chore: update alarm names/descriptions/metric filters to be more accu… (

#1131) * chore: update alarm names/descriptions/metric filters to be more accurate * chore: get rid of the sqs prefix as it doesnt really make any sense * chore: renamed things one more time!
cds-snc · Jan 24, 2024 · 9876564 · 9876564
1 parent c6e119e
commit 9876564
Show file tree

Hide file tree

Showing 2 changed files with 13 additions and 13 deletions.
diff --git a/aws/eks/cloudwatch_alarms.tf b/aws/eks/cloudwatch_alarms.tf
@@ -908,29 +908,29 @@ resource "aws_cloudwatch_metric_alarm" "karpenter-replicas-unavailable" {
   }
 }
 
-resource "aws_cloudwatch_metric_alarm" "queues-not-active-1-minute-warning" {
+resource "aws_cloudwatch_metric_alarm" "aggregating-queues-not-active-1-minute-warning" {
   count               = var.cloudwatch_enabled ? 1 : 0
-  alarm_name          = "queues-not-active-1-minute-warning"
-  alarm_description   = "Queues have not been active for one minute"
+  alarm_name          = "aggregating-queues-not-active-1-minute-warning"
+  alarm_description   = "Beat inbox tasks have not been active for one minute"
   comparison_operator = "LessThanThreshold"
   evaluation_periods  = "1"
-  metric_name         = aws_cloudwatch_log_metric_filter.queues-are-active[0].metric_transformation[0].name
-  namespace           = aws_cloudwatch_log_metric_filter.queues-are-active[0].metric_transformation[0].namespace
+  metric_name         = aws_cloudwatch_log_metric_filter.aggregating-queues-are-active[0].metric_transformation[0].name
+  namespace           = aws_cloudwatch_log_metric_filter.aggregating-queues-are-active[0].metric_transformation[0].namespace
   period              = "60"
   statistic           = "Sum"
   threshold           = 1
   treat_missing_data  = "breaching"
   alarm_actions       = [var.sns_alert_warning_arn]
 }
 
-resource "aws_cloudwatch_metric_alarm" "queues-not-active-5-minutes-critical" {
+resource "aws_cloudwatch_metric_alarm" "aggregating-queues-not-active-5-minutes-critical" {
   count               = var.cloudwatch_enabled ? 1 : 0
-  alarm_name          = "queues-not-active-5-minutes-critical"
-  alarm_description   = "Queues have not been active for 5 minutes"
+  alarm_name          = "aggregating-queues-not-active-5-minutes-critical"
+  alarm_description   = "Beat inbox tasks have not been active for 5 minutes"
   comparison_operator = "LessThanThreshold"
   evaluation_periods  = "1"
-  metric_name         = aws_cloudwatch_log_metric_filter.queues-are-active[0].metric_transformation[0].name
-  namespace           = aws_cloudwatch_log_metric_filter.queues-are-active[0].metric_transformation[0].namespace
+  metric_name         = aws_cloudwatch_log_metric_filter.aggregating-queues-are-active[0].metric_transformation[0].name
+  namespace           = aws_cloudwatch_log_metric_filter.aggregating-queues-are-active[0].metric_transformation[0].namespace
   period              = "300"
   statistic           = "Sum"
   threshold           = 1

diff --git a/aws/eks/cloudwatch_log.tf b/aws/eks/cloudwatch_log.tf
@@ -154,14 +154,14 @@ resource "aws_cloudwatch_log_metric_filter" "documentation-evicted-pods" {
   }
 }
 
-resource "aws_cloudwatch_log_metric_filter" "queues-are-active" {
+resource "aws_cloudwatch_log_metric_filter" "aggregating-queues-are-active" {
   count          = var.cloudwatch_enabled ? 1 : 0
-  name           = "queues-are-active"
+  name           = "aggregating-queues-are-active"
   pattern        = "Batch saving with"
   log_group_name = aws_cloudwatch_log_group.notification-canada-ca-eks-application-logs[0].name
 
   metric_transformation {
-    name      = "queues-are-active"
+    name      = "aggregating-queues-are-active"
     namespace = "LogMetrics"
     value     = "1"
   }