Split alarm modules into files

Author: francisco-videira-nhs

infrastructure/terraform/modules/alarms/alarms-apigw/alarm-5xx.tf

@@ -0,0 +1,22 @@
+resource "aws_cloudwatch_metric_alarm" "five_xx" {
+  alarm_name        = "${var.alarm_prefix}-apigw-5xx"
+  alarm_description = "RELIABILITY: API Gateway 5xx responses"
+
+  namespace   = "AWS/ApiGateway"
+  metric_name = "5XXError"
+  statistic   = "Sum"
+  period      = var.error_5xx_period_seconds
+
+  evaluation_periods  = var.error_5xx_evaluation_periods
+  threshold           = var.error_5xx_threshold
+  comparison_operator = "GreaterThanThreshold"
+  treat_missing_data  = "notBreaching"
+
+  dimensions = local.api_dimensions
+
+  actions_enabled           = false
+  alarm_actions             = []
+  ok_actions                = []
+  insufficient_data_actions = []
+  tags                      = var.tags
+}

infrastructure/terraform/modules/alarms/alarms-apigw/alarm-latency-anomaly.tf

@@ -0,0 +1,34 @@
+resource "aws_cloudwatch_metric_alarm" "latency_anomaly" {
+  alarm_name          = "${var.alarm_prefix}-apigw-latency-anomaly"
+  alarm_description   = "RELIABILITY: API Gateway latency anomaly"
+  comparison_operator = "GreaterThanUpperThreshold"
+  evaluation_periods  = var.latency_evaluation_periods
+  datapoints_to_alarm = var.latency_datapoints_to_alarm
+  threshold_metric_id = "ad1"
+  treat_missing_data  = "notBreaching"
+
+  actions_enabled           = false
+  alarm_actions             = []
+  ok_actions                = []
+  insufficient_data_actions = []
+  tags                      = var.tags
+
+  metric_query {
+    id = "m1"
+    metric {
+      metric_name = "Latency"
+      namespace   = "AWS/ApiGateway"
+      stat        = "Average"
+      period      = var.latency_period_seconds
+      dimensions  = local.api_dimensions
+    }
+    return_data = true
+  }
+
+  metric_query {
+    id          = "ad1"
+    expression  = "ANOMALY_DETECTION_BAND(m1, ${var.latency_anomaly_sensitivity})"
+    label       = "Latency (expected)"
+    return_data = true
+  }
+}

infrastructure/terraform/modules/alarms/alarms-apigw/alarm-latency-threshold.tf

@@ -0,0 +1,22 @@
+resource "aws_cloudwatch_metric_alarm" "latency_threshold" {
+  alarm_name        = "${var.alarm_prefix}-apigw-latency-threshold"
+  alarm_description = "RELIABILITY: API Gateway latency above threshold"
+
+  namespace   = "AWS/ApiGateway"
+  metric_name = "Latency"
+  statistic   = "Average"
+  period      = var.latency_period_seconds
+
+  evaluation_periods  = var.latency_evaluation_periods
+  threshold           = var.latency_threshold_ms
+  comparison_operator = "GreaterThanThreshold"
+  treat_missing_data  = "notBreaching"
+
+  dimensions = local.api_dimensions
+
+  actions_enabled           = false
+  alarm_actions             = []
+  ok_actions                = []
+  insufficient_data_actions = []
+  tags                      = var.tags
+}

infrastructure/terraform/modules/alarms/alarms-apigw/locals.tf

@@ -0,0 +1,6 @@
+locals {
+  api_dimensions = {
+    ApiName = var.api_name
+    Stage   = var.stage_name
+  }
+}

infrastructure/terraform/modules/alarms/alarms-apigw/main.tf

@@ -0,0 +1,22 @@
+resource "aws_cloudwatch_metric_alarm" "read_throttle" {
+  alarm_name        = "${var.alarm_prefix}-ddb-${var.table_name}-read-throttle"
+  alarm_description = "RELIABILITY: DynamoDB read throttling"
+
+  namespace   = "AWS/DynamoDB"
+  metric_name = "ReadThrottleEvents"
+  statistic   = "Sum"
+  period      = var.period_seconds
+
+  evaluation_periods  = var.evaluation_periods
+  threshold           = var.read_throttle_threshold
+  comparison_operator = "GreaterThanThreshold"
+  treat_missing_data  = "notBreaching"
+
+  dimensions = { TableName = var.table_name }
+
+  actions_enabled           = false
+  alarm_actions             = []
+  ok_actions                = []
+  insufficient_data_actions = []
+  tags                      = var.tags
+}

