Nomis: tweak nomis alarm thresholds to allow for DB refershes (#4431)

* Tweak nomis alarms to allow for DB refershes * typo
ministryofjustice · Jan 3, 2024 · 41c6c3c · 41c6c3c
1 parent 078db3e
commit 41c6c3c
Show file tree

Hide file tree

Showing 2 changed files with 10 additions and 10 deletions.
diff --git a/terraform/environments/nomis/locals_database.tf b/terraform/environments/nomis/locals_database.tf
@@ -31,18 +31,18 @@ locals {
       module.baseline_presets.cloudwatch_metric_alarms_by_sns_topic["dso_pagerduty"].ec2_instance_cwagent_collectd_service_status_os,
       module.baseline_presets.cloudwatch_metric_alarms_by_sns_topic["dba_pagerduty"].ec2_instance_cwagent_collectd_service_status_app,
       local.environment == "production" ? {} : {
-        cpu-utilization-high = {
-          comparison_operator = "GreaterThanOrEqualToThreshold"
+        cpu-utilization-high = merge(module.baseline_presets.cloudwatch_metric_alarms_by_sns_topic["dba_pagerduty"].ec2["cpu-utilization-high"], {
           evaluation_periods  = "480"
           datapoints_to_alarm = "480"
-          metric_name         = "CPUUtilization"
-          namespace           = "AWS/EC2"
-          period              = "60"
-          statistic           = "Maximum"
           threshold           = "95"
-          alarm_description   = "Triggers if the average cpu remains at 95% utilization or above for 8 hours to allow for DB refreshes"
-          alarm_actions       = ["dba_pagerduty"]
-        }
+          alarm_description   = "Triggers if the average cpu remains at 95% utilization or above for 8 hours to allow for DB refreshes. See https://dsdmoj.atlassian.net/wiki/spaces/DSTT/pages/4326064583"
+        })
+        cpu-iowait-high = merge(module.baseline_presets.cloudwatch_metric_alarms_by_sns_topic["dba_pagerduty"].ec2_cwagent_linux["cpu-iowait-high"], {
+          evaluation_periods  = "480"
+          datapoints_to_alarm = "480"
+          threshold           = "40"
+          alarm_description   = "Triggers if the amount of CPU time spent waiting for I/O to complete is continually high for 8 hours allowing for DB refreshes.  See https://dsdmoj.atlassian.net/wiki/spaces/DSTT/pages/4325900634"
+        })
       },
     )
     connectivity_test = merge(

diff --git a/terraform/modules/baseline_presets/cloudwatch_metric_alarms.tf b/terraform/modules/baseline_presets/cloudwatch_metric_alarms.tf
@@ -31,7 +31,7 @@ locals {
         period              = "86400"
         statistic           = "Minimum"
         threshold           = "14"
-        alarm_description   = "Triggers if an ACM certificate has notautomatically renewed and is expiring soon. Automatic renewal should happen 60 days prior to expiration. See https://dsdmoj.atlassian.net/wiki/spaces/DSTT/pages/4615340266"
+        alarm_description   = "Triggers if an ACM certificate has not automatically renewed and is expiring soon. Automatic renewal should happen 60 days prior to expiration. See https://dsdmoj.atlassian.net/wiki/spaces/DSTT/pages/4615340266"
         alarm_actions       = var.options.cloudwatch_metric_alarms_default_actions
       }
     }