ROX-15136: filter aws metrics by cluster name (#857)

stackrox · Apr 11, 2023 · f42db1e · f42db1e
1 parent 7f665da
commit f42db1e
Show file tree

Hide file tree

Showing 3 changed files with 136 additions and 216 deletions.
diff --git a/dp-terraform/helm/rhacs-terraform/charts/cloudwatch/templates/01-operator-03-config-map.yaml b/dp-terraform/helm/rhacs-terraform/charts/cloudwatch/templates/01-operator-03-config-map.yaml
@@ -4,215 +4,134 @@ metadata:
   name: cloudwatch-exporter-config
   namespace: {{ include "cloudwatch.namespace" . }}
 data:
-  config.yaml: |-
-    region: us-east-1
-    metrics:
-      - aws_metric_name: DatabaseConnections
-        aws_namespace: AWS/RDS
-        aws_dimensions:
-          - DBInstanceIdentifier
-        aws_statistics:
-          - Average
-          - Maximum
-          - Minimum
-        aws_extended_statistics:
-          - p90
-          - p95
-          - p99
-        use_get_metric_data: true
-
-      - aws_metric_name: ServerlessDatabaseCapacity
-        aws_namespace: AWS/RDS
-        aws_dimensions:
-          - DBInstanceIdentifier
-        aws_statistics:
-          - Average
-          - Maximum
-          - Minimum
-        aws_extended_statistics:
-          - p90
-          - p95
-          - p99
-        use_get_metric_data: true
-
-      - aws_metric_name: ACUUtilization
-        aws_namespace: AWS/RDS
-        aws_dimensions:
-          - DBInstanceIdentifier
-        aws_statistics:
-          - Average
-          - Maximum
-          - Minimum
-        aws_extended_statistics:
-          - p90
-          - p95
-          - p99
-        use_get_metric_data: true
-
-      - aws_metric_name: FreeableMemory
-        aws_namespace: AWS/RDS
-        aws_dimensions:
-          - DBInstanceIdentifier
-        aws_statistics:
-          - Average
-          - Maximum
-          - Minimum
-        aws_extended_statistics:
-          - p90
-          - p95
-          - p99
-        use_get_metric_data: true
-
-      - aws_metric_name: CPUUtilization
-        aws_namespace: AWS/RDS
-        aws_dimensions:
-          - DBInstanceIdentifier
-        aws_statistics:
-          - Average
-          - Maximum
-          - Minimum
-        aws_extended_statistics:
-          - p90
-          - p95
-          - p99
-        use_get_metric_data: true
-
-      - aws_metric_name: ReadLatency
-        aws_namespace: AWS/RDS
-        aws_dimensions:
-          - DBInstanceIdentifier
-        aws_statistics:
-          - Average
-          - Maximum
-          - Minimum
-        aws_extended_statistics:
-          - p90
-          - p95
-          - p99
-        use_get_metric_data: true
-
-      - aws_metric_name: ReadThroughput
-        aws_namespace: AWS/RDS
-        aws_dimensions:
-          - DBInstanceIdentifier
-        aws_statistics:
-          - Average
-          - Maximum
-          - Minimum
-        aws_extended_statistics:
-          - p90
-          - p95
-          - p99
-        use_get_metric_data: true
-
-      - aws_metric_name: WriteLatency
-        aws_namespace: AWS/RDS
-        aws_dimensions:
-          - DBInstanceIdentifier
-        aws_statistics:
-          - Average
-          - Maximum
-          - Minimum
-        aws_extended_statistics:
-          - p90
-          - p95
-          - p99
-        use_get_metric_data: true
-
-      - aws_metric_name: WriteThroughput
-        aws_namespace: AWS/RDS
-        aws_dimensions:
-          - DBInstanceIdentifier
-        aws_statistics:
-          - Average
-          - Maximum
-          - Minimum
-        aws_extended_statistics:
-          - p90
-          - p95
-          - p99
-        use_get_metric_data: true
-
-      - aws_metric_name: NetworkThroughput
-        aws_namespace: AWS/RDS
-        aws_dimensions:
-          - DBInstanceIdentifier
-        aws_statistics:
-          - Average
-          - Maximum
-          - Minimum
-        aws_extended_statistics:
-          - p90
-          - p95
-          - p99
-        use_get_metric_data: true
-
-      - aws_metric_name: AuroraReplicaLag
-        aws_namespace: AWS/RDS
-        aws_dimensions:
-          - DBInstanceIdentifier
-        aws_statistics:
-          - Average
-          - Maximum
-          - Minimum
-        aws_extended_statistics:
-          - p90
-          - p95
-          - p99
-        use_get_metric_data: true
-
-      - aws_metric_name: MaximumUsedTransactionIDs
-        aws_namespace: AWS/RDS
-        aws_dimensions:
-          - DBInstanceIdentifier
-        aws_statistics:
-          - Average
-          - Maximum
-          - Minimum
-        aws_extended_statistics:
-          - p90
-          - p95
-          - p99
-        use_get_metric_data: true
-
-      - aws_metric_name: TransactionLogsDiskUsage
-        aws_namespace: AWS/RDS
-        aws_dimensions:
-          - DBInstanceIdentifier
-        aws_statistics:
-          - Average
-          - Maximum
-          - Minimum
-        aws_extended_statistics:
-          - p90
-          - p95
-          - p99
-        use_get_metric_data: true
-
-      - aws_metric_name: Deadlocks
-        aws_namespace: AWS/RDS
-        aws_dimensions:
-          - DBInstanceIdentifier
-        aws_statistics:
-          - Average
