]> git.openstreetmap.org Git - chef.git/blob - cookbooks/prometheus/templates/default/cloudwatch.yml.erb
Only try and create standby.signal if the cluster exists
[chef.git] / cookbooks / prometheus / templates / default / cloudwatch.yml.erb
1 apiVersion: v1alpha1
2 sts-region: eu-west-1
3 discovery:
4   jobs:
5     - type: AWS/S3
6       regions:
7         - eu-west-1
8         - eu-west-2
9       roles:
10         - roleArn: "arn:aws:iam::173189593406:role/osm-cloudwatch-export-role" # osm-main
11       period: 86400
12       length: 86400
13       metrics:
14         - name: BucketSizeBytes
15           statistics: [Average]
16         - name: NumberOfObjects
17           statistics: [Average]
18     - type: AWS/S3
19       regions:
20         - eu-west-1
21         - eu-west-2
22       roles:
23         - roleArn: "arn:aws:iam::173189593406:role/osm-cloudwatch-export-role" # osm-main
24       period: 300
25       length: 300
26       metrics:
27         - name: AllRequests
28           statistics: [Sum]
29         - name: GetRequests
30           statistics: [Sum]
31         - name: PutRequests
32           statistics: [Sum]
33         - name: DeleteRequests
34           statistics: [Sum]
35         - name: HeadRequests
36           statistics: [Sum]
37         - name: PostRequests
38           statistics: [Sum]
39         - name: ListRequests
40           statistics: [Sum]
41         - name: BytesDownloaded
42           statistics: [Sum]
43         - name: BytesUploaded
44           statistics: [Sum]
45         - name: 4xxErrors
46           statistics: [Sum]
47         - name: 5xxErrors
48           statistics: [Sum]
49         - name: FirstByteLatency
50           statistics: [Minimum, Maximum, Average]
51         - name: TotalRequestLatency
52           statistics: [Minimum, Maximum, Average]
53         - name: ReplicationLatency
54           statistics: [Maximum]
55         - name: BytesPendingReplication
56           statistics: [Maximum]
57         - name: OperationsPendingReplication
58           statistics: [Maximum]
59         - name: OperationsFailedReplication
60           statistics: [Sum]
61     - type: AWS/S3
62       regions:
63         - eu-central-1
64         - us-west-2
65       roles:
66         - roleArn: "arn:aws:iam::630658470130:role/osm-cloudwatch-export-role" # osm-planet (pds)
67       period: 86400
68       length: 86400
69       metrics:
70         - name: BucketSizeBytes
71           statistics: [Average]
72         - name: NumberOfObjects
73           statistics: [Average]
74     - type: AWS/S3
75       regions:
76         - eu-central-1
77         - us-west-2
78       roles:
79         - roleArn: "arn:aws:iam::630658470130:role/osm-cloudwatch-export-role" # osm-planet (pds)
80       period: 300
81       length: 300
82       metrics:
83         - name: AllRequests
84           statistics: [Sum]
85         - name: GetRequests
86           statistics: [Sum]
87         - name: PutRequests
88           statistics: [Sum]
89         - name: DeleteRequests
90           statistics: [Sum]
91         - name: HeadRequests
92           statistics: [Sum]
93         - name: PostRequests
94           statistics: [Sum]
95         - name: ListRequests
96           statistics: [Sum]
97         - name: BytesDownloaded
98           statistics: [Sum]
99         - name: BytesUploaded
100           statistics: [Sum]
101         - name: 4xxErrors
102           statistics: [Sum]
103         - name: 5xxErrors
104           statistics: [Sum]
105         - name: FirstByteLatency
106           statistics: [Minimum, Maximum, Average]
107         - name: TotalRequestLatency
108           statistics: [Minimum, Maximum, Average]
109         - name: ReplicationLatency
110           statistics: [Maximum]
111         - name: BytesPendingReplication
112           statistics: [Maximum]
113         - name: OperationsPendingReplication
114           statistics: [Maximum]
115         - name: OperationsFailedReplication
116           statistics: [Sum]
117     - type: AWS/S3
118       regions:
119         - eu-north-1
120       roles:
121         - roleArn: "arn:aws:iam::674083635870:role/osm-cloudwatch-export-role" # osm-backup
122         - roleArn: "arn:aws:iam::949022809175:role/osm-cloudwatch-export-role" # osm-vector
123         - roleArn: "arn:aws:iam::289997608296:role/osm-cloudwatch-export-role" # osm-service-logs
124       period: 86400
125       length: 86400
126       metrics:
127         - name: BucketSizeBytes
128           statistics: [Average]
129         - name: NumberOfObjects
130           statistics: [Average]
131     - type: AWS/S3
132       regions:
133         - eu-north-1
134       roles:
135         - roleArn: "arn:aws:iam::674083635870:role/osm-cloudwatch-export-role" # osm-backup
136         - roleArn: "arn:aws:iam::949022809175:role/osm-cloudwatch-export-role" # osm-vector
137         - roleArn: "arn:aws:iam::289997608296:role/osm-cloudwatch-export-role" # osm-service-logs
138       period: 300
139       length: 300
140       metrics:
141         - name: AllRequests
142           statistics: [Sum]
143         - name: GetRequests
144           statistics: [Sum]
145         - name: PutRequests
146           statistics: [Sum]
147         - name: DeleteRequests
148           statistics: [Sum]
149         - name: HeadRequests
150           statistics: [Sum]
151         - name: PostRequests
152           statistics: [Sum]
153         - name: ListRequests
154           statistics: [Sum]
155         - name: BytesDownloaded
156           statistics: [Sum]
157         - name: BytesUploaded
158           statistics: [Sum]
159         - name: 4xxErrors
