X-Git-Url: https://git.openstreetmap.org/chef.git/blobdiff_plain/b71ac1f7665947274b96382fa8f2c9f0a3af67d2..2835205c501da417ff80ce4f06dc32d14c8c38bb:/cookbooks/prometheus/recipes/server.rb

diff --git a/cookbooks/prometheus/recipes/server.rb b/cookbooks/prometheus/recipes/server.rb
index 9c215778e..9066c2ec8 100644
--- a/cookbooks/prometheus/recipes/server.rb
+++ b/cookbooks/prometheus/recipes/server.rb
@@ -18,24 +18,39 @@
 #
 
 include_recipe "apache"
-include_recipe "apt"
+include_recipe "apt::grafana"
+include_recipe "apt::timescaledb"
 include_recipe "networking"
 include_recipe "timescaledb"
 
 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
 
+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 = "2.31.1"
-alertmanager_version = "0.23.0"
+prometheus_version = "2.41.0"
+alertmanager_version = "0.24.0"
+karma_version = "0.111"
 
 directory "/opt/prometheus-server" do
   owner "root"
@@ -58,7 +73,7 @@ archive_file "#{cache_dir}/prometheus.linux-amd64.tar.gz" do
   strip_components 1
   owner "root"
   group "root"
-  subscribes :extract, "remote_file[#{cache_dir}/prometheus.linux-amd64.tar.gz]"
+  subscribes :extract, "remote_file[#{cache_dir}/prometheus.linux-amd64.tar.gz]", :immediately
 end
 
 remote_file "#{cache_dir}/alertmanager.linux-amd64.tar.gz" do
@@ -76,19 +91,46 @@ archive_file "#{cache_dir}/alertmanager.linux-amd64.tar.gz" do
   strip_components 1
   owner "root"
   group "root"
-  subscribes :extract, "remote_file[#{cache_dir}/alertmanager.linux-amd64.tar.gz]"
+  subscribes :extract, "remote_file[#{cache_dir}/alertmanager.linux-amd64.tar.gz]", :immediately
 end
 
-package %w[
-  prometheus
-  prometheus-alertmanager
-]
+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
 
-promscale_version = "0.6.2"
+promscale_version = "0.17.0"
+promscale_extension_version = "0.8.0-1"
 
 database_version = node[:timescaledb][:database_version]
 database_cluster = "#{database_version}/main"
 
+package %W[
+  prometheus
+  promscale-extension-postgresql-#{database_version}
+]
+
+package "promscale-extension-postgresql-#{database_version}" do
+  version promscale_extension_version
+end
+
+apt_preference "promscale-extension-postgresql" do
+  pin "version #{promscale_extension_version}"
+  pin_priority "1100"
+end
+
 postgresql_user "prometheus" do
   cluster database_cluster
   superuser true
@@ -105,13 +147,6 @@ directory "/opt/promscale" do
   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"
@@ -130,30 +165,10 @@ 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" }
+  exec_start "/opt/promscale/bin/promscale --db.uri postgresql:///promscale?host=/run/postgresql&port=5432 --db.connections-max 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
+  sandbox :enable_network => true
+  restrict_address_families "AF_UNIX"
 end
 
 if node[:prometheus][:promscale]
@@ -162,24 +177,14 @@ if node[:prometheus][:promscale]
     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
+  allowed_ips = gateway.ipaddresses(:role => :internal).map(&:subnet)
 
   node.default[:networking][:wireguard][:peers] << {
     :public_key => gateway[:networking][:wireguard][:public_key],
@@ -204,17 +209,26 @@ 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]
+      scrape_interval = exporter[:scrape_interval]
+      scrape_timeout = exporter[:scrape_timeout]
       metric_relabel = exporter[:metric_relabel] || []
     else
       name = key
       address = exporter
+      sni = nil
+      scrape_interval = nil
+      scrape_timeout = nil
       metric_relabel = []
     end
 
     jobs[name] ||= []
     jobs[name] << {
       :address => address,
+      :sni => sni,
       :instance => client.name.split(".").first,
+      :scrape_interval => scrape_interval,
+      :scrape_timeout => scrape_timeout,
       :metric_relabel => metric_relabel
     }
   end
@@ -223,7 +237,7 @@ search(:node, "recipes:prometheus\\:\\:default").sort_by(&:name).each do |client
     snmp_targets << {
       :instance => instance,
       :target => details[:address],
-      :module => details[:module],
+      :modules => details[:modules],
       :address => client[:prometheus][:addresses]["snmp"],
       :labels => Array(details[:labels])
     }
@@ -259,7 +273,8 @@ 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"
+  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=30d"
+  timeout_stop_sec 300
   notifies :restart, "service[prometheus]"
 end
 
@@ -282,15 +297,24 @@ 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-executable" do
-  service "prometheus-alertmanager"
-  dropin "executable"
+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"
@@ -298,18 +322,58 @@ template "/etc/prometheus/alertmanager.yml" do
   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
 
-template "/etc/prometheus/amtool.yml" do
+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"
 
 template "/etc/grafana/grafana.ini" do
@@ -335,6 +399,7 @@ end
 
 apache_site "prometheus.openstreetmap.org" do
   template "apache.erb"
+  variables :admin_hosts => admins["hosts"]
 end
 
 template "/etc/cron.daily/prometheus-backup" do