terraform.aws-baseline-infra/modules/ManagementGovernance/Monitoring.EC2/main.tf

resource "aws_cloudwatch_metric_alarm" "ec2-StatusCheckFailed_System" {
  alarm_name                = "${var.cw-alarm-prefix}:EC2:StatusCheckFailed_System:${var.ec2-instance-id}"
  comparison_operator       = "GreaterThanThreshold"
  evaluation_periods        = "1"
  metric_name               = "StatusCheckFailed_System"
  period                    = "300"
  statistic                 = "Maximum"
  threshold                 = 0
  alarm_description         = "EC2:StatusCheckFailed_System"
  namespace                 = "AWS/EC2"
  insufficient_data_actions = []
  actions_enabled           = var.actions-enabled
  alarm_actions             = [var.sns-targets.alarm-actions-emergency]
  ok_actions                = [var.sns-targets.alarm-actions-emergency]
  dimensions = {
    InstanceId = var.ec2-instance-id
  }
  tags = var.default-tags
  lifecycle {
    ignore_changes = [tags]
  }
}

resource "aws_cloudwatch_metric_alarm" "ec2-StatusCheckFailed_Instance" {
  alarm_name                = "${var.cw-alarm-prefix}:EC2:StatusCheckFailed_Instance:${var.ec2-instance-id}"
  comparison_operator       = "GreaterThanThreshold"
  evaluation_periods        = "1"
  metric_name               = "StatusCheckFailed_Instance"
  period                    = "300"
  statistic                 = "Maximum"
  threshold                 = 0
  alarm_description         = "EC2:StatusCheckFailed_Instance"
  namespace                 = "AWS/EC2"
  insufficient_data_actions = []
  actions_enabled           = var.actions-enabled
  alarm_actions             = [var.sns-targets.alarm-actions-emergency]
  ok_actions                = [var.sns-targets.alarm-actions-emergency]
  dimensions = {
    InstanceId = var.ec2-instance-id
  }
  tags = var.default-tags
  lifecycle {
    ignore_changes = [tags]
  }
}

resource "aws_cloudwatch_metric_alarm" "ec2-CPUUtilization" {
  alarm_name                = "${var.cw-alarm-prefix}:EC2:CPUUtilization:${var.ec2-instance-id}"
  comparison_operator       = "GreaterThanThreshold"
  evaluation_periods        = "6"
  metric_name               = "CPUUtilization"
  period                    = "300"
  statistic                 = "Average"
  threshold                 = var.threshold-CPUUtilization
  alarm_description         = "EC2:CPUUtilization"
  namespace                 = "AWS/EC2"
  insufficient_data_actions = []
  actions_enabled           = var.actions-enabled
  alarm_actions             = [var.sns-targets.alarm-actions-urgent]
  ok_actions                = [var.sns-targets.alarm-actions-urgent]
  treat_missing_data        = "notBreaching"
  dimensions = {
    InstanceId = var.ec2-instance-id
  }
  tags = var.default-tags
  lifecycle {
    ignore_changes = [tags]
  }
}

# cwagent metrics
data "aws_instance" "ec2-instance" {
  instance_id = var.ec2-instance-id
}

resource "aws_cloudwatch_metric_alarm" "ec2-mem_free" {
  alarm_name                = "${var.cw-alarm-prefix}:EC2:mem_free:${var.ec2-instance-id}"
  comparison_operator       = "LessThanThreshold"
  evaluation_periods        = "2"
  metric_name               = "mem_free"
  period                    = "900"
  statistic                 = "Average"
  threshold                 = var.threshold-mem_free
  alarm_description         = "EC2:mem_free"
  namespace                 = "CWAgent"
  insufficient_data_actions = []
  actions_enabled           = var.actions-enabled
  alarm_actions             = [var.sns-targets.alarm-actions-standard]
  ok_actions                = [var.sns-targets.alarm-actions-standard]
  dimensions = {
    InstanceId   = var.ec2-instance-id
    ImageId      = data.aws_instance.ec2-instance.ami
    InstanceType = data.aws_instance.ec2-instance.instance_type
  }
  tags = var.default-tags
  lifecycle {
    ignore_changes = [tags]
  }
}

