Update site power limit for Amsterdam to 3.5kVA

[chef.git] / cookbooks / prometheus / recipes / server.rb

diff --git a/cookbooks/prometheus/recipes/server.rb b/cookbooks/prometheus/recipes/server.rb
index 02bc587d4fb082516f5e22a84b6d6758cc808c1b..1c4f8e2d20f8a9500e91a8bc0ae04609f88d5355 100644 (file)
--- a/cookbooks/prometheus/recipes/server.rb
+++ b/cookbooks/prometheus/recipes/server.rb
@@ -48,9 +48,9 @@ end
 
 cache_dir = Chef::Config[:file_cache_path]
 
-prometheus_version = "2.39.1"
+prometheus_version = "2.41.0"
 alertmanager_version = "0.24.0"
-karma_version = "0.108"
+karma_version = "0.111"
 
 directory "/opt/prometheus-server" do
   owner "root"
@@ -73,7 +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
@@ -91,7 +91,7 @@ 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
 
 remote_file "#{cache_dir}/karma-linux-amd64.tar.gz" do
@@ -108,20 +108,29 @@ archive_file "#{cache_dir}/karma-linux-amd64.tar.gz" do
   overwrite true
   owner "root"
   group "root"
-  subscribes :extract, "remote_file[#{cache_dir}/karma-linux-amd64.tar.gz]"
+  subscribes :extract, "remote_file[#{cache_dir}/karma-linux-amd64.tar.gz]", :immediately
 end
 
-promscale_version = "0.16.0"
+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
-  prometheus-alertmanager
   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
@@ -175,9 +184,7 @@ else
 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],
@@ -266,7 +273,7 @@ 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 --storage.tsdb.retention.time=30d"
+  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
@@ -293,10 +300,14 @@ service "prometheus" do
   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
 
@@ -311,9 +322,16 @@ 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
 
@@ -351,7 +369,7 @@ end
 
 service "prometheus-karma" do
   action [:enable, :start]
-  subscribes :reload, "template[/etc/prometheus/karma.yml]"
+  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
@@ -373,6 +391,7 @@ 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"]