data "template_file" "filter" { template = "$${filter}" vars { filter = "${var.filter_tags_use_defaults == "true" ? format("dd_monitoring:enabled,dd_aws_elb:enabled,env:%s", var.environment) : "${var.filter_tags_custom}"}" } } resource "datadog_monitor" "ELB_no_healthy_instances" { name = "[${var.environment}] ELB no healthy instances" message = "${var.message}" query = < ${var.elb_4xx_threshold_critical} EOF type = "metric alert" thresholds { warning = "${var.elb_4xx_threshold_warning}" critical = "${var.elb_4xx_threshold_critical}" } notify_no_data = false evaluation_delay = "${var.evaluation_delay}" renotify_interval = 0 notify_audit = false timeout_h = 0 include_tags = true locked = false require_full_window = false new_host_delay = "${var.evaluation_delay}" no_data_timeframe = 20 tags = ["env:${var.environment}", "resource:elb", "team:aws", "provider:aws"] } resource "datadog_monitor" "ELB_too_much_5xx" { name = "[${var.environment}] ELB 5xx errors too high {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)" message = "${var.message}" query = < ${var.elb_5xx_threshold_critical} EOF type = "metric alert" thresholds { warning = "${var.elb_5xx_threshold_warning}" critical = "${var.elb_5xx_threshold_critical}" } notify_no_data = false evaluation_delay = "${var.evaluation_delay}" renotify_interval = 0 notify_audit = false timeout_h = 0 include_tags = true locked = false require_full_window = false new_host_delay = "${var.evaluation_delay}" no_data_timeframe = 20 tags = ["env:${var.environment}", "resource:elb", "team:aws", "provider:aws"] } resource "datadog_monitor" "ELB_too_much_4xx_backend" { name = "[${var.environment}] ELB backend 4xx errors too high {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)" message = "${var.message}" query = < ${var.elb_backend_4xx_threshold_critical} EOF type = "metric alert" thresholds { warning = "${var.elb_backend_4xx_threshold_warning}" critical = "${var.elb_backend_4xx_threshold_critical}" } notify_no_data = false evaluation_delay = "${var.evaluation_delay}" renotify_interval = 0 notify_audit = false timeout_h = 0 include_tags = true locked = false require_full_window = false new_host_delay = "${var.evaluation_delay}" no_data_timeframe = 20 tags = ["env:${var.environment}", "resource:elb", "team:aws", "provider:aws"] } resource "datadog_monitor" "ELB_too_much_5xx_backend" { name = "[${var.environment}] ELB backend 5xx errors too high {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)" message = "${var.message}" query = < ${var.elb_backend_5xx_threshold_critical} EOF type = "metric alert" thresholds { warning = "${var.elb_backend_5xx_threshold_warning}" critical = "${var.elb_backend_5xx_threshold_critical}" } notify_no_data = false evaluation_delay = "${var.evaluation_delay}" renotify_interval = 0 notify_audit = false timeout_h = 0 include_tags = true locked = false require_full_window = false new_host_delay = "${var.evaluation_delay}" no_data_timeframe = 20 tags = ["env:${var.environment}", "resource:elb", "team:aws", "provider:aws"] } resource "datadog_monitor" "ELB_backend_latency" { name = "[${var.environment}] ELB latency too high {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)" message = "${var.message}" query = < ${var.elb_backend_latency_critical} EOF type = "metric alert" thresholds { warning = "${var.elb_backend_latency_warning}" critical = "${var.elb_backend_latency_critical}" } notify_no_data = false evaluation_delay = "${var.evaluation_delay}" renotify_interval = 0 notify_audit = false timeout_h = 0 include_tags = true locked = false require_full_window = false new_host_delay = "${var.evaluation_delay}" no_data_timeframe = 20 tags = ["env:${var.environment}", "resource:elb", "team:aws", "provider:aws"] }