X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/bfb0580b2ad65b9de88dfb78a014d29d498692df..7dea747820c21c7d45070bff703d58534aada3ba:/cookbooks/prometheus/recipes/server.rb

diff --git a/cookbooks/prometheus/recipes/server.rb b/cookbooks/prometheus/recipes/server.rb
index f3ff4d9ca..9414eec12 100644
--- a/cookbooks/prometheus/recipes/server.rb
+++ b/cookbooks/prometheus/recipes/server.rb
@@ -20,6 +20,7 @@
 include_recipe "apache"
 include_recipe "apt"
 include_recipe "networking"
+include_recipe "timescaledb"
 
 passwords = data_bag_item("prometheus", "passwords")
 tokens = data_bag_item("prometheus", "tokens")
@@ -31,9 +32,107 @@ prometheus_exporter "fastly" do
   environment "FASTLY_API_TOKEN" => tokens["fastly"]
 end
 
-package "prometheus"
+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
+
+service "promscale" do
+  action [:enable, :start]
+  subscribes :restart, "remote_file[/opt/promscale/bin/promscale]"
+  subscribes :restart, "systemd_service[promscale]"
+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
+
+service "promscale-maintenance.timer" do
+  action [:enable, :start]
+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"
@@ -48,27 +147,112 @@ search(:node, "recipes:prometheus\\:\\:default").sort_by(&:name).each do |client
     if exporter.is_a?(Hash)
       name = exporter[:name]
       address = exporter[:address]
+      metric_relabel = exporter[:metric_relabel] || []
     else
       name = key
       address = exporter
+      metric_relabel = []
     end
 
     jobs[name] ||= []
-    jobs[name] << { :address => address, :name => client.name }
+    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
+
+template "/etc/default/prometheus" do
+  source "default.prometheus.erb"
+  owner "root"
+  group "root"
+  mode "644"
+end
+
 template "/etc/prometheus/prometheus.yml" do
   source "prometheus.yml.erb"
   owner "root"
   group "root"
   mode "644"
-  variables :jobs => 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 :restart, "template[/etc/default/prometheus]"
   subscribes :reload, "template[/etc/prometheus/prometheus.yml]"
+  subscribes :reload, "template[/etc/prometheus/alert_rules.yml]"
+end
+
+template "/etc/default/prometheus-alertmanager" do
+  source "default.alertmanager.erb"
+  owner "root"
+  group "root"
+  mode "644"
+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 :restart, "template[/etc/default/prometheus-alertmanager]"
+  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"
@@ -97,3 +281,10 @@ 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