X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/778dafe1b60675abedacec30bbbec1e174fec5b0..bba9cc3fa8c80291c9db5188f19cd2030595007a:/cookbooks/prometheus/recipes/server.rb

diff --git a/cookbooks/prometheus/recipes/server.rb b/cookbooks/prometheus/recipes/server.rb
index d069cd8e7..1d3a8aa5a 100644
--- a/cookbooks/prometheus/recipes/server.rb
+++ b/cookbooks/prometheus/recipes/server.rb
@@ -37,13 +37,13 @@ package %w[
   prometheus-alertmanager
 ]
 
-promscale_version = "0.1.4"
+promscale_version = "0.2.1"
 
-database_cluster = node[:timescaledb][:cluster]
+database_cluster = "#{node[:timescaledb][:database_version]}/main"
 
 postgresql_user "prometheus" do
   cluster database_cluster
-  createrole true
+  superuser true
 end
 
 postgresql_database "promscale" do
@@ -51,11 +51,6 @@ postgresql_database "promscale" do
   owner "prometheus"
 end
 
-postgresql_extension "timescaledb" do
-  cluster database_cluster
-  database "promscale"
-end
-
 directory "/opt/promscale" do
   owner "root"
   group "root"
@@ -121,12 +116,6 @@ service "promscale" do
   subscribes :restart, "systemd_service[promscale]"
 end
 
-postgresql_extension "promscale" do
-  cluster database_cluster
-  database "promscale"
-  notifies :restart, "service[promscale]"
-end
-
 systemd_service "promscale-maintenance" do
   description "Promscale Maintenace"
   type "simple"
@@ -144,7 +133,24 @@ systemd_timer "promscale-maintenance" do
   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"
@@ -159,16 +165,58 @@ 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"
@@ -181,13 +229,21 @@ template "/etc/prometheus/prometheus.yml" do
   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
@@ -210,6 +266,13 @@ service "prometheus-alertmanager" do
   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"
 
 template "/etc/grafana/grafana.ini" do
@@ -236,3 +299,10 @@ 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