Merge branch 'main' into fix/callback-failure-queries

cds-snc · Oct 1, 2024 · 913f823 · 913f823
2 parents 8b44986 + ff2ef9e
commit 913f823
Show file tree

Hide file tree

Showing 4 changed files with 46 additions and 3 deletions.
diff --git a/.github/workflows/infrastructure_version.txt b/.github/workflows/infrastructure_version.txt
@@ -1 +1 @@
-2.16.21
+2.16.24
diff --git a/.github/workflows/ossf-scorecard.yml b/.github/workflows/ossf-scorecard.yml
@@ -25,7 +25,7 @@ jobs:
           persist-credentials: false
 
       - name: "Run analysis"
-        uses: ossf/scorecard-action@6c4912ed9e5f80cfda40164b92753f21f0892cab
+        uses: ossf/scorecard-action@72803a12483ed6f4f7c34f804818169f50162e37
         with:
           results_file: ossf-results.json
           results_format: json

diff --git a/aws/eks/cloudwatch_alarms.tf b/aws/eks/cloudwatch_alarms.tf
@@ -972,7 +972,7 @@ resource "aws_cloudwatch_metric_alarm" "service-callback-too-many-failures-warni
 
 resource "aws_cloudwatch_metric_alarm" "service-callback-too-many-failures-critical" {
   count               = var.cloudwatch_enabled ? 1 : 0
-  alarm_name          = "service-callback-too-many-failures-warning"
+  alarm_name          = "service-callback-too-many-failures-critical"
   alarm_description   = "Service reached the max number of callback retries 100 times in 10 minutes"
   comparison_operator = "GreaterThanOrEqualToThreshold"
   evaluation_periods  = "1"
@@ -984,3 +984,33 @@ resource "aws_cloudwatch_metric_alarm" "service-callback-too-many-failures-criti
   treat_missing_data  = "notBreaching"
   alarm_actions       = [var.sns_alert_critical_arn]
 }
+
+resource "aws_cloudwatch_metric_alarm" "throttling-exception-warning" {
+  count               = var.cloudwatch_enabled ? 1 : 0
+  alarm_name          = "throttling-exception-warning"
+  alarm_description   = "Have received a throttling exception in the last minute"
+  comparison_operator = "GreaterThanOrEqualToThreshold"
+  evaluation_periods  = "1"
+  metric_name         = aws_cloudwatch_log_metric_filter.throttling-exceptions[0].metric_transformation[0].name
+  namespace           = aws_cloudwatch_log_metric_filter.throttling-exceptions[0].metric_transformation[0].namespace
+  period              = 60
+  statistic           = "Sum"
+  threshold           = 1
+  treat_missing_data  = "notBreaching"
+  alarm_actions       = [var.sns_alert_warning_arn]
+}
+
+resource "aws_cloudwatch_metric_alarm" "many-throttling-exceptions-warning" {
+  count               = var.cloudwatch_enabled ? 1 : 0
+  alarm_name          = "many-throttling-exceptions-warning"
+  alarm_description   = "Have received 100 throttling exception in the last minute"
+  comparison_operator = "GreaterThanOrEqualToThreshold"
+  evaluation_periods  = "1"
+  metric_name         = aws_cloudwatch_log_metric_filter.throttling-exceptions[0].metric_transformation[0].name
+  namespace           = aws_cloudwatch_log_metric_filter.throttling-exceptions[0].metric_transformation[0].namespace
+  period              = 60
+  statistic           = "Sum"
+  threshold           = 100
+  treat_missing_data  = "notBreaching"
+  alarm_actions       = [var.sns_alert_warning_arn]
+}
diff --git a/aws/eks/cloudwatch_log.tf b/aws/eks/cloudwatch_log.tf
@@ -192,3 +192,16 @@ resource "aws_cloudwatch_log_metric_filter" "callback-request-failures" {
     value     = "1"
   }
 }
+
+resource "aws_cloudwatch_log_metric_filter" "throttling-exceptions" {
+  count          = var.cloudwatch_enabled ? 1 : 0
+  name           = "throttling-exceptions"
+  pattern        = "ThrottlingException"
+  log_group_name = aws_cloudwatch_log_group.notification-canada-ca-eks-application-logs[0].name
+
+  metric_transformation {
+    name      = "throttling-exceptions"
+    namespace = "LogMetrics"
+    value     = "1"
+  }
+}