Switch to using upstream builds for prometheus and alertmanager

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

diff --git a/cookbooks/prometheus/recipes/server.rb b/cookbooks/prometheus/recipes/server.rb
index e4689dbf15346468154e4b3e8a2b73a51368369b..9c215778e202a0ee035d9ccf60601f0105b9a330 100644 (file)
--- a/cookbooks/prometheus/recipes/server.rb
+++ b/cookbooks/prometheus/recipes/server.rb
@@ -32,18 +32,66 @@ prometheus_exporter "fastly" do
   environment "FASTLY_API_TOKEN" => tokens["fastly"]
 end
 
   environment "FASTLY_API_TOKEN" => tokens["fastly"]
 end
 

+cache_dir = Chef::Config[:file_cache_path]
+
+prometheus_version = "2.31.1"
+alertmanager_version = "0.23.0"
+
+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
+

 package %w[
   prometheus
   prometheus-alertmanager
 ]
 
 package %w[
   prometheus
   prometheus-alertmanager
 ]
 

-promscale_version = "0.1.4"
+promscale_version = "0.6.2"

 
 

-database_cluster = node[:timescaledb][:cluster]
+database_version = node[:timescaledb][:database_version]
+database_cluster = "#{database_version}/main"

 
 postgresql_user "prometheus" do
   cluster database_cluster
 
 postgresql_user "prometheus" do
   cluster database_cluster

-  createrole true
+  superuser true

 end
 
 postgresql_database "promscale" do
 end
 
 postgresql_database "promscale" do


@@ -51,41 +99,17 @@ postgresql_database "promscale" do
   owner "prometheus"
 end
 
   owner "prometheus"
 end
 

-postgresql_extension "timescaledb" do
-  cluster database_cluster
-  database "promscale"
-end
-

 directory "/opt/promscale" do
   owner "root"
   group "root"
   mode "755"
 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"
-  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"
   group "root"

-  subscribes :run, "git[/opt/promscale/extension]", :immediately
-  notifies :restart, "service[postgresql]", :immediately
+  mode "644"

 end
 
 directory "/opt/promscale/bin" do
 end
 
 directory "/opt/promscale/bin" do


@@ -106,7 +130,7 @@ systemd_service "promscale" do
   description "Promscale Connector"
   type "simple"
   user "prometheus"
   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 "/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" }
   limit_nofile 16384
   private_tmp true
   # 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" }
   limit_nofile 16384
   private_tmp true


@@ -115,20 +139,8 @@ systemd_service "promscale" do
   no_new_privileges true
 end
 
   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]"
-end
-

 systemd_service "promscale-maintenance" do
 systemd_service "promscale-maintenance" do

-  description "Promscale Maintenace"
+  description "Promscale Maintenance"

   type "simple"
   user "prometheus"
   exec_start "/usr/bin/psql --command='CALL prom_api.execute_maintenance()' promscale"
   type "simple"
   user "prometheus"
   exec_start "/usr/bin/psql --command='CALL prom_api.execute_maintenance()' promscale"


@@ -144,7 +156,40 @@ systemd_timer "promscale-maintenance" do
   on_unit_inactive_sec 1800
 end
 
   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|
+  allowed_ips = gateway.interfaces(:role => :internal).map do |interface|
+    "#{interface[:network]}/#{interface[:prefix]}"
+  end
+
+  node.default[:networking][:wireguard][:peers] << {
+    :public_key => gateway[:networking][:wireguard][:public_key],
+    :allowed_ips => allowed_ips,
+    :endpoint => "#{gateway.name}:51820"
+  }
+end
+

 jobs = {}
 jobs = {}

+snmp_targets = []

 
 search(:node, "recipes:prometheus\\:\\:default").sort_by(&:name).each do |client|
   if client[:prometheus][:mode] == "wireguard"
 
 search(:node, "recipes:prometheus\\:\\:default").sort_by(&:name).each do |client|
   if client[:prometheus][:mode] == "wireguard"


@@ -173,13 +218,49 @@ search(:node, "recipes:prometheus\\:\\:default").sort_by(&:name).each do |client
       :metric_relabel => metric_relabel
     }
   end
       :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
 
 end
 

-template "/etc/default/prometheus" do
-  source "default.prometheus.erb"
+template "/etc/prometheus/ssl.yml" do
+  source "ssl.yml.erb"

   owner "root"
   group "root"
   mode "644"
   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
 end
 
 template "/etc/prometheus/prometheus.yml" do


@@ -187,7 +268,7 @@ template "/etc/prometheus/prometheus.yml" do
   owner "root"
   group "root"
   mode "644"
   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
 end
 
 template "/etc/prometheus/alert_rules.yml" do


@@ -199,16 +280,15 @@ end
 
 service "prometheus" do
   action [:enable, :start]
 
 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]"
 end
 
   subscribes :reload, "template[/etc/prometheus/prometheus.yml]"
   subscribes :reload, "template[/etc/prometheus/alert_rules.yml]"
 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
 end
 
 template "/etc/prometheus/alertmanager.yml" do


@@ -220,7 +300,6 @@ end
 
 service "prometheus-alertmanager" do
   action [:enable, :start]
 
 service "prometheus-alertmanager" do
   action [:enable, :start]

-  subscribes :restart, "template[/etc/default/prometheus-alertmanager]"

   subscribes :reload, "template[/etc/prometheus/alertmanager.yml]"
 end
 
   subscribes :reload, "template[/etc/prometheus/alertmanager.yml]"
 end
 


@@ -257,3 +336,10 @@ end
 apache_site "prometheus.openstreetmap.org" do
   template "apache.erb"
 end
 apache_site "prometheus.openstreetmap.org" do
   template "apache.erb"
 end

+
+template "/etc/cron.daily/prometheus-backup" do
+  source "backup.cron.erb"
+  owner "root"
+  group "root"
+  mode "750"
+end