X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/ef3e963c0c8532dcc2d2c132138e7e220db17b9f..c87a97890768b4427f078d7deda0d1edf4bb39ce:/cookbooks/prometheus/recipes/server.rb?ds=sidebyside

diff --git a/cookbooks/prometheus/recipes/server.rb b/cookbooks/prometheus/recipes/server.rb
index 6e112a000..b4a4c36da 100644
--- a/cookbooks/prometheus/recipes/server.rb
+++ b/cookbooks/prometheus/recipes/server.rb
@@ -32,11 +32,14 @@ prometheus_exporter "fastly" do
   environment "FASTLY_API_TOKEN" => tokens["fastly"]
 end
 
-package "prometheus"
+package %w[
+  prometheus
+  prometheus-alertmanager
+]
 
 promscale_version = "0.1.4"
 
-database_cluster = node[:timescaledb][:cluster]
+database_cluster = "#{node[:timescaledb][:database_version]}/main"
 
 postgresql_user "prometheus" do
   cluster database_cluster
@@ -81,8 +84,8 @@ execute "/opt/promscale/extension/Makefile" do
   cwd "/opt/promscale/extension"
   user "root"
   group "root"
-  subscribes :run, "git[/opt/promscale/extension]", :immediate
-  notifies :restart, "service[postgresql]", :immediate
+  subscribes :run, "git[/opt/promscale/extension]", :immediately
+  notifies :restart, "service[postgresql]", :immediately
 end
 
 directory "/opt/promscale/bin" do
@@ -104,7 +107,8 @@ systemd_service "promscale" do
   type "simple"
   user "prometheus"
   exec_start "/opt/promscale/bin/promscale --db-host /run/postgresql --db-port 5432 --db-user prometheus --db-name promscale --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" }
+  # 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
@@ -140,8 +144,24 @@ systemd_timer "promscale-maintenance" do
   on_unit_inactive_sec 1800
 end
 
+service "promscale-maintenance.timer" do
+  action [:enable, :start]
+end
+
 jobs = {}
 
+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
+
 search(:node, "recipes:prometheus\\:\\:default").sort_by(&:name).each do |client|
   if client[:prometheus][:mode] == "wireguard"
     node.default[:networking][:wireguard][:peers] << {
@@ -155,16 +175,29 @@ 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
 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"
@@ -173,9 +206,45 @@ template "/etc/prometheus/prometheus.yml" do
   variables :jobs => jobs
 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"
@@ -204,3 +273,10 @@ 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