X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/eecaf26698348adf3d3b64ea680ad82f959a5ae4..9d1da497d442b16de5ab68a9cf74f832059734e7:/cookbooks/prometheus/recipes/server.rb?ds=inline

diff --git a/cookbooks/prometheus/recipes/server.rb b/cookbooks/prometheus/recipes/server.rb
index f3ff4d9ca..a480a8323 100644
--- a/cookbooks/prometheus/recipes/server.rb
+++ b/cookbooks/prometheus/recipes/server.rb
@@ -18,22 +18,110 @@
 #
 
 include_recipe "apache"
-include_recipe "apt"
+include_recipe "apt::grafana"
 include_recipe "networking"
 
 passwords = data_bag_item("prometheus", "passwords")
 tokens = data_bag_item("prometheus", "tokens")
+admins = data_bag_item("apache", "admins")
 
 prometheus_exporter "fastly" do
   port 8080
-  listen_switch "endpoint"
-  listen_type "url"
+  listen_switch "listen"
   environment "FASTLY_API_TOKEN" => tokens["fastly"]
 end
 
-package "prometheus"
+prometheus_exporter "fastly_healthcheck" do
+  port 9696
+  scrape_interval "1m"
+  environment "FASTLY_API_TOKEN" => tokens["fastly"]
+end
+
+prometheus_exporter "statuscake" do
+  port 9595
+  scrape_interval "5m"
+  scrape_timeout "2m"
+  environment "STATUSCAKE_APIKEY" => tokens["statuscake"]
+end
+
+cache_dir = Chef::Config[:file_cache_path]
+
+prometheus_version = "2.45.0"
+alertmanager_version = "0.25.0"
+karma_version = "0.114"
+
+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]", :immediately
+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
+
+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]", :immediately
+end
+
+remote_file "#{cache_dir}/karma-linux-amd64.tar.gz" do
+  source "https://github.com/prymitive/karma/releases/download/v#{karma_version}/karma-linux-amd64.tar.gz"
+  owner "root"
+  group "root"
+  mode "644"
+  backup false
+end
+
+archive_file "#{cache_dir}/karma-linux-amd64.tar.gz" do
+  action :nothing
+  destination "/opt/prometheus-server/karma"
+  overwrite true
+  owner "root"
+  group "root"
+  subscribes :extract, "remote_file[#{cache_dir}/karma-linux-amd64.tar.gz]", :immediately
+end
+
+search(:node, "roles:gateway") do |gateway|
+  allowed_ips = gateway.ipaddresses(:role => :internal).map(&:subnet)
+
+  node.default[:networking][:wireguard][:peers] << {
+    :public_key => gateway[:networking][:wireguard][:public_key],
+    :allowed_ips => allowed_ips,
+    :endpoint => "#{gateway.name}:51820"
+  }
+end
 
 jobs = {}
+junos_targets = []
+snmp_targets = []
 
 search(:node, "recipes:prometheus\\:\\:default").sort_by(&:name).each do |client|
   if client[:prometheus][:mode] == "wireguard"
@@ -48,27 +136,183 @@ search(:node, "recipes:prometheus\\:\\:default").sort_by(&:name).each do |client
     if exporter.is_a?(Hash)
       name = exporter[:name]
       address = exporter[:address]
+      sni = exporter[:sni]
+      labels = Array(exporter[:labels])
+      scrape_interval = exporter[:scrape_interval]
+      scrape_timeout = exporter[:scrape_timeout]
+      metric_relabel = exporter[:metric_relabel] || []
     else
       name = key
       address = exporter
+      sni = nil
+      labels = []
+      scrape_interval = nil
+      scrape_timeout = nil
+      metric_relabel = []
     end
 
