></a> [records](#module\_records) | terraform-aws-modules/route53/aws//modules/records | ~> 4.0 |
 | <a name="module_route53_records"></a> [route53\_records](#module\_route53\_records) | terraform-aws-modules/acm/aws | ~> 5.0 |
+| <a name="module_sns_batch_fail"></a> [sns\_batch\_fail](#module\_sns\_batch\_fail) | terraform-aws-modules/sns/aws | ~> 6.1 |
 
 ## Resources
 
@@ -34,6 +36,7 @@
 | [aws_cloudwatch_log_group.this](https://registry.terraform.io/providers/hashicorp/aws/latest/docs/resources/cloudwatch_log_group) | resource |
 | [aws_lb_listener_rule.this](https://registry.terraform.io/providers/hashicorp/aws/latest/docs/resources/lb_listener_rule) | resource |
 | [aws_lb_target_group.this](https://registry.terraform.io/providers/hashicorp/aws/latest/docs/resources/lb_target_group) | resource |
+| [aws_caller_identity.current](https://registry.terraform.io/providers/hashicorp/aws/latest/docs/data-sources/caller_identity) | data source |
 | [aws_canonical_user_id.current](https://registry.terraform.io/providers/hashicorp/aws/latest/docs/data-sources/canonical_user_id) | data source |
 | [aws_cloudfront_log_delivery_canonical_user_id.cloudfront](https://registry.terraform.io/providers/hashicorp/aws/latest/docs/data-sources/cloudfront_log_delivery_canonical_user_id) | data source |
 | [aws_route53_zone.public](https://registry.terraform.io/providers/hashicorp/aws/latest/docs/data-sources/route53_zone) | data source |

infra/terraform/modules/service/batch.tf

@@ -1,3 +1,5 @@
+data "aws_caller_identity" "current" {}
+
 locals {
   default_retry_policy = {
     attempts = 1
@@ -155,9 +157,97 @@ module "eventbridge" {
   }
 
   schedules = local.schedules
+
+}
+
+module "eventbridge_sns" {
+  source  = "terraform-aws-modules/eventbridge/aws"
+  version = "~> 3.7"
+
+  create_bus = false
+
+  role_name = "vol-app-${var.environment}-batch-fail-role"
+
+  rules = {
+    batch-fail-sns = {
+      name        = "${var.environment}-batch-fail-event"
+      description = "Capture failed Batch Events sent to SNS"
+      event_pattern = jsonencode({
+        "source" : ["aws.batch"],
+        "detail-type" : ["Batch Job State Change"],
+        "detail" : {
+          "status" : [
+            "FAILED"
+          ],
+          "jobName" : [{
+            "wildcard" : "vol-app-${var.environment}-*"
+          }]
+        }
+      })
+      enabled = true
+    }
+  }
+
+  targets = {
+    batch-fail-sns = [
+      {
+        name = "batch-fail-event"
+        arn  = module.sns_batch_fail.topic_arn
+      }
+    ]
+  }
+
+}
+
+module "sns_batch_fail" {
+  source  = "terraform-aws-modules/sns/aws"
+  version = "~> 6.1"
+
+  name            = "vol-app-${var.environment}-batch-fail-topic"
+  use_name_prefix = true
+  display_name    = "batch-event-failed"
+
+
+  create_topic_policy         = true
+  enable_default_topic_policy = true
+  topic_policy_statements = {
+    pub = {
+      actions = ["sns:Publish"]
+      principals = [{
+        type = "AWS"
+        identifiers = [
+          "arn:aws:iam::${data.aws_caller_identity.current.account_id}:root"
+        ]
+      }]
+    },
+
+    sub = {
+      actions = [
+        "sns:Subscribe",
+        "sns:Receive",
+      ]
+
+      principals = [{
+        type        = "Service"
+        identifiers = ["events.amazonaws.com"]
+      }]
+
+      conditions = [{
+        test     = "ArnLike"
+        variable = "aws:SourceArn"
+        values   = [module.eventbridge_sns.eventbridge_bus_arn]
+      }]
+    }
+  }
+
+  tags = {
+    "Name" = "vol-app-${var.environment}-aws-sns-batch-fail"
+
+  }
+
 }
 
 resource "aws_cloudwatch_log_group" "this" {
   name              = "/aws/batch/vol-app-${var.environment}"
   retention_in_days = 1
-}
+}