Class: Terrafying::Components::Prometheus

Inherits:
Terrafying::Context
  • Object
show all
Defined in:
lib/terrafying/components/prometheus.rb

Instance Attribute Summary collapse

Class Method Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(vpc:, thanos_name: 'thanos', thanos_version: 'v0.4.0', prom_name: 'prometheus', prom_version: 'v2.10.0', instances: 2, instance_type: 't3a.small', thanos_instance_type: 't3a.small', prometheus_tsdb_retention: '1d') ⇒ Prometheus

Returns a new instance of Prometheus.



20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
 
# File 'lib/terrafying/components/prometheus.rb', line 20

def initialize(
  vpc:,
  thanos_name: 'thanos',
  thanos_version: 'v0.4.0',
  prom_name: 'prometheus',
  prom_version: 'v2.10.0',
  instances: 2,
  instance_type: 't3a.small',
  thanos_instance_type: 't3a.small',
  prometheus_tsdb_retention: '1d'
)
  super()
  @vpc = vpc
  @thanos_name = thanos_name
  @thanos_version = thanos_version
  @prom_name = prom_name
  @prom_version = prom_version
  @instances = instances
  @prometheus_instance_type = instance_type
  @thanos_instance_type = thanos_instance_type
  @prometheus_tsdb_retention = prometheus_tsdb_retention
end

Instance Attribute Details

#prometheus ⇒ Object (readonly)

Returns the value of attribute prometheus.



10
11
12
 
# File 'lib/terrafying/components/prometheus.rb', line 10

def prometheus
  @prometheus
end

#security_group ⇒ Object (readonly)

Returns the value of attribute security_group.



10
11
12
 
# File 'lib/terrafying/components/prometheus.rb', line 10

def security_group
  @security_group
end

Class Method Details

.create_in(options) ⇒ Object 



12
13
14
 
# File 'lib/terrafying/components/prometheus.rb', line 12

def self.create_in(options)
  new(**options).tap(&:create)
end

.find_in(options) ⇒ Object 



16
17
18
 
# File 'lib/terrafying/components/prometheus.rb', line 16

def self.find_in(options)
  new(**options).tap(&:find)
end

Instance Method Details

#cloudwatch_alarm(name, namespace, dimensions) ⇒ Object 



334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
 
# File 'lib/terrafying/components/prometheus.rb', line 334

def cloudwatch_alarm(name, namespace, dimensions)
  resource 'aws_cloudwatch_metric_alarm', name,
           alarm_name: name,
           comparison_operator: 'GreaterThanOrEqualToThreshold',
           evaluation_periods: '1',
           metric_name: 'UnHealthyHostCount',
           namespace: namespace,
           period: '180',
           threshold: '1',
           statistic: 'Minimum',
           alarm_description: "Monitoring #{name} target group host health",
           dimensions: dimensions,
           alarm_actions: ['arn:aws:sns:eu-west-1:136393635417:prometheus_cloudwatch_topic'],
           ok_actions: ['arn:aws:sns:eu-west-1:136393635417:prometheus_cloudwatch_topic']
end

#create ⇒ Object 



50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
 
# File 'lib/terrafying/components/prometheus.rb', line 50

def create
  prometheus_thanos_sidecar_hostname = tf_safe(@prom_name)
  prometheus_thanos_sidecar_srv_fqdn = "_grpc._tcp.#{@vpc.zone.qualify prometheus_thanos_sidecar_hostname}"
  @prometheus_instance_vcpu_count = aws.instance_type_vcpu_count(@prometheus_instance_type)
  @thanos_instance_vcpu_count = aws.instance_type_vcpu_count(@thanos_instance_type)
  @thanos = create_thanos(prometheus_thanos_sidecar_srv_fqdn)
  create_thanos_cloudwatch_alert(@thanos)

  @prometheus = create_prom

  @security_group = @prometheus.egress_security_group

  # Form SRV record with thanos-sidecars
  @vpc.zone.add_srv_in(self, prometheus_thanos_sidecar_hostname, 'grpc', 10_901, 'tcp', @prom_service.domain_names.drop(1))

  # Allow Prometheus to scrape Thanos Query
  @thanos.used_by(@prometheus) { |port| port[:upstream_port] == 10_902 }
  # Allow Thanos Query instance to reach Prometheus running Thanos Sidecar
  @prometheus.used_by(@thanos) { |port| port[:upstream_port] == 10_901 }
  # Allow connections from VPC to Thanos Query services
  @thanos.used_by_cidr(@vpc.cidr) { |port| [10_902, 10_901].include? port[:upstream_port] }
end

#create_prom ⇒ Object 



73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
 
# File 'lib/terrafying/components/prometheus.rb', line 73

def create_prom
  @prom_service = add! Terrafying::Components::Service.create_in(
    @vpc, @prom_name,
    ports: [
      {
        type: 'tcp',
        number: 9090
      },
      {
        type: 'tcp',
        number: 10_902
      },
      {
        type: 'tcp',
        number: 10_901
      }
    ],
    instance_type: @prometheus_instance_type,
    iam_policy_statements: thanos_store_access,
    instances: [{}] * @instances,
    units: [prometheus_unit, thanos_sidecar_unit],
    files: [prometheus_conf, thanos_bucket],
    tags: {
      prometheus_port: 9090,
      prometheus_path: '/metrics',
      prometheus_port_0: 10_902,
      prometheus_path_0: '/metrics'
    }
  )
end