     jobs[name] ||= []
-    jobs[name] << { :address => address, :name => client.name }
+    jobs[name] << {
+      :address => address,
+      :sni => sni,
+      :instance => client.name.split(".").first,
+      :labels => labels,
+      :scrape_interval => scrape_interval,
+      :scrape_timeout => scrape_timeout,
+      :metric_relabel => metric_relabel
+    }
+  end
+
+  Hash(client[:prometheus][:junos]).each do |instance, details|
+    junos_targets << {
+      :instance => instance,
+      :target => details[:address],
+      :address => client[:prometheus][:addresses]["junos"],
+      :labels => Array(details[:labels])
+    }
+  end
+
+  Hash(client[:prometheus][:snmp]).each do |instance, details|
+    snmp_targets << {
+      :instance => instance,
+      :target => details[:address],
+      :modules => details[:modules],
+      :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
+
+package "prometheus"
+
+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 --storage.tsdb.retention.time=540d"
+  timeout_stop_sec 300
+  notifies :restart, "service[prometheus]"
+end
+
 template "/etc/prometheus/prometheus.yml" do
   source "prometheus.yml.erb"
   owner "root"
   group "root"
   mode "644"
-  variables :jobs => jobs
+  variables :jobs => jobs, :junos_targets => junos_targets, :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]"
+  subscribes :restart, "archive_file[#{cache_dir}/prometheus.linux-amd64.tar.gz]"
+end
+
+systemd_service "prometheus-alertmanager" do
+  description "Prometheus alert manager"
+  type "simple"
+  user "prometheus"
+  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"
+  exec_reload "/bin/kill -HUP $MAINPID"
+  timeout_stop_sec 20
+  restart "on-failure"
+  notifies :restart, "service[prometheus-alertmanager]"
+end
+
+link "/usr/local/bin/promtool" do
+  to "/opt/prometheus-server/prometheus/promtool"
+end
+
+template "/etc/prometheus/alertmanager.yml" do
+  source "alertmanager.yml.erb"
+  owner "root"
+  group "root"
+  mode "644"
+end
+
+directory "/var/lib/prometheus/alertmanager" do
+  owner "prometheus"
+  group "prometheus"
+  mode "755"
+end
+
+service "prometheus-alertmanager" do
+  action [:enable, :start]
+  subscribes :reload, "template[/etc/prometheus/alertmanager.yml]"
+  subscribes :restart, "systemd_service[prometheus-alertmanager]"
+  subscribes :restart, "archive_file[#{cache_dir}/alertmanager.linux-amd64.tar.gz]"
+end
+
+directory "/etc/amtool" do
+  owner "root"
+  group "root"
+  mode "755"
+end
+
+template "/etc/amtool/config.yml" do
+  source "amtool.yml.erb"
+  owner "root"
+  group "root"
+  mode "644"
+end
+
+link "/usr/local/bin/amtool" do
+  to "/opt/prometheus-server/alertmanager/amtool"
+end
+
+template "/etc/prometheus/karma.yml" do
+  source "karma.yml.erb"
+  owner "root"
+  group "root"
+  mode "644"
+end
+
+systemd_service "prometheus-karma" do
+  description "Alert dashboard for Prometheus Alertmanager"
+  user "prometheus"
+  exec_start "/opt/prometheus-server/karma/karma-linux-amd64 --config.file=/etc/prometheus/karma.yml"
+  sandbox :enable_network => true
+  restart "on-failure"
+end
+
+service "prometheus-karma" do
+  action [:enable, :start]
+  subscribes :restart, "template[/etc/prometheus/karma.yml]"
+  subscribes :restart, "archive_file[#{cache_dir}/karma-linux-amd64.tar.gz]"
+  subscribes :restart, "systemd_service[prometheus-karma]"
 end
 
 package "grafana-enterprise"
@@ -88,6 +332,7 @@ end
 
 apache_module "alias"
 apache_module "proxy_http"
+apache_module "proxy_wstunnel"
 
 ssl_certificate "prometheus.openstreetmap.org" do
   domains ["prometheus.openstreetmap.org", "prometheus.osm.org"]
@@ -96,4 +341,12 @@ end
 
 apache_site "prometheus.openstreetmap.org" do
   template "apache.erb"
+  variables :admin_hosts => admins["hosts"]
+end
+
+template "/etc/cron.daily/prometheus-backup" do
+  source "backup.cron.erb"
+  owner "root"
+  group "root"
+  mode "750"
 end