cloudwatch: Own job for Replication Metrics

[chef.git] / cookbooks / prometheus / templates / default / cloudwatch.yml.erb

diff --git a/cookbooks/prometheus/templates/default/cloudwatch.yml.erb b/cookbooks/prometheus/templates/default/cloudwatch.yml.erb
index 810186e0737f9000d6e4028e39d897ca9a993335..14550314d09f6fa3be2f96e0c3aeb350797123c1 100644 (file)
--- a/cookbooks/prometheus/templates/default/cloudwatch.yml.erb
+++ b/cookbooks/prometheus/templates/default/cloudwatch.yml.erb
@@ -45,14 +45,22 @@ discovery:
           statistics: [Minimum, Maximum, Average]
         - name: TotalRequestLatency
           statistics: [Minimum, Maximum, Average]
           statistics: [Minimum, Maximum, Average]
         - name: TotalRequestLatency
           statistics: [Minimum, Maximum, Average]

+        - name: OperationsFailedReplication
+          statistics: [Sum]
+    - type: AWS/S3
+      regions:
+        - eu-north-1
+      roles:
+        - roleArn: "arn:aws:iam::173189593406:role/osm-cloudwatch-export-role" # osm-main
+      period: 300
+      length: 300
+      metrics:

         - name: ReplicationLatency
           statistics: [Maximum]
         - name: BytesPendingReplication
           statistics: [Maximum]
         - name: OperationsPendingReplication
           statistics: [Maximum]
         - name: ReplicationLatency
           statistics: [Maximum]
         - name: BytesPendingReplication
           statistics: [Maximum]
         - name: OperationsPendingReplication
           statistics: [Maximum]

-        - name: OperationsFailedReplication
-          statistics: [Sum]

     - type: AWS/S3
       regions:
         - eu-central-1
     - type: AWS/S3
       regions:
         - eu-central-1


@@ -146,11 +154,42 @@ discovery:
           statistics: [Minimum, Maximum, Average]
         - name: TotalRequestLatency
           statistics: [Minimum, Maximum, Average]
           statistics: [Minimum, Maximum, Average]
         - name: TotalRequestLatency
           statistics: [Minimum, Maximum, Average]

-        - name: ReplicationLatency
-          statistics: [Maximum]
-        - name: BytesPendingReplication
-          statistics: [Maximum]
-        - name: OperationsPendingReplication
-          statistics: [Maximum]
-        - name: OperationsFailedReplication
+    - type: AWS/EC2
+      regions:
+        - us-east-2
+      roles:
+        - roleArn: "arn:aws:iam::683740446523:role/osm-cloudwatch-export-role" # osm-render
+      period: 300
+      length: 300
+      metrics:
+        - name: CPUUtilization
+          statistics: [Average]
+        - name: DiskReadOps
+          statistics: [Sum]
+        - name: DiskWriteOps
+          statistics: [Sum]
+        - name: DiskReadBytes
+          statistics: [Sum]
+        - name: DiskWriteBytes
+          statistics: [Sum]
+        - name: NetworkIn
+          statistics: [Sum]
+        - name: NetworkOut
+          statistics: [Sum]
+        - name: NetworkPacketsIn
+          statistics: [Sum]
+        - name: NetworkPacketsOut
+          statistics: [Sum]
+    - type: AWS/Billing
+      regions:
+        - us-east-1
+      roles:
+        - roleArn: "arn:aws:iam::173189593406:role/osm-cloudwatch-export-role" # osm-main
+        - roleArn: "arn:aws:iam::674083635870:role/osm-cloudwatch-export-role" # osm-backup
+        - roleArn: "arn:aws:iam::683740446523:role/osm-cloudwatch-export-role" # osm-render
+        # Note osm-planet is billed direct to AWS Open Data Program
+      period: 21600
+      length: 21600
+      metrics:
+        - name: EstimatedCharges

           statistics: [Sum]
           statistics: [Sum]