X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/22014524ab710947798fed63c81cafad02bcfb45..b71ac1f7665947274b96382fa8f2c9f0a3af67d2:/cookbooks/prometheus/recipes/server.rb

diff --git a/cookbooks/prometheus/recipes/server.rb b/cookbooks/prometheus/recipes/server.rb
index 2fcaddf47..9c215778e 100644
--- a/cookbooks/prometheus/recipes/server.rb
+++ b/cookbooks/prometheus/recipes/server.rb
@@ -19,31 +19,295 @@
 
 include_recipe "apache"
 include_recipe "apt"
+include_recipe "networking"
+include_recipe "timescaledb"
 
 passwords = data_bag_item("prometheus", "passwords")
+tokens = data_bag_item("prometheus", "tokens")
 
-package "prometheus"
+prometheus_exporter "fastly" do
+  port 8080
+  listen_switch "endpoint"
+  listen_type "url"
+  environment "FASTLY_API_TOKEN" => tokens["fastly"]
+end
+
+cache_dir = Chef::Config[:file_cache_path]
+
+prometheus_version = "2.31.1"
+alertmanager_version = "0.23.0"
+
+directory "/opt/prometheus-server" do
+  owner "root"
+  group "root"
+  mode "755"
+end
+
+remote_file "#{cache_dir}/prometheus.linux-amd64.tar.gz" do
+  source "https://github.com/prometheus/prometheus/releases/download/v#{prometheus_version}/prometheus-#{prometheus_version}.linux-amd64.tar.gz"
+  owner "root"
+  group "root"
+  mode "644"
+  backup false
+end
+
+archive_file "#{cache_dir}/prometheus.linux-amd64.tar.gz" do
+  action :nothing
+  destination "/opt/prometheus-server/prometheus"
+  overwrite true
+  strip_components 1
+  owner "root"
+  group "root"
+  subscribes :extract, "remote_file[#{cache_dir}/prometheus.linux-amd64.tar.gz]"
+end
+
+remote_file "#{cache_dir}/alertmanager.linux-amd64.tar.gz" do
+  source "https://github.com/prometheus/alertmanager/releases/download/v#{alertmanager_version}/alertmanager-#{alertmanager_version}.linux-amd64.tar.gz"
+  owner "root"
+  group "root"
+  mode "644"
+  backup false
+end
 
-clients = search(:node, "recipes:prometheus\\:\\:default").sort_by(&:name)
+archive_file "#{cache_dir}/alertmanager.linux-amd64.tar.gz" do
+  action :nothing
+  destination "/opt/prometheus-server/alertmanager"
+  overwrite true
+  strip_components 1
+  owner "root"
+  group "root"
+  subscribes :extract, "remote_file[#{cache_dir}/alertmanager.linux-amd64.tar.gz]"
+end
+
+package %w[
+  prometheus
+  prometheus-alertmanager
+]
+
+promscale_version = "0.6.2"
+
+database_version = node[:timescaledb][:database_version]
+database_cluster = "#{database_version}/main"
+
+postgresql_user "prometheus" do
+  cluster database_cluster
+  superuser true
+end
+
+postgresql_database "promscale" do
+  cluster database_cluster
+  owner "prometheus"
+end
+
+directory "/opt/promscale" do
+  owner "root"
+  group "root"
+  mode "755"
+end
+
+cookbook_file "/usr/lib/postgresql/#{database_version}/lib/promscale.so" do
+  source "postgresql-#{database_version}-promscale.so"
+  owner "root"
+  group "root"
+  mode "644"
+end
+
+directory "/opt/promscale/bin" do
+  owner "root"
+  group "root"
+  mode "755"
+end
+
+remote_file "/opt/promscale/bin/promscale" do
+  action :create
+  source "https://github.com/timescale/promscale/releases/download/#{promscale_version}/promscale_#{promscale_version}_Linux_x86_64"
+  owner "root"
+  group "root"
+  mode "755"
+end
+
+systemd_service "promscale" do
+  description "Promscale Connector"
+  type "simple"
+  user "prometheus"
+  exec_start "/opt/promscale/bin/promscale --db-uri postgresql:///promscale?host=/run/postgresql&port=5432 --db-connections-max 400"
+  # exec_start lazy { "/opt/promscale/bin/promscale --db-host /run/postgresql --db-port #{node[:postgresql][:clusters][database_cluster][:port]} --db-user prometheus --db-name promscale --db-max-connections 400" }
+  limit_nofile 16384
+  private_tmp true
+  protect_system "strict"
+  protect_home true
+  no_new_privileges true
+end
+
+systemd_service "promscale-maintenance" do
+  description "Promscale Maintenance"
+  type "simple"
+  user "prometheus"
+  exec_start "/usr/bin/psql --command='CALL prom_api.execute_maintenance()' promscale"
+  private_tmp true
+  protect_system "strict"
+  protect_home true
+  no_new_privileges true
+end
+
+systemd_timer "promscale-maintenance" do
+  description "Promscale Maintenace"
+  on_active_sec 1800
+  on_unit_inactive_sec 1800
+end
+
+if node[:prometheus][:promscale]
+  service "promscale" do
+    action [:enable, :start]
+    subscribes :restart, "remote_file[/opt/promscale/bin/promscale]"
+    subscribes :restart, "systemd_service[promscale]"
+  end
+
+  service "promscale-maintenance.timer" do
+    action [:enable, :start]
+  end
+else
+  service "promscale" do
+    action [:disable, :stop]
+  end
+
+  service "promscale-maintenance.timer" do
+    action [:disable, :stop]
+  end
+end
+
+search(:node, "roles:gateway") do |gateway|
+  allowed_ips = gateway.interfaces(:role => :internal).map do |interface|
+    "#{interface[:network]}/#{interface[:prefix]}"
+  end
+
+  node.default[:networking][:wireguard][:peers] << {
+    :public_key => gateway[:networking][:wireguard][:public_key],
+    :allowed_ips => allowed_ips,
+    :endpoint => "#{gateway.name}:51820"
+  }
+end
+
+jobs = {}
+snmp_targets = []
+
+search(:node, "recipes:prometheus\\:\\:default").sort_by(&:name).each do |client|
+  if client[:prometheus][:mode] == "wireguard"
+    node.default[:networking][:wireguard][:peers] << {
+      :public_key => client[:networking][:wireguard][:public_key],
+      :allowed_ips => client[:networking][:wireguard][:address],
+      :endpoint => "#{client.name}:51820"
+    }
+  end
+
+  client[:prometheus][:exporters].each do |key, exporter|
+    if exporter.is_a?(Hash)
+      name = exporter[:name]
+      address = exporter[:address]
+      metric_relabel = exporter[:metric_relabel] || []
+    else
+      name = key
+      address = exporter
+      metric_relabel = []
+    end
 