infrastructure/terraform/modules/alarms/alarms-ddb/alarm-read-throttle.tf

@@ -1,26 +1,3 @@
-resource "aws_cloudwatch_metric_alarm" "read_throttle" {
-  alarm_name        = "${var.alarm_prefix}-ddb-${var.table_name}-read-throttle"
-  alarm_description = "RELIABILITY: DynamoDB read throttling"
-
-  namespace   = "AWS/DynamoDB"
-  metric_name = "ReadThrottleEvents"
-  statistic   = "Sum"
-  period      = var.period_seconds
-
-  evaluation_periods  = var.evaluation_periods
-  threshold           = var.read_throttle_threshold
-  comparison_operator = "GreaterThanThreshold"
-  treat_missing_data  = "notBreaching"
-
-  dimensions = { TableName = var.table_name }
-
-  actions_enabled           = false
-  alarm_actions             = []
-  ok_actions                = []
-  insufficient_data_actions = []
-  tags                      = var.tags
-}
-
 resource "aws_cloudwatch_metric_alarm" "write_throttle" {
   alarm_name        = "${var.alarm_prefix}-ddb-${var.table_name}-write-throttle"
   alarm_description = "RELIABILITY: DynamoDB write throttling"

…raform/modules/alarms/alarms-ddb/main.tf …larms/alarms-ddb/alarm-write-throttle.tfinfrastructure/terraform/modules/alarms/alarms-ddb/main.tf renamed to infrastructure/terraform/modules/alarms/alarms-ddb/alarm-write-throttle.tf infrastructure/terraform/modules/alarms/alarms-ddb/main.tf renamed to infrastructure/terraform/modules/alarms/alarms-ddb/alarm-write-throttle.tf

@@ -0,0 +1,21 @@
+resource "aws_cloudwatch_metric_alarm" "error_logs" {
+  count             = var.enable_error_log_metric ? 1 : 0
+  alarm_name        = "${var.alarm_prefix}-lambda-${var.function_name}-error-logs"
+  alarm_description = "ERROR: Lambda error logs detected"
+
+  namespace   = var.error_log_metric_namespace
+  metric_name = "${var.error_log_metric_name_prefix}${var.function_name}"
+  statistic   = "Sum"
+  period      = var.period_seconds
+
+  evaluation_periods  = var.error_log_evaluation_periods
+  threshold           = var.error_log_threshold
+  comparison_operator = "GreaterThanThreshold"
+  treat_missing_data  = "notBreaching"
+
+  actions_enabled           = false
+  alarm_actions             = []
+  ok_actions                = []
+  insufficient_data_actions = []
+  tags                      = var.tags
+}

infrastructure/terraform/modules/alarms/alarms-lambda/alarm-error-logs.tf

@@ -0,0 +1,22 @@
+resource "aws_cloudwatch_metric_alarm" "errors" {
+  alarm_name        = "${var.alarm_prefix}-lambda-${var.function_name}-errors"
+  alarm_description = "ERROR: Lambda errors"
+
+  namespace   = "AWS/Lambda"
+  metric_name = "Errors"
+  statistic   = "Sum"
+  period      = var.period_seconds
+
+  evaluation_periods  = var.evaluation_periods
+  threshold           = var.errors_threshold
+  comparison_operator = "GreaterThanThreshold"
+  treat_missing_data  = "notBreaching"
+
+  dimensions = { FunctionName = var.function_name }
+
+  actions_enabled           = false
+  alarm_actions             = []
+  ok_actions                = []
+  insufficient_data_actions = []
+  tags                      = var.tags
+}

infrastructure/terraform/modules/alarms/alarms-lambda/alarm-errors.tf

@@ -0,0 +1,22 @@
+resource "aws_cloudwatch_metric_alarm" "throttles" {
+  alarm_name        = "${var.alarm_prefix}-lambda-${var.function_name}-throttles"
+  alarm_description = "RELIABILITY: Lambda throttles"
+
+  namespace   = "AWS/Lambda"
+  metric_name = "Throttles"
+  statistic   = "Sum"
+  period      = var.period_seconds
+
+  evaluation_periods  = var.evaluation_periods
+  threshold           = var.throttles_threshold
+  comparison_operator = "GreaterThanThreshold"
+  treat_missing_data  = "notBreaching"
+
+  dimensions = { FunctionName = var.function_name }
+
+  actions_enabled           = false
+  alarm_actions             = []
+  ok_actions                = []
+  insufficient_data_actions = []
+  tags                      = var.tags
+}