tools/salt-install/config_examples/multi_host/aws/pillars/prometheus_server.sls

   1 ---
   2 # Copyright (C) The Arvados Authors. All rights reserved.
   3 #
   4 # SPDX-License-Identifier: AGPL-3.0
   5
   6 {%- set controller_nodes = "__CONTROLLER_NODES__".split(',') %}
   7 {%- set enable_balancer = ("__ENABLE_BALANCER__"|to_bool) %}
   8 {%- set data_retention_time = "__PROMETHEUS_DATA_RETENTION_TIME__" %}
   9
  10 ### PROMETHEUS
  11 prometheus:
  12   wanted:
  13     component:
  14       - prometheus
  15       - alertmanager
  16       - node_exporter
  17   pkg:
  18     use_upstream_repo: false
  19     use_upstream_archive: true
  20     component:
  21       prometheus:
  22         service:
  23            args:
  24              storage.tsdb.retention.time: {{ data_retention_time }}
  25         config:
  26           global:
  27             scrape_interval: 15s
  28             evaluation_interval: 15s
  29           rule_files:
  30             - rules.yml
  31
  32           scrape_configs:
  33             - job_name: prometheus
  34               # metrics_path defaults to /metrics
  35               # scheme defaults to http.
  36               static_configs:
  37               - targets: ['localhost:9090']
  38                 labels:
  39                   instance: mon.__CLUSTER__
  40                   cluster: __CLUSTER__
  41
  42             ## Arvados unique jobs
  43             - job_name: arvados_ws
  44               bearer_token: __MANAGEMENT_TOKEN__
  45               scheme: https
  46               static_configs:
  47                 - targets: ['ws.__DOMAIN__:443']
  48                   labels:
  49                     instance: ws.__CLUSTER__
  50                     cluster: __CLUSTER__
  51             - job_name: arvados_controller
  52               bearer_token: __MANAGEMENT_TOKEN__
  53               {%- if enable_balancer %}
  54               scheme: http
  55               {%- else %}
  56               scheme: https
  57               {%- endif %}
  58               static_configs:
  59                 {%- if enable_balancer %}
  60                   {%- for controller in controller_nodes %}
  61                 - targets: ['{{ controller }}']
  62                   labels:
  63                     instance: {{ controller.split('.')[0] }}.__CLUSTER__
  64                     cluster: __CLUSTER__
  65                   {%- endfor %}
  66                 {%- else %}
  67                 - targets: ['__DOMAIN__:443']
  68                   labels:
  69                     instance: controller.__CLUSTER__
  70                     cluster: __CLUSTER__
  71                 {%- endif %}
  72             - job_name: keep_web
  73               bearer_token: __MANAGEMENT_TOKEN__
  74               scheme: https
  75               static_configs:
  76                 - targets: ['keep.__DOMAIN__:443']
  77                   labels:
  78                     instance: keep-web.__CLUSTER__
  79                     cluster: __CLUSTER__
  80             - job_name: keep_balance
  81               bearer_token: __MANAGEMENT_TOKEN__
  82               static_configs:
  83                 - targets: ['__KEEPBALANCE_INT_IP__:9005']
  84                   labels:
  85                     instance: keep-balance.__CLUSTER__
  86                     cluster: __CLUSTER__
  87             - job_name: keepstore
  88               bearer_token: __MANAGEMENT_TOKEN__
  89               static_configs:
  90                 - targets: ['__KEEPSTORE0_INT_IP__:25107']
  91                   labels:
  92                     instance: keep0.__CLUSTER__
  93                     cluster: __CLUSTER__
  94             - job_name: arvados_dispatch_cloud
  95               bearer_token: __MANAGEMENT_TOKEN__
  96               static_configs:
  97                 - targets: ['__DISPATCHER_INT_IP__:9006']
  98                   labels:
  99                     instance: arvados-dispatch-cloud.__CLUSTER__
 100                     cluster: __CLUSTER__
 101
 102             {%- if "__DATABASE_INT_IP__" != "" %}
 103             # Database
 104             - job_name: postgresql
 105               static_configs:
 106                 - targets: [
 107                     '__DATABASE_INT_IP__:9187',
 108                     '__DATABASE_INT_IP__:3903'
 109                   ]
 110                   labels:
 111                     instance: database.__CLUSTER__
 112                     cluster: __CLUSTER__
 113             {%- endif %}
 114
 115             # Nodes
 116             {%- set node_list = "__NODELIST__".split(',') %}
 117             {%- set nodes = [] %}
 118             {%- for node in node_list %}
 119               {%- set _ = nodes.append(node.split('.')[0]) %}
 120             {%- endfor %}
 121             - job_name: node
 122               static_configs:
 123                 {% for node in nodes %}
 124                 - targets: [ "{{ node }}.__DOMAIN__:9100" ]
 125                   labels:
 126                     instance: "{{ node }}.__CLUSTER__"
 127                     cluster: __CLUSTER__
 128                 {% endfor %}