#create_thanos(prometheus_thanos_sidecar_srv_fqdn) ⇒ Object 



104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
 
# File 'lib/terrafying/components/prometheus.rb', line 104

def create_thanos(prometheus_thanos_sidecar_srv_fqdn)
  @thanos_service = add! Terrafying::Components::Service.create_in(
    @vpc, @thanos_name,
    ports: [
      {
        type: 'tcp',
        number: 10_902,
        health_check: {
          protocol: 'HTTP',
          path: '/-/healthy'
        }
      },
      {
        type: 'tcp',
        number: 10_901,
        health_check: {
          protocol: 'TCP'
        }
      }
    ],
    instance_type: @thanos_instance_type,
    units: [thanos_unit(prometheus_thanos_sidecar_srv_fqdn)],
    instances: [{}] * @instances,
    loadbalancer: true,
    tags: {
      prometheus_port: 10_902,
      prometheus_path: '/metrics'
    }
  )
end

#create_thanos_cloudwatch_alert(service) ⇒ Object 



350
351
352
353
354
355
356
 
# File 'lib/terrafying/components/prometheus.rb', line 350

def create_thanos_cloudwatch_alert(service)
  service.load_balancer.targets.each_with_index do |target, i|
    cloudwatch_alarm "#{service.name}_#{i}", 'AWS/NetworkELB',
                     LoadBalancer: output_of('aws_lb', service.load_balancer.name, 'arn_suffix'),
                     TargetGroup: target.target_group.to_s.gsub(/id/, 'arn_suffix')
  end
end

#expose_in(vpc) ⇒ Object 



324
325
326
327
328
329
330
331
332
 
# File 'lib/terrafying/components/prometheus.rb', line 324

def expose_in(vpc)
  @endpoint_service ||= @thanos.with_endpoint_service(acceptance_required: false)

  options = {}
  endpoint = add! @endpoint_service.expose_in(vpc, options)
  endpoint.used_by_cidr(vpc.cidr)

  endpoint
end

#find ⇒ Object 



43
44
45
46
47
48
 
# File 'lib/terrafying/components/prometheus.rb', line 43

def find
  @security_group = aws.security_group_in_vpc(
    @vpc.id,
    "staticset-#{@vpc.name}-#{@prom_name}"
  )
end

#prometheus_conf ⇒ Object 



206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
 
# File 'lib/terrafying/components/prometheus.rb', line 206

def prometheus_conf
  {
    path: '/opt/prometheus/prometheus.yml',
    mode: 0o644,
    contents: ERB.new(<<~'END', 0, '-', '_').result(binding)
      global:
        external_labels:
          monitor: prometheus
          cluster: <%= @vpc.name %>
          replica: {{HOST}}
        scrape_interval: 15s
      scrape_configs:
      # While AWS EC2 instance support up to 50 tags, we wouldn't be able
      # to fit such a long configuration file into user_data of the
      # instance; user_data is limited to just 16k.
      # This configuration support scraping up to 5 ports per instance:
      <%- prom_tag_name_suffixes = [''] + (0..3).map {|i| "_#{i}"} -%>
      <%- prom_tag_name_suffixes.each do |suffix| -%>
      - job_name: ec2<%= suffix %>
        params:
          format: ["prometheus"]
        ec2_sd_configs:
        - region: eu-west-1
          filters:
          - name: vpc-id
            values: ["<%= @vpc.id %>"]
          - # by using the same ec2_sd_configs we are able to share single
            # provider instance thanks to SD configuration coalescing
            # therefore "prometheus_port" tag must always be present on
            # the instance to be discovered (i. e. "prometheus_port_*" tag
            # would not be sufficient if "  " tag is missing)
            name: tag-key
            values: ["prometheus_port"]
        relabel_configs:
        - source_labels: [__meta_ec2_tag_prometheus_port<%= suffix %>]
          regex: (.+)
          action: keep
        - source_labels: [__meta_ec2_private_ip, __meta_ec2_tag_prometheus_port<%= suffix %>]
          target_label: __address__
          separator: ':'
        - source_labels: [__meta_ec2_tag_prometheus_path<%= suffix %>]
          target_label: __metrics_path__
          regex: (.+)
        - source_labels: [__meta_ec2_instance_id]
          target_label: instance_id
        - source_labels: [__meta_ec2_tag_envoy_cluster]
          target_label: envoy_cluster
      <%- end -%>
    END
  }
end

