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

// The following checks requires container insights

resource "aws_cloudwatch_metric_alarm" "eks-pod_cpu_utilization" {
  for_each                  = toset(var.pod-names)
  alarm_name                = "${var.cw-alarm-prefix}:EKS:${var.cluster-name}:${each.value}:${var.settings.alarm1.metric}"
  comparison_operator       = var.settings.alarm1.comparison_operator
  evaluation_periods        = var.settings.alarm1.evaluation_periods
  metric_name               = var.settings.alarm1.metric
  period                    = var.settings.alarm1.period
  statistic                 = var.settings.alarm1.statistic
  threshold                 = var.settings.alarm1.threshold
  alarm_description         = "EKS:${var.settings.alarm1.metric}"
  namespace                 = "ContainerInsights"
  insufficient_data_actions = []
  actions_enabled           = var.actions-enabled
  alarm_actions             = [var.settings.alarm1.action]
  ok_actions                = [var.settings.alarm1.action]
  dimensions = {
    "PodName"     = each.value
    "ClusterName" = var.cluster-name
    "Namespace"   = var.eks-namespace
  }
  tags = var.default-tags
  lifecycle {
    ignore_changes = [tags]
  }
}

resource "aws_cloudwatch_metric_alarm" "eks-pod_memory_utilization" {
  for_each = toset(var.pod-names)

  alarm_name                = "${var.cw-alarm-prefix}:EKS:${var.cluster-name}:${each.value}:${var.settings.alarm2.metric}"
  comparison_operator       = "GreaterThanThreshold"
  evaluation_periods        = "3"
  metric_name               = var.settings.alarm2.metric
  period                    = var.settings.alarm2.period
  statistic                 = var.settings.alarm2.statistic
  threshold                 = var.settings.alarm2.threshold
  alarm_description         = "EKS:${var.settings.alarm2.metric}"
  namespace                 = "ContainerInsights"
  insufficient_data_actions = []
  actions_enabled           = var.actions-enabled
  alarm_actions             = [var.settings.alarm2.action]
  ok_actions                = [var.settings.alarm2.action]
  dimensions = {
    "PodName"     = each.value
    "ClusterName" = var.cluster-name
    "Namespace"   = var.eks-namespace
  }
  tags = var.default-tags
  lifecycle {
    ignore_changes = [tags]
  }
}

resource "aws_cloudwatch_metric_alarm" "eks-pod_number_of_container_restarts" {
  for_each = toset(var.pod-names)

  alarm_name                = "${var.cw-alarm-prefix}:EKS:${var.cluster-name}:${each.value}:${var.settings.alarm3.metric}"
  comparison_operator       = "GreaterThanThreshold"
  evaluation_periods        = "3"
  metric_name               = var.settings.alarm3.metric
  period                    = var.settings.alarm3.period
  statistic                 = var.settings.alarm3.statistic
  threshold                 = var.settings.alarm3.threshold
  alarm_description         = "EKS:${var.settings.alarm3.metric}"
  namespace                 = "ContainerInsights"
  insufficient_data_actions = []
  actions_enabled           = var.actions-enabled
  alarm_actions             = [var.settings.alarm3.action]
  ok_actions                = [var.settings.alarm3.action]
  dimensions = {
    "PodName"     = each.value
    "ClusterName" = var.cluster-name
    "Namespace"   = var.eks-namespace
  }
  tags = var.default-tags
  lifecycle {
    ignore_changes = [tags]
  }
}