X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/54e60013a7d0c78409ef63538ff65f6a2e30b20e..32f0f0ef975a814886142e25acc96f9d0b045457:/cookbooks/prometheus/recipes/server.rb

diff --git a/cookbooks/prometheus/recipes/server.rb b/cookbooks/prometheus/recipes/server.rb
index 822dfb19f..2df77a603 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
@@ -32,14 +33,80 @@ prometheus_exporter "fastly" do
   environment "FASTLY_API_TOKEN" => tokens["fastly"]
 end
 
+cache_dir = Chef::Config[:file_cache_path]
+
+prometheus_version = "2.31.1"
+alertmanager_version = "0.23.0"
+karma_version = "0.93"
+
+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 = "0.5.0"
+promscale_version = "0.7.1"
 
-database_cluster = "#{node[:timescaledb][:database_version]}/main"
+database_version = node[:timescaledb][:database_version]
+database_cluster = "#{database_version}/main"
 
 postgresql_user "prometheus" do
   cluster database_cluster
@@ -57,30 +124,11 @@ directory "/opt/promscale" do
   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 "0.2.0"
-  user "root"
-  group "root"
-end
-
-execute "/opt/promscale/extension/Makefile" do
-  action :nothing
-  command "make install"
-  cwd "/opt/promscale/extension"
-  user "root"
+cookbook_file "/usr/lib/postgresql/#{database_version}/lib/promscale.so" do
+  source "postgresql-#{database_version}-promscale.so"
+  owner "root"
   group "root"
-  subscribes :run, "git[/opt/promscale/extension]", :immediately
-  notifies :restart, "service[postgresql]", :immediately
+  mode "644"
 end
 
 directory "/opt/promscale/bin" do
@@ -110,12 +158,6 @@ systemd_service "promscale" do
   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
-
 systemd_service "promscale-maintenance" do
   description "Promscale Maintenance"
   type "simple"
@@ -133,8 +175,24 @@ systemd_timer "promscale-maintenance" do
   on_unit_inactive_sec 1800
 end
 
-service "promscale-maintenance.timer" do
-  action [:enable, :start]
+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|
@@ -217,11 +275,11 @@ prometheus_exporter "ssl" do
   register_target false
 end
 
-template "/etc/default/prometheus" do
-  source "default.prometheus.erb"
-  owner "root"
-  group "root"
-  mode "644"
+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
@@ -241,16 +299,16 @@ 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]"
+  subscribes :restart, "archive_file[#{cache_dir}/prometheus.linux-amd64.tar.gz]"
 end
 
-template "/etc/default/prometheus-alertmanager" do
-  source "default.alertmanager.erb"
-  owner "root"
-  group "root"
-  mode "644"
+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
@@ -262,8 +320,8 @@ end
 
 service "prometheus-alertmanager" do
   action [:enable, :start]
-  subscribes :restart, "template[/etc/default/prometheus-alertmanager]"
   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
@@ -273,6 +331,31 @@ template "/etc/prometheus/amtool.yml" do
   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"
 
 template "/etc/grafana/grafana.ini" do
@@ -298,6 +381,7 @@ end
 
 apache_site "prometheus.openstreetmap.org" do
   template "apache.erb"
+  variables :admin_hosts => admins["hosts"]
 end
 
 template "/etc/cron.daily/prometheus-backup" do