Improve monitoring of primary database for api, tile and nominatim

[chef.git] / cookbooks / postgresql / recipes / default.rb

diff --git a/cookbooks/postgresql/recipes/default.rb b/cookbooks/postgresql/recipes/default.rb
index affe9f502d50084d620cb7c09794805943b44f25..3bf0ad154e6ca39e09a1819a4ca5e06c5bad386c 100644 (file)
--- a/cookbooks/postgresql/recipes/default.rb
+++ b/cookbooks/postgresql/recipes/default.rb
@@ -35,7 +35,6 @@ node[:postgresql][:versions].each do |version|
 
   standby_mode = settings[:standby_mode] || defaults[:standby_mode]
   primary_conninfo = settings[:primary_conninfo] || defaults[:primary_conninfo]
-  restore_command = settings[:restore_command] || defaults[:restore_command]
 
   passwords = if primary_conninfo
                 data_bag_item(primary_conninfo[:passwords][:bag],
@@ -87,28 +86,7 @@ node[:postgresql][:versions].each do |version|
     only_if { ::Dir.exist?("/var/lib/postgresql/#{version}/main") }
   end
 
-  if version.to_f < 12 && (restore_command || standby_mode == "on")
-    template "/var/lib/postgresql/#{version}/main/recovery.conf" do
-      source "recovery.conf.erb"
-      owner "postgres"
-      group "postgres"
-      mode "640"
-      variables :standby_mode => standby_mode,
-                :primary_conninfo => primary_conninfo,
-                :restore_command => restore_command,
-                :passwords => passwords
-      notifies :reload, "service[postgresql]"
-      only_if { ::Dir.exist?("/var/lib/postgresql/#{version}/main") }
-    end
-  else
-    template "/var/lib/postgresql/#{version}/main/recovery.conf" do
-      action :delete
-      notifies :reload, "service[postgresql]"
-      only_if { ::Dir.exist?("/var/lib/postgresql/#{version}/main") }
-    end
-  end
-
-  if version.to_f > 11 && standby_mode == "on"
+  if standby_mode == "on"
     file "/var/lib/postgresql/#{version}/main/standby.signal" do
       owner "postgres"
       group "postgres"
@@ -130,39 +108,63 @@ ohai_plugin "postgresql" do
   template "ohai.rb.erb"
 end
 
+template "/etc/prometheus/exporters/postgres_queries.yml" do
+  source "postgres_queries.yml.erb"
+  owner "root"
+  group "root"
+  mode "644"
+end
+
 package "pgtop"
 package "libdbd-pg-perl"
 
 clusters = node[:postgresql][:clusters] || []
 
 clusters.each do |name, details|
-  suffix = name.tr("/", ":")
+  prometheus_suffix = name.tr("/", "-")
+  prometheus_database = node[:postgresql][:monitor_database]
+
+  prometheus_exporter "postgres" do
+    port 10000 + details[:port].to_i
+    service "postgres-#{prometheus_suffix}"
+    labels "cluster" => name
+    scrape_interval "1m"
+    scrape_timeout "1m"
+    user "postgres"
+    options "--no-collector.process_idle --extend.query-path=/etc/prometheus/exporters/postgres_queries.yml"
+    environment "DATA_SOURCE_NAME" => "postgres:///#{prometheus_database}?host=/run/postgresql&port=#{details[:port]}"
+    restrict_address_families "AF_UNIX"
+    remove_ipc false
+    subscribes :restart, "template[/etc/prometheus/exporters/postgres_queries.yml]"
+  end
+
+  munin_suffix = name.tr("/", ":")
 
-  munin_plugin "postgres_bgwriter_#{suffix}" do
+  munin_plugin "postgres_bgwriter_#{munin_suffix}" do
     target "postgres_bgwriter"
     conf "munin.erb"
     conf_variables :port => details[:port]
   end
 
-  munin_plugin "postgres_checkpoints_#{suffix}" do
+  munin_plugin "postgres_checkpoints_#{munin_suffix}" do
     