#prometheus_unit ⇒ Object 



135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
 
# File 'lib/terrafying/components/prometheus.rb', line 135

def prometheus_unit
  {
    name: 'prometheus.service',
    contents: <<~PROM_UNIT
      [Install]
      WantedBy=multi-user.target
       [Unit]
      Description=Prometheus Service
      After=docker.service
      Requires=docker.service
       [Service]
      ExecStartPre=-/usr/bin/docker network create --driver bridge prom
      ExecStartPre=-/usr/bin/docker kill prometheus
      ExecStartPre=-/usr/bin/docker rm prometheus
      ExecStartPre=/usr/bin/docker pull quay.io/prometheus/prometheus:#{@prom_version}
      ExecStartPre=-/usr/bin/sed -i "s/{{HOST}}/%H/" /opt/prometheus/prometheus.yml
      ExecStartPre=/usr/bin/install -d -o nobody -g nobody -m 0755 /opt/prometheus/data
      ExecStart=/usr/bin/docker run --name prometheus \
        -p 9090:9090 \
        --network=prom \
        -v /opt/prometheus:/opt/prometheus \
        quay.io/prometheus/prometheus:#{@prom_version} \
        --storage.tsdb.path=/opt/prometheus/data \
        --storage.tsdb.retention.time=#{@prometheus_tsdb_retention} \
        --storage.tsdb.min-block-duration=2h \
        --storage.tsdb.max-block-duration=2h \
        --storage.tsdb.no-lockfile \
        --storage.remote.read-concurrent-limit=#{@prometheus_instance_vcpu_count} \
        --query.max-concurrency=#{@prometheus_instance_vcpu_count} \
        --config.file=/opt/prometheus/prometheus.yml \
        --web.console.templates=/etc/prometheus/consoles \
        --web.console.libraries=/etc/prometheus/console_libraries \
        --web.enable-lifecycle \
        --log.level=warn
      Restart=always
      RestartSec=30
    PROM_UNIT
  }
end

#thanos_bucket ⇒ Object 



288
289
290
291
292
293
294
295
296
297
298
299
 
# File 'lib/terrafying/components/prometheus.rb', line 288

def thanos_bucket
  {
    path: '/opt/thanos/bucket.yml',
    mode: 0o644,
    contents: <<~S3CONF
      type: S3
      config:
          bucket: uswitch-thanos-store
          endpoint: s3.eu-west-1.amazonaws.com
    S3CONF
  }
end

#thanos_sidecar_unit ⇒ Object 



175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
 
# File 'lib/terrafying/components/prometheus.rb', line 175

def thanos_sidecar_unit
  {
    name: 'thanos.service',
    contents: <<~THANOS_SIDE
      [Install]
      WantedBy=multi-user.target
       [Unit]
      Description=Thanos Service
      After=docker.service prometheus.service
      Requires=docker.service prometheus.service
       [Service]
      ExecStartPre=-/usr/bin/docker kill thanos
      ExecStartPre=-/usr/bin/docker rm thanos
      ExecStartPre=/usr/bin/docker pull improbable/thanos:#{@thanos_version}
      ExecStart=/usr/bin/docker run --name thanos \
        -p 10901-10902:10901-10902 \
        -v /opt/prometheus:/opt/prometheus \
        -v /opt/thanos:/opt/thanos \
        --network=prom \
        improbable/thanos:#{@thanos_version} \
        sidecar \
        --prometheus.url=http://prometheus:9090 \
        --tsdb.path=/opt/prometheus/data \
        --objstore.config-file=/opt/thanos/bucket.yml \
        --log.level=warn
      Restart=always
      RestartSec=30
    THANOS_SIDE
  }
end

#thanos_store_access ⇒ Object 



301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
 
# File 'lib/terrafying/components/prometheus.rb', line 301

def thanos_store_access
  [
    {
      Action: ['ec2:DescribeInstances'],
      Effect: 'Allow',
      Resource: '*'
    },
    {
      Action: [
        's3:ListBucket',
        's3:GetObject',
        's3:DeleteObject',
        's3:PutObject'
      ],
      Effect: 'Allow',
      Resource: [
        'arn:aws:s3:::uswitch-thanos-store/*',
        'arn:aws:s3:::uswitch-thanos-store'
      ]
    }
  ]
end

#thanos_unit(prometheus_thanos_sidecar_srv_fqdn) ⇒ Object 



258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
 
# File 'lib/terrafying/components/prometheus.rb', line 258

def thanos_unit(prometheus_thanos_sidecar_srv_fqdn)
  {
    name: 'thanos.service',
    contents: <<~THANOS_UNIT
      [Install]
      WantedBy=multi-user.target

      [Unit]
      Description=Thanos Service
      After=docker.service
      Requires=docker.service

      [Service]
      ExecStartPre=-/usr/bin/docker kill thanos
      ExecStartPre=-/usr/bin/docker rm thanos
      ExecStartPre=/usr/bin/docker pull improbable/thanos:#{@thanos_version}
      ExecStart=/usr/bin/docker run --name thanos \
        -p 10901-10902:10901-10902 \
        improbable/thanos:#{@thanos_version} \
        query \
        --query.replica-label=replica \
        --query.max-concurrent=#{@thanos_instance_vcpu_count} \
        --store=dnssrv+#{prometheus_thanos_sidecar_srv_fqdn} \
        --log.level=warn
      Restart=always
      RestartSec=30
    THANOS_UNIT
  }
end