Add prometheus SNMP monitoring of switch and PDUs in Amsterdam

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

diff --git a/cookbooks/prometheus/recipes/server.rb b/cookbooks/prometheus/recipes/server.rb
index bbb7fbae5d7bc369e22554f2f1d759bf98b7899c..c8241a1f1d3ed0f5a8adcceee5cd360eef37b05c 100644 (file)
--- a/cookbooks/prometheus/recipes/server.rb
+++ b/cookbooks/prometheus/recipes/server.rb
@@ -32,15 +32,18 @@ prometheus_exporter "fastly" do
   environment "FASTLY_API_TOKEN" => tokens["fastly"]
 end
 
-package "prometheus"
+package %w[
+  prometheus
+  prometheus-alertmanager
+]
 
-promscale_version = "0.1.4"
+promscale_version = "0.2.0"
 
-database_cluster = node[:timescaledb][:cluster]
+database_cluster = "#{node[:timescaledb][:database_version]}/main"
 
 postgresql_user "prometheus" do
   cluster database_cluster
-  createrole true
+  superuser true
 end
 
 postgresql_database "promscale" do
@@ -48,11 +51,6 @@ 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"
@@ -105,6 +103,7 @@ systemd_service "promscale" do
   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" }
+  limit_nofile 16384
   private_tmp true
   protect_system "strict"
   protect_home true
@@ -117,12 +116,6 @@ service "promscale" do
   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
   description "Promscale Maintenace"
   type "simple"
@@ -140,7 +133,24 @@ systemd_timer "promscale-maintenance" do
   on_unit_inactive_sec 1800
 end
 
+service "promscale-maintenance.timer" do
+  action [:enable, :start]
+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 = {}
+snmp_targets = []
 
 search(:node, "recipes:prometheus\\:\\:default").sort_by(&:name).each do |client|
   if client[:prometheus][:mode] == "wireguard"
@@ -155,27 +165,111 @@ search(:node, "recipes:prometheus\\:\\:default").sort_by(&:name).each do |client
     if exporter.is_a?(Hash)
       name = exporter[:name]
       address = exporter[:address]
+      metric_relabel = exporter[:metric_relabel] || []
     else
       name = key
       address = exporter
+      metric_relabel = []
     end
 
     jobs[name] ||= []
-    jobs[name] << { :address => address, :name => client.name }
+    jobs[name] << {
+      :address => address,
+      :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"]
+    }
+  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
+
+template "/etc/default/prometheus" do
+  source "default.prometheus.erb"
+  owner "root"
+  group "root"
+  mode "644"
+end
+
 template "/etc/prometheus/prometheus.yml" do
   source "prometheus.yml.erb"
   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 :restart, "template[/etc/default/prometheus]"
   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"
+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 :restart, "template[/etc/default/prometheus-alertmanager]"
+  subscribes :reload, "template[/etc/prometheus/alertmanager.yml]"
+end
+
+template "/etc/prometheus/amtool.yml" do
+  source "amtool.yml.erb"
+  owner "root"
+  group "root"
+  mode "644"
 end
 
 package "grafana-enterprise"
@@ -204,3 +298,10 @@ 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