160           statistics: [Sum]
161         - name: 5xxErrors
162           statistics: [Sum]
163         - name: FirstByteLatency
164           statistics: [Minimum, Maximum, Average]
165         - name: TotalRequestLatency
166           statistics: [Minimum, Maximum, Average]
167         - name: ReplicationLatency
168           statistics: [Maximum]
169         - name: BytesPendingReplication
170           statistics: [Maximum]
171         - name: OperationsPendingReplication
172           statistics: [Maximum]
173         - name: OperationsFailedReplication
174           statistics: [Sum]
175     - type: AWS/EC2
176       regions:
177         - us-east-2
178       roles:
179         - roleArn: "arn:aws:iam::683740446523:role/osm-cloudwatch-export-role" # osm-render
180       period: 300
181       length: 300
182       metrics:
183         - name: CPUUtilization
184           statistics: [Average]
185         - name: DiskReadOps
186           statistics: [Sum]
187         - name: DiskWriteOps
188           statistics: [Sum]
189         - name: DiskReadBytes
190           statistics: [Sum]
191         - name: DiskWriteBytes
192           statistics: [Sum]
193         - name: NetworkIn
194           statistics: [Sum]
195         - name: NetworkOut
196           statistics: [Sum]
197         - name: NetworkPacketsIn
198           statistics: [Sum]
199         - name: NetworkPacketsOut
200           statistics: [Sum]
201     - type: AWS/Billing
202       regions:
203         - us-east-1
204       roles:
205         - roleArn: "arn:aws:iam::173189593406:role/osm-cloudwatch-export-role" # osm-main
206         - roleArn: "arn:aws:iam::674083635870:role/osm-cloudwatch-export-role" # osm-backup
207         - roleArn: "arn:aws:iam::683740446523:role/osm-cloudwatch-export-role" # osm-render
208         - roleArn: "arn:aws:iam::949022809175:role/osm-cloudwatch-export-role" # osm-vector
209         - roleArn: "arn:aws:iam::289997608296:role/osm-cloudwatch-export-role" # osm-service-logs
210         # Note osm-planet is billed direct to AWS Open Data Program
211       period: 21600
212       length: 21600
213       metrics:
214         - name: EstimatedCharges
215           statistics: [Sum]
216 static:
217   - namespace: AWS/S3
218     name: gps-images-replication
219     regions:
220       - eu-north-1
221     roles:
222       - roleArn: "arn:aws:iam::173189593406:role/osm-cloudwatch-export-role" # osm-main
223     dimensions:
224       - name: RuleId
225         value: full-bucket-replication
226       - name: SourceBucket
227         value: openstreetmap-gps-images
228       - name: DestinationBucket
229         value: openstreetmap-gps-images-replicate
230     metrics:
231       - name: ReplicationLatency
232         statistics: [Maximum]
233         period: 300
234         length: 300
235       - name: BytesPendingReplication
236         statistics: [Maximum]
237         period: 300
238         length: 300
239       - name: OperationsPendingReplication
240         statistics: [Maximum]
241         period: 300
242         length: 300
243   - namespace: AWS/S3
244     name: gps-traces-replication
245     regions:
246       - eu-north-1
247     roles:
248       - roleArn: "arn:aws:iam::173189593406:role/osm-cloudwatch-export-role" # osm-main
249     dimensions:
250       - name: RuleId
251         value: full-bucket-replication
252       - name: SourceBucket
253         value: openstreetmap-gps-traces
254       - name: DestinationBucket
255         value: openstreetmap-gps-traces-replicate
256     metrics:
257       - name: ReplicationLatency
258         statistics: [Maximum]
259         period: 300
260         length: 300
261       - name: BytesPendingReplication
262         statistics: [Maximum]
263         period: 300
264         length: 300
265       - name: OperationsPendingReplication
266         statistics: [Maximum]
267         period: 300
268         length: 300
269   - namespace: AWS/S3
270     name: user-avatars-replication
271     regions:
272       - eu-north-1
273     roles:
274       - roleArn: "arn:aws:iam::173189593406:role/osm-cloudwatch-export-role" # osm-main
275     dimensions:
276       - name: RuleId
277         value: full-bucket-replication
278       - name: SourceBucket
279         value: openstreetmap-user-avatars
280       - name: DestinationBucket
281         value: openstreetmap-user-avatars-replicate
282     metrics:
283       - name: ReplicationLatency
284         statistics: [Maximum]
285         period: 300
286         length: 300
287       - name: BytesPendingReplication
288         statistics: [Maximum]
289         period: 300
290         length: 300
291       - name: OperationsPendingReplication
292         statistics: [Maximum]
293         period: 300
294         length: 300
295   - namespace: AWS/S3
296     name: aws-cloudtrail-replication
297     regions:
298       - eu-north-1
299     roles:
300       - roleArn: "arn:aws:iam::173189593406:role/osm-cloudwatch-export-role" # osm-main
301     dimensions:
302       - name: RuleId
303         value: full-bucket-replication
304       - name: SourceBucket
305         value: openstreetmap-aws-cloudtrail
306       - name: DestinationBucket
307         value: openstreetmap-aws-cloudtrail-replicate
308     metrics:
309       - name: ReplicationLatency
310         statistics: [Maximum]
311         period: 300
312         length: 300
313       - name: BytesPendingReplication
314         statistics: [Maximum]
315         period: 300
316         length: 300
317       - name: OperationsPendingReplication
318         statistics: [Maximum]
319         period: 300
320         length: 300