Statically configure backup 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 43eb9e3ae682fa4209f84158113a822993bd6547..6a8757e40d66fd363bf9ab87814920771377eb41 100644 (file)
--- a/cookbooks/prometheus/templates/default/cloudwatch.yml.erb
+++ b/cookbooks/prometheus/templates/default/cloudwatch.yml.erb
@@ -1,4 +1,5 @@
 apiVersion: v1alpha1
+sts-region: eu-west-1
 discovery:
   jobs:
     - type: AWS/S3
@@ -6,18 +7,139 @@ discovery:
         - eu-west-1
         - eu-west-2
       roles:
-        - roleArn: "arn:aws:iam::173189593406:role/osm-cloudwatch-export-role"
+        - roleArn: "arn:aws:iam::173189593406:role/osm-cloudwatch-export-role" # osm-main
+      roundingPeriod: 86400
+      period: 86400
+      length: 86400
+      metrics:
+        - name: BucketSizeBytes
+          statistics: [Average]
+        - name: NumberOfObjects
+          statistics: [Average]
+    - type: AWS/S3
+      regions:
+        - eu-west-1
+        - eu-west-2
+      roles:
+        - roleArn: "arn:aws:iam::173189593406:role/osm-cloudwatch-export-role" # osm-main
+      roundingPeriod: 300
+      period: 300
+      length: 300
+      metrics:
+        - name: AllRequests
+          statistics: [Sum]
+        - name: GetRequests
+          statistics: [Sum]
+        - name: PutRequests
+          statistics: [Sum]
+        - name: DeleteRequests
+          statistics: [Sum]
+        - name: HeadRequests
+          statistics: [Sum]
+        - name: PostRequests
+          statistics: [Sum]
+        - name: ListRequests
+          statistics: [Sum]
+        - name: BytesDownloaded
+          statistics: [Sum]
+        - name: BytesUploaded
+          statistics: [Sum]
+        - name: 4xxErrors
+          statistics: [Sum]
+        - name: 5xxErrors
+          statistics: [Sum]
+        - name: FirstByteLatency
+          statistics: [Minimum, Maximum, Average]
+        - name: TotalRequestLatency
+          statistics: [Minimum, Maximum, Average]
+        - name: ReplicationLatency
+          statistics: [Maximum]
+        - name: BytesPendingReplication
+          statistics: [Maximum]
+        - name: OperationsPendingReplication
+          statistics: [Maximum]
+        - name: OperationsFailedReplication
+          statistics: [Sum]
+    - type: AWS/S3
+      regions:
+        - eu-central-1
+        - us-west-2
+      roles:
+        - roleArn: "arn:aws:iam::630658470130:role/osm-cloudwatch-export-role" # osm-planet (pds)
+      roundingPeriod: 86400
+      period: 86400
+      length: 86400
+      metrics:
+        - name: BucketSizeBytes
+          statistics: [Average]
+        - name: NumberOfObjects
+          statistics: [Average]
+    - type: AWS/S3
+      regions:
+        - eu-central-1
+        - us-west-2
+      roles:
+        - roleArn: "arn:aws:iam::630658470130:role/osm-cloudwatch-export-role" # osm-planet (pds)
+      roundingPeriod: 300
       period: 300
       length: 300
+      metrics:
+        - name: AllRequests
+          statistics: [Sum]
+        - name: GetRequests
+          statistics: [Sum]
+        - name: PutRequests
+          statistics: [Sum]
+        - name: DeleteRequests
+          statistics: [Sum]
+        - name: HeadRequests
+          statistics: [Sum]
+        - name: PostRequests
+          statistics: [Sum]
+        - name: ListRequests
+          statistics: [Sum]
+        - name: BytesDownloaded
+          statistics: [Sum]
+        - name: BytesUploaded
+          statistics: [Sum]
+        - name: 4xxErrors
+          statistics: [Sum]
+        - name: 5xxErrors
+          statistics: [Sum]
+        - name: FirstByteLatency
+          statistics: [Minimum, Maximum, Average]
+        - name: TotalRequestLatency
+          statistics: [Minimum, Maximum, Average]
+        - name: ReplicationLatency
+          statistics: [Maximum]
+        - name: BytesPendingReplication
+          statistics: [Maximum]
+        - name: OperationsPendingReplication
+          statistics: [Maximum]
+        - name: OperationsFailedReplication
+          statistics: [Sum]
+    - type: AWS/S3
+      regions:
+        - eu-north-1
+      roles:
+        - roleArn: "arn:aws:iam::674083635870:role/osm-cloudwatch-export-role" # osm-backup
+      roundingPeriod: 86400
+      period: 86400