-prometheus_jobs = clients.each_with_object({}) do |client, jobs|
-  client[:prometheus][:exporters].each do |name, address|
     jobs[name] ||= []
-    jobs[name] << address
+    jobs[name] << {
+      :address => address,
+      :instance => client.name.split(".").first,
+      :metric_relabel => metric_relabel
+    }
+  end
+
+  Hash(client[:prometheus][:snmp]).each do |instance, details|
+    snmp_targets << {
+      :instance => instance,
+      :target => details[:address],
+      :module => details[:module],
+      :address => client[:prometheus][:addresses]["snmp"],
+      :labels => Array(details[:labels])
+    }
+  end
+end
+
+certificates = search(:node, "letsencrypt:certificates").each_with_object({}) do |n, c|
+  n[:letsencrypt][:certificates].each do |name, details|
+    c[name] ||= details.merge(:nodes => [])
+
+    c[name][:nodes] << {
+      :name => n[:fqdn],
+      :address => n.external_ipaddress || n.internal_ipaddress
+    }
   end
 end
 
+template "/etc/prometheus/ssl.yml" do
+  source "ssl.yml.erb"
+  owner "root"
+  group "root"
+  mode "644"
+  variables :certificates => certificates
+end
+
+prometheus_exporter "ssl" do
+  address "127.0.0.1"
+  port 9219
+  options "--config.file=/etc/prometheus/ssl.yml"
+  register_target false
+end
+
+systemd_service "prometheus-executable" do
+  service "prometheus"
+  dropin "executable"
+  exec_start "/opt/prometheus-server/prometheus/prometheus --config.file=/etc/prometheus/prometheus.yml --web.external-url=https://prometheus.openstreetmap.org/prometheus --storage.tsdb.path=/var/lib/prometheus/metrics2"
+  notifies :restart, "service[prometheus]"
+end
+
 template "/etc/prometheus/prometheus.yml" do
   source "prometheus.yml.erb"
   owner "root"
   group "root"
   mode "644"
-  variables :jobs => prometheus_jobs
+  variables :jobs => jobs, :snmp_targets => snmp_targets, :certificates => certificates
+end
+
+template "/etc/prometheus/alert_rules.yml" do
+  source "alert_rules.yml.erb"
+  owner "root"
+  group "root"
+  mode "644"
 end
 
 service "prometheus" do
   action [:enable, :start]
   subscribes :reload, "template[/etc/prometheus/prometheus.yml]"
+  subscribes :reload, "template[/etc/prometheus/alert_rules.yml]"
+end
+
+systemd_service "prometheus-alertmanager-executable" do
+  service "prometheus-alertmanager"
+  dropin "executable"
+  exec_start "/opt/prometheus-server/alertmanager/alertmanager --config.file=/etc/prometheus/alertmanager.yml --storage.path=/var/lib/prometheus/alertmanager --web.external-url=https://prometheus.openstreetmap.org/alertmanager"
+  notifies :restart, "service[prometheus-alertmanager]"
+end
+
+template "/etc/prometheus/alertmanager.yml" do
+  source "alertmanager.yml.erb"
+  owner "root"
+  group "root"
+  mode "644"
+end
+
+service "prometheus-alertmanager" do
+  action [:enable, :start]
+  subscribes :reload, "template[/etc/prometheus/alertmanager.yml]"
+end
+
+template "/etc/prometheus/amtool.yml" do
+  source "amtool.yml.erb"
+  owner "root"
+  group "root"
+  mode "644"
 end
 
 package "grafana-enterprise"
@@ -58,6 +322,7 @@ end
 
 service "grafana-server" do
   action [:enable, :start]
+  subscribes :restart, "template[/etc/grafana/grafana.ini]"
 end
 
 apache_module "alias"
@@ -71,3 +336,10 @@ end
 apache_site "prometheus.openstreetmap.org" do
   template "apache.erb"
 end
+
+template "/etc/cron.daily/prometheus-backup" do
+  source "backup.cron.erb"
+  owner "root"
+  group "root"
+  mode "750"
+end