-          - Maximum
-          - Minimum
-        aws_extended_statistics:
-          - p90
-          - p95
-          - p99
-        use_get_metric_data: true
-
-      - aws_metric_name: BufferCacheHitRatio
-        aws_namespace: AWS/RDS
-        aws_dimensions:
-          - DBInstanceIdentifier
-        aws_statistics:
-          - Average
-          - Maximum
-          - Minimum
-        aws_extended_statistics:
-          - p90
-          - p95
-          - p99
-        use_get_metric_data: true
+  config.yml: |-
+    apiVersion: v1alpha1
+    discovery:
+      jobs:
+        - type: rds
+          regions:
+            - us-east-1
+          searchTags:
+            - key: DataplaneClusterName
+              value: ^{{ .Values.clusterName }}$
+          metrics:
+            - name: DatabaseConnections
+              statistics:
+                - Average
+                - Maximum
+                - Minimum
+                - p90
+                - p95
+                - p99
+            - name: ServerlessDatabaseCapacity
+              statistics:
+                - Average
+                - Maximum
+                - Minimum
+                - p90
+                - p95
+                - p99
+            - name: ACUUtilization
+              statistics:
+                - Average
+                - Maximum
+                - Minimum
+                - p90
+                - p95
+                - p99
+            - name: FreeableMemory
+              statistics:
+                - Average
+                - Maximum
+                - Minimum
+                - p90
+                - p95
+                - p99
+            - name: CPUUtilization
+              statistics:
+                - Average
+                - Maximum
+                - Minimum
+                - p90
+                - p95
+                - p99
+            - name: ReadLatency
+              statistics:
+                - Average
+                - Maximum
+                - Minimum
+                - p90
+                - p95
+                - p99
+            - name: ReadThroughput
+              statistics:
+                - Average
+                - Maximum
+                - Minimum
+                - p90
+                - p95
+                - p99
+            - name: WriteLatency
+              statistics:
+                - Average
+                - Maximum
+                - Minimum
+                - p90
+                - p95
+                - p99
+            - name: WriteThroughput
+              statistics:
+                - Average
+                - Maximum
+                - Minimum
+                - p90
+                - p95
+                - p99
+            - name: NetworkThroughput
+              statistics:
+                - Average
+                - Maximum
+                - Minimum
+                - p90
+                - p95
+                - p99
+            - name: AuroraReplicaLag
+              statistics:
+                - Average
+                - Maximum
+                - Minimum
+                - p90
+                - p95
+                - p99
+            - name: MaximumUsedTransactionIDs
+              statistics:
+                - Average
+                - Maximum
+                - Minimum
+                - p90
+                - p95
+                - p99
+            - name: TransactionLogsDiskUsage
+              statistics:
+                - Average
+                - Maximum
+                - Minimum
+                - p90
+                - p95
+                - p99
+            - name: Deadlocks
+              statistics:
+                - Average
+                - Maximum
+                - Minimum
+                - p90
+                - p95
+                - p99
+            - name: BufferCacheHitRatio
+              statistics:
+                - Average
+                - Maximum
+                - Minimum
+                - p90
+                - p95
+                - p99
diff --git a/dp-terraform/helm/rhacs-terraform/charts/cloudwatch/templates/01-operator-04-deployment.yaml b/dp-terraform/helm/rhacs-terraform/charts/cloudwatch/templates/01-operator-04-deployment.yaml
@@ -29,6 +29,8 @@ spec:
             runAsNonRoot: true
             seccompProfile:
               type: RuntimeDefault
+          args:
+            - "--config.file=/tmp/config.yml"
           env:
             - name: AWS_ACCESS_KEY_ID
               valueFrom:
@@ -41,12 +43,11 @@ spec:
                   name: rhacs-cloudwatch-exporter
                   key: AWS_SECRET_ACCESS_KEY
           volumeMounts:
-            - mountPath: /config/config.yml
+            - mountPath: /tmp
               name: config
-              subPath: config.yaml
           ports:
             - name: monitoring
-              containerPort: 9106
+              containerPort: 5000
       volumes:
         - name: config
           configMap:

diff --git a/dp-terraform/helm/rhacs-terraform/charts/cloudwatch/values.yaml b/dp-terraform/helm/rhacs-terraform/charts/cloudwatch/values.yaml
@@ -5,4 +5,4 @@ aws:
 
 clusterName: ""
 environment: ""
-image: "quay.io/prometheus/cloudwatch-exporter:v0.15.1"
+image: "ghcr.io/nerdswords/yet-another-cloudwatch-exporter:v0.48.0-alpha"