Fix CloudWatch Agent dimensions and alarms

Author: electroly

project/TASKS.md

@@ -31,7 +31,7 @@
     - [x] Give the instance profile permission to access the filesystem, if needed
     - [x] Update the EC2 instance terraform to mount the EFS filesystem
     - [x] Update env.example to include the correct mount path for EFS for LETS_ENCRYPT_CERTS_DIR
-- [ ] The CloudWatch Agent configuration in user-data.sh is wrong.
-    - [ ] The mem_used_percent metric is appearing with a host dimension, with a name like "ip-10-0-1-200". We want "instance id" as the sole dimension.
-    - [ ] The disk_used_percent metric is appearing with device, fstype, host, and path dimensions. We want "instance id" and "path" as the two dimensions.
-    - [ ] Update the CloudWatch alarms accordingly; they are pointing at nonexistent metrics right now.
+- [x] The CloudWatch Agent configuration in user-data.sh is wrong.
+    - [x] The mem_used_percent metric is appearing with a host dimension, with a name like "ip-10-0-1-200". We want "instance id" as the sole dimension.
+    - [x] The disk_used_percent metric is appearing with device, fstype, host, and path dimensions. We want "instance id" and "path" as the two dimensions.
+    - [x] Update the CloudWatch alarms accordingly; they are pointing at nonexistent metrics right now.

terraform/main.tf

@@ -343,8 +343,6 @@ resource "aws_cloudwatch_metric_alarm" "disk_usage" {
 
   dimensions = {
     InstanceId = aws_instance.api.id
-    device     = "/dev/root"
-    fstype     = "ext4"
     path       = "/"
   }
 

terraform/user-data.sh

@@ -48,7 +48,14 @@ cat > /opt/aws/amazon-cloudwatch-agent/etc/amazon-cloudwatch-agent.json << EOF
           "/"
         ]
       }
-    }
+    },
+    "append_dimensions": {
+      "InstanceId": "${aws:InstanceId}"
+    },
+    "aggregation_dimensions": [
+      ["InstanceId"],
+      ["InstanceId", "path"]
+    ]
   }
 }
 EOF