Merge pull request #197 from stackrox/master

sync: master to stage

resources/prometheus/prometheus-rules.yaml

@@ -291,6 +291,37 @@ spec:
             description: |
               Remaining DB instances: {{ with query "acs_fleetshard_central_db_instances_max - acs_fleetshard_central_db_instances_used" }}{{ . | first | value | humanize }}{{ end }}. A quota increase must be requested from AWS.
             sop_url: "https://gitlab.cee.redhat.com/stackrox/acs-managed-service-runbooks/blob/master/sops/dp-029-increase-aws-rds-limits.md"
+
+    - name: aws-ses
+      rules:
+        - alert: AWSSESReputationBounceRateTooHigh
+          expr: avg_over_time(aws_ses_reputation_bounce_rate_sum[1h]) > 0.05
+          labels:
+            severity: warning
+          annotations:
+            summary: 'AWS SES Bounce rate too high'
+            description: 'The SES bounce rate is {{ $value | humanizePercentage }}, which is higher than the maximum limit of 5%.'
+            sop_url: "https://gitlab.cee.redhat.com/stackrox/acs-managed-service-runbooks/blob/master/sops/dp-040-aws-ses-violation.md"
+        - alert: AWSSESReputationComplaintRateTooHigh
+          expr: avg_over_time(aws_ses_reputation_complaint_rate_sum[1h]) > 0.01
+          labels:
+            severity: warning
+          annotations:
+            summary: 'AWS SES Complaint rate too high'
+            description: 'The SES complaint rate is {{ $value | humanizePercentage }}, which is higher than the maximum limit of 1%.'
+            sop_url: "https://gitlab.cee.redhat.com/stackrox/acs-managed-service-runbooks/blob/master/sops/dp-040-aws-ses-violation.md"
+        - alert: AWSSESSendRateTooHigh
+          # cloudwatch is setup to collect aws_ses_send_sum in 2 minute intervals (120 seconds)
+          # and the max send rate is 14/s.
+          expr: (max_over_time(aws_ses_send_sum[1h]) / 120) > 12
+          labels:
+              severity: warning
+          annotations:
+              summary: 'AWS SES Send rate too high'
+              description: 'The maximum send rate over the last hour is {{ $value }} messages/second, which is dangerously approaching the maximum limit of 14 per second.'
+              sop_url: "https://gitlab.cee.redhat.com/stackrox/acs-managed-service-runbooks/blob/master/sops/dp-040-aws-ses-violation.md"
+
+
     - name: rhacs-probe
       rules:
         - alert: RHACSProbeRunFailed

resources/prometheus/unit_tests/AWSSESQuota.yaml

@@ -0,0 +1,60 @@
+rule_files:
+  - /tmp/prometheus-rules-test.yaml
+
+evaluation_interval: 1m
+
+tests:
+  - interval: 1m
+    input_series:
+      - series: aws_ses_reputation_bounce_rate_sum
+        values: "0.05+0x60 0.06+0x90"
+    alert_rule_test:
+      - eval_time: 60m
+        alertname: AWSSESReputationBounceRateTooHigh
+        exp_alerts: []
+      - eval_time: 130m
+        alertname: AWSSESReputationBounceRateTooHigh
+        exp_alerts:
+          - exp_labels:
+              alertname: AWSSESReputationBounceRateTooHigh
+              severity: warning
+            exp_annotations:
+              summary: 'AWS SES Bounce rate too high'
+              description: 'The SES bounce rate is 6%, which is higher than the maximum limit of 5%.'
+              sop_url: "https://gitlab.cee.redhat.com/stackrox/acs-managed-service-runbooks/blob/master/sops/dp-040-aws-ses-violation.md"
+  - interval: 1m
+    input_series:
+      - series: aws_ses_reputation_complaint_rate_sum
+        values: "0.01+0x60 0.02+0x90"
+    alert_rule_test:
+      - eval_time: 60m
+        alertname: AWSSESReputationComplaintRateTooHigh
+        exp_alerts: []
+      - eval_time: 130m
+        alertname: AWSSESReputationComplaintRateTooHigh
+        exp_alerts:
+          - exp_labels:
+              alertname: AWSSESReputationComplaintRateTooHigh
+              severity: warning
+            exp_annotations:
+              summary: 'AWS SES Complaint rate too high'
+              description: 'The SES complaint rate is 2%, which is higher than the maximum limit of 1%.'
+              sop_url: "https://gitlab.cee.redhat.com/stackrox/acs-managed-service-runbooks/blob/master/sops/dp-040-aws-ses-violation.md"
+  - interval: 2m
+    input_series:
+      - series: aws_ses_send_sum
+        values: "0+0x5 1680x20"
+    alert_rule_test:
+      - eval_time: 10m
+        alertname: AWSSESSendRateTooHigh
+        exp_alerts: []
+      - eval_time: 30m
+        alertname: AWSSESSendRateTooHigh
+        exp_alerts:
+          - exp_labels:
+              alertname: AWSSESSendRateTooHigh
+              severity: warning
+            exp_annotations:
+              summary: 'AWS SES Send rate too high'
+              description: 'The maximum send rate over the last hour is 14 messages/second, which is dangerously approaching the maximum limit of 14 per second.'
+              sop_url: "https://gitlab.cee.redhat.com/stackrox/acs-managed-service-runbooks/blob/master/sops/dp-040-aws-ses-violation.md"