git.squeep.com Git - awsible/blob - infrastructure/modules/management-stack/alarms.tf

   1 resource "aws_cloudwatch_metric_alarm" "unconsumed" {
   2         alarm_name = "${length(var.sqs_events_name) > 0 ? var.sqs_events_name : var.management_service_name}${length(var.sqs_events_name) > 0 ? "" : "-events"}-unconsumed"
   3         alarm_description = "Alert when queue has aging messages."
   4         metric_name = "ApproximateAgeOfOldestMessage"
   5         comparison_operator = "GreaterThanOrEqualToThreshold"
   6         evaluation_periods = 10
   7         period = 60
   8         namespace = "AWS/SQS"
   9         statistic = "Maximum"
  10         threshold = 1800
  11         dimensions {
  12                 QueueName = "${length(var.sqs_events_name) > 0 ? var.sqs_events_name : var.management_service_name}${length(var.sqs_events_name) > 0 ? "" : "-events"}"
  13         }
  14         alarm_actions = ["${aws_sns_topic.management-alerts.arn}"]
  15 }
  16
  17 resource "aws_cloudwatch_metric_alarm" "failures-present" {
  18         alarm_name = "${length(var.sqs_events_name) > 0 ? var.sqs_events_name : var.management_service_name}${length(var.sqs_events_name) > 0 ? "" : "-events"}-failed-present"
  19         alarm_description = "Alert when a message has failed."
  20         metric_name = "ApproximateNumberOfMessagesVisible"
  21         comparison_operator = "GreaterThanOrEqualToThreshold"
  22         evaluation_periods = 2
  23         period = 60
  24         namespace = "AWS/SQS"
  25         statistic = "Sum"
  26         threshold = 1
  27         dimensions {
  28                 QueueName = "${length(var.sqs_events_name) > 0 ? var.sqs_events_name : var.management_service_name}${length(var.sqs_events_name) > 0 ? "" : "-events"}-failed"
  29         }
  30         alarm_actions = ["${aws_sns_topic.management-alerts.arn}"]
  31 }
  32
  33 resource "aws_cloudwatch_metric_alarm" "unhealthy" {
  34         alarm_name = "UnhealthyHostCount"
  35         alarm_description = "Alert when any ELB has unhealthy hosts."
  36         metric_name = "UnHealthyHostCount"
  37         comparison_operator = "GreaterThanOrEqualToThreshold"
  38         evaluation_periods = 15
  39         period = 60
  40         namespace = "AWS/ELB"
  41         statistic = "Maximum"
  42         threshold = 1
  43         alarm_actions = ["${aws_sns_topic.management-alerts.arn}"]
  44 }