resource "aws_cloudwatch_metric_alarm" "ec2-swap_free" {
  # zero is fine as most ec2 instances are deployed without any swap
  alarm_name                = "${var.cw-alarm-prefix}:EC2:swap_free:${var.ec2-instance-id}"
  comparison_operator       = "LessThanThreshold"
  evaluation_periods        = "2"
  threshold                 = var.threshold-swap_free
  alarm_description         = "EC2:swap_free"
  insufficient_data_actions = []
  actions_enabled           = var.actions-enabled
  alarm_actions             = [var.sns-targets.alarm-actions-standard]
  ok_actions                = [var.sns-targets.alarm-actions-standard]
  metric_query {
    id = "m1"
    metric {
      metric_name = "swap_free"
      namespace   = "CWAgent"
      period      = 900
      stat        = "Average"
      dimensions = {
        InstanceId   = var.ec2-instance-id
        ImageId      = data.aws_instance.ec2-instance.ami
        InstanceType = data.aws_instance.ec2-instance.instance_type
      }
    }
  }
  metric_query {
    id          = "e1"
    expression  = "IF(m1==0, ${var.threshold-swap_free}, m1)"
    label       = "swap_free_if_not_zero"
    return_data = "true"
  }

  tags = var.default-tags
  lifecycle {
    ignore_changes = [tags]
  }
}


# get device dimension from cw metrics
data "external" "disk-device" {
  program = ["bash", "${path.module}/get-cwagent-device.sh"]
  query = {
    input = var.ec2-instance-id
  }
}

resource "aws_cloudwatch_metric_alarm" "ec2-disk_free" {
  alarm_name                = "${var.cw-alarm-prefix}:EC2:disk_free:${var.ec2-instance-id}"
  comparison_operator       = "LessThanThreshold"
  evaluation_periods        = "2"
  metric_name               = "disk_free"
  period                    = "900"
  statistic                 = "Average"
  threshold                 = var.threshold-disk_free
  alarm_description         = "EC2:disk_free"
  namespace                 = "CWAgent"
  insufficient_data_actions = []
  actions_enabled           = var.actions-enabled
  alarm_actions             = [var.sns-targets.alarm-actions-urgent]
  ok_actions                = [var.sns-targets.alarm-actions-urgent]
  dimensions = {
    InstanceId   = var.ec2-instance-id
    ImageId      = data.aws_instance.ec2-instance.ami
    InstanceType = data.aws_instance.ec2-instance.instance_type
    device       = data.external.disk-device.result.device
    fstype       = data.external.disk-device.result.fstype
    path         = "/"
  }
  tags = var.default-tags
  lifecycle {
    ignore_changes = [tags]
  }
}


resource "aws_cloudwatch_metric_alarm" "ec2-disk_inodes_free" {
  alarm_name                = "${var.cw-alarm-prefix}:EC2:disk_inodes_free:${var.ec2-instance-id}"
  comparison_operator       = "LessThanThreshold"
  evaluation_periods        = "2"
  metric_name               = "disk_inodes_free"
  period                    = "300"
  statistic                 = "Average"
  threshold                 = var.threshold-disk_inodes_free
  alarm_description         = "EC2:disk_inodes_free"
  namespace                 = "CWAgent"
  insufficient_data_actions = []
  actions_enabled           = var.actions-enabled
  alarm_actions             = [var.sns-targets.alarm-actions-urgent]
  ok_actions                = [var.sns-targets.alarm-actions-urgent]
  dimensions = {
    InstanceId   = var.ec2-instance-id
    ImageId      = data.aws_instance.ec2-instance.ami
    InstanceType = data.aws_instance.ec2-instance.instance_type
    device       = data.external.disk-device.result.device
    fstype       = data.external.disk-device.result.fstype
    path         = "/"
  }
  tags = var.default-tags
  lifecycle {
    ignore_changes = [tags]
  }
}

resource "aws_cloudwatch_metric_alarm" "ec2-processes_total" {
  alarm_name                = "${var.cw-alarm-prefix}:EC2:processes_total:${var.ec2-instance-id}"
  comparison_operator       = "GreaterThanThreshold"
  evaluation_periods        = "2"
  metric_name               = "processes_total"
  period                    = "900"
  statistic                 = "Average"
  threshold                 = var.threshold-processes_total
  alarm_description         = "EC2:processes_total"
  namespace                 = "CWAgent"
  insufficient_data_actions = []
  actions_enabled           = var.actions-enabled
  alarm_actions             = [var.sns-targets.alarm-actions-urgent]
  ok_actions                = [var.sns-targets.alarm-actions-urgent]
  dimensions = {
    InstanceId   = var.ec2-instance-id
    ImageId      = data.aws_instance.ec2-instance.ami
    InstanceType = data.aws_instance.ec2-instance.instance_type
  }
  tags = var.default-tags
  lifecycle {
    ignore_changes = [tags]
  }
}