+      length: 86400
       metrics:
         - name: BucketSizeBytes
           statistics: [Average]
-          period: 86400
-          length: 86400
         - name: NumberOfObjects
           statistics: [Average]
-          period: 86400
-          length: 86400
+    - type: AWS/S3
+      regions:
+        - eu-north-1
+      roles:
+        - roleArn: "arn:aws:iam::674083635870:role/osm-cloudwatch-export-role" # osm-backup
+      roundingPeriod: 300
+      period: 300
+      length: 300
+      metrics:
         - name: AllRequests
           statistics: [Sum]
         - name: GetRequests
@@ -52,3 +174,149 @@ discovery:
           statistics: [Maximum]
         - name: OperationsFailedReplication
           statistics: [Sum]
+    - type: AWS/EC2
+      regions:
+        - us-east-2
+      roles:
+        - roleArn: "arn:aws:iam::683740446523:role/osm-cloudwatch-export-role" # osm-render
+      roundingPeriod: 300
+      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
+      roundingPeriod: 21600
+      period: 21600
+      length: 21600
+      metrics:
+        - name: EstimatedCharges
+          statistics: [Sum]
+static:
+  - namespace: AWS/S3
+    name: gps-images-replication
+    regions:
+      - eu-north-1
+    roles:
+      - roleArn: "arn:aws:iam::173189593406:role/osm-cloudwatch-export-role" # osm-main
+    dimensions:
+      - name: RuleId
+        value: full-bucket-replication
+      - name: SourceBucket
+        value: openstreetmap-gps-images
+      - name: DestinationBucket
+        value: openstreetmap-gps-images-replicate
+    metrics:
+      - name: ReplicationLatency
+        statistics: [Maximum]
+        period: 300
+        length: 300
+      - name: BytesPendingReplication
+        statistics: [Maximum]
+        period: 300
+        length: 300
+      - name: OperationsPendingReplication
+        statistics: [Maximum]
+        period: 300
+        length: 300
+  - namespace: AWS/S3
+    name: gps-traces-replication
+    regions:
+      - eu-north-1
+    roles:
+      - roleArn: "arn:aws:iam::173189593406:role/osm-cloudwatch-export-role" # osm-main
+    dimensions:
+      - name: RuleId
+        value: full-bucket-replication
+      - name: SourceBucket
+        value: openstreetmap-gps-traces
+      - name: DestinationBucket
+        value: openstreetmap-gps-traces-replicate
+    metrics:
+      - name: ReplicationLatency
+        statistics: [Maximum]
+        period: 300
+        length: 300
+      - name: BytesPendingReplication
+        statistics: [Maximum]
+        period: 300
+        length: 300
+      - name: OperationsPendingReplication
+        statistics: [Maximum]
+        period: 300
+        length: 300
+  - namespace: AWS/S3
+    name: user-avatars-replication
+    regions:
+      - eu-north-1
+    roles:
+      - roleArn: "arn:aws:iam::173189593406:role/osm-cloudwatch-export-role" # osm-main
+    dimensions:
+      - name: RuleId
+        value: full-bucket-replication
+      - name: SourceBucket
+        value: openstreetmap-user-avatars
+      - name: DestinationBucket
+        value: openstreetmap-user-avatars-replicate
+    metrics:
+      - name: ReplicationLatency
+        statistics: [Maximum]
+        period: 300
+        length: 300
+      - name: BytesPendingReplication
+        statistics: [Maximum]
+        period: 300
+        length: 300
+      - name: OperationsPendingReplication
+        statistics: [Maximum]
+        period: 300
+        length: 300
+  - namespace: AWS/S3
+    name: aws-cloudtrail-replication
+    regions:
+      - eu-north-1
+    roles:
+      - roleArn: "arn:aws:iam::173189593406:role/osm-cloudwatch-export-role" # osm-main
+    dimensions:
+      - name: RuleId
+        value: full-bucket-replication
+      - name: SourceBucket
+        value: openstreetmap-aws-cloudtrail
+      - name: DestinationBucket
+        value: openstreetmap-aws-cloudtrail-replicate
+    metrics:
+      - name: ReplicationLatency
+        statistics: [Maximum]
+        period: 300
+        length: 300
+      - name: BytesPendingReplication
+        statistics: [Maximum]
+        period: 300
+        length: 300
+      - name: OperationsPendingReplication
+        statistics: [Maximum]
+        period: 300
+        length: 300