X-Git-Url: https://git.openstreetmap.org/chef.git/blobdiff_plain/3b16ed0abdd8dd032d6d9b6d66f01d59d6749bfb..8ba2e5010812ff169ada588834ef5606d166d526:/cookbooks/prometheus/recipes/server.rb

diff --git a/cookbooks/prometheus/recipes/server.rb b/cookbooks/prometheus/recipes/server.rb
index bbb7fbae5..045e9ff34 100644
--- a/cookbooks/prometheus/recipes/server.rb
+++ b/cookbooks/prometheus/recipes/server.rb
@@ -24,6 +24,7 @@ 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
@@ -32,15 +33,85 @@ prometheus_exporter "fastly" do
   environment "FASTLY_API_TOKEN" => tokens["fastly"]
 end
 
-package "prometheus"
+cache_dir = Chef::Config[:file_cache_path]
 
-promscale_version = "0.1.4"
+prometheus_version = "2.31.1"
+alertmanager_version = "0.23.0"
+karma_version = "0.93"
 
-database_cluster = node[:timescaledb][:cluster]
+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]"
+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]"
+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]"
+end
+
+package %w[
+  prometheus
+  prometheus-alertmanager
+]
+
+promscale_version = "0.10.0"
+promscale_extension_version = "0.3.2"
+
+database_version = node[:timescaledb][:database_version]
+database_cluster = "#{database_version}/main"
 
 postgresql_user "prometheus" do
   cluster database_cluster
-  createrole true
+  superuser true
 end
 
 postgresql_database "promscale" do
@@ -48,41 +119,23 @@ 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"
   mode "755"
 end
 
-package %w[
-  make
-  gcc
-  clang-9
-  llvm-9
-  cargo
-]
-
-git "/opt/promscale/extension" do
-  action :sync
-  repository "https://github.com/timescale/promscale_extension.git"
-  revision "0.1.1"
-  user "root"
+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
 
-execute "/opt/promscale/extension/Makefile" do
-  action :nothing
-  command "make install"
-  cwd "/opt/promscale/extension"
-  user "root"
-  group "root"
-  subscribes :run, "git[/opt/promscale/extension]", :immediately
-  notifies :restart, "service[postgresql]", :immediately
+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
@@ -103,44 +156,40 @@ systemd_service "promscale" do
   description "Promscale Connector"
   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 "/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
 
-service "promscale" do
-  action [:enable, :start]
-  subscribes :restart, "remote_file[/opt/promscale/bin/promscale]"
-  subscribes :restart, "systemd_service[promscale]"
-end
-
-postgresql_extension "promscale" do
-  cluster database_cluster
-  database "promscale"
-  notifies :restart, "service[promscale]"
+if node[:prometheus][:promscale]
+  service "promscale" do
+    action [:enable, :start]
+    subscribes :restart, "remote_file[/opt/promscale/bin/promscale]"
+    subscribes :restart, "systemd_service[promscale]"
+  end
+else
+  service "promscale" do
+    action [:disable, :stop]
+  end
 end
 
-systemd_service "promscale-maintenance" do
-  description "Promscale Maintenace"
-  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
+search(:node, "roles:gateway") do |gateway|
+  allowed_ips = gateway.interfaces(:role => :internal).map do |interface|
+    "#{interface[:network]}/#{interface[:prefix]}"
+  end
 
-systemd_timer "promscale-maintenance" do
-  description "Promscale Maintenace"
-  on_active_sec 1800
-  on_unit_inactive_sec 1800
+  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"
@@ -155,14 +204,66 @@ 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]
+      metric_relabel = exporter[:metric_relabel] || []
     else
       name = key
       address = exporter
+      sni = nil
+      metric_relabel = []
     end
 
     jobs[name] ||= []
-    jobs[name] << { :address => address, :name => client.name }
+    jobs[name] << {
+      :address => address,
+      :sni => sni,
+      :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
+
+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"
+  notifies :restart, "service[prometheus]"
 end
 
 template "/etc/prometheus/prometheus.yml" do
@@ -170,12 +271,73 @@ 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 :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"
+  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"
+  notifies :restart, "service[prometheus-alertmanager]"
+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 :reload, "template[/etc/prometheus/alertmanager.yml]"
+  subscribes :restart, "archive_file[#{cache_dir}/alertmanager.linux-amd64.tar.gz]"
+end
+
+template "/etc/prometheus/amtool.yml" do
+  source "amtool.yml.erb"
+  owner "root"
+  group "root"
+  mode "644"
+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"
+  private_tmp true
+  private_devices true
+  protect_system "full"
+  protect_home true
+  no_new_privileges true
+  restart "on-failure"
+end
+
+service "prometheus-karma" do
+  action [:enable, :start]
+  subscribes :reload, "template[/etc/prometheus/karma.yml]"
+  subscribes :restart, "archive_file[#{cache_dir}/karma-linux-amd64.tar.gz]"
 end
 
 package "grafana-enterprise"
@@ -203,4 +365,12 @@ 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