X-Git-Url: https://git.openstreetmap.org/chef.git/blobdiff_plain/9083b42d96701a996123e7275170d64b27f66bfa..10b1e5bbc72ae497f41b911af431cbb757a12f36:/cookbooks/prometheus/recipes/server.rb

diff --git a/cookbooks/prometheus/recipes/server.rb b/cookbooks/prometheus/recipes/server.rb
index 0b6e2a0d2..84b34afc2 100644
--- a/cookbooks/prometheus/recipes/server.rb
+++ b/cookbooks/prometheus/recipes/server.rb
@@ -18,7 +18,8 @@
 #
 
 include_recipe "apache"
-include_recipe "apt"
+include_recipe "apt::grafana"
+include_recipe "apt::timescaledb"
 include_recipe "networking"
 include_recipe "timescaledb"
 
@@ -28,16 +29,28 @@ 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"
-karma_version = "0.93"
+prometheus_version = "2.39.1"
+alertmanager_version = "0.24.0"
+karma_version = "0.108"
 
 directory "/opt/prometheus-server" do
   owner "root"
@@ -98,17 +111,17 @@ archive_file "#{cache_dir}/karma-linux-amd64.tar.gz" do
   subscribes :extract, "remote_file[#{cache_dir}/karma-linux-amd64.tar.gz]"
 end
 
-package %w[
-  prometheus
-  prometheus-alertmanager
-]
-
-promscale_version = "0.10.0"
-promscale_extension_version = "0.3.2"
+promscale_version = "0.16.0"
 
 database_version = node[:timescaledb][:database_version]
 database_cluster = "#{database_version}/main"
 
+package %W[
+  prometheus
+  prometheus-alertmanager
+  promscale-extension-postgresql-#{database_version}
+]
+
 postgresql_user "prometheus" do
   cluster database_cluster
   superuser true
@@ -125,19 +138,6 @@ directory "/opt/promscale" do
   mode "755"
 end
 
-remote_file "#{cache_dir}/promscale_extension.pg#{database_version}.x86_64.deb" do
-  source "https://github.com/timescale/promscale_extension/releases/download/#{promscale_extension_version}/promscale_extension-#{promscale_extension_version}.pg#{database_version}.x86_64.deb"
-  owner "root"
-  group "root"
-  mode "644"
-  backup false
-end
-
-dpkg_package "promscale-extension-postgresql-#{database_version}" do
-  source "#{cache_dir}/promscale_extension.pg#{database_version}.x86_64.deb"
-  version "#{promscale_extension_version}-1"
-end
-
 directory "/opt/promscale/bin" do
   owner "root"
   group "root"
@@ -159,46 +159,22 @@ systemd_service "promscale" do
   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
+  private_devices 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
-end
-
 if node[:prometheus][:promscale]
   service "promscale" do
     action [:enable, :start]
     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|
@@ -229,17 +205,26 @@ 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
@@ -248,7 +233,7 @@ 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])
     }
@@ -284,7 +269,8 @@ 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
 
@@ -317,6 +303,10 @@ systemd_service "prometheus-alertmanager-executable" do
   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"
@@ -330,13 +320,23 @@ service "prometheus-alertmanager" do
   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"
@@ -350,7 +350,7 @@ systemd_service "prometheus-karma" do
   exec_start "/opt/prometheus-server/karma/karma-linux-amd64 --config.file=/etc/prometheus/karma.yml"
   private_tmp true
   private_devices true
-  protect_system "full"
+  protect_system "strict"
   protect_home true
   no_new_privileges true
   restart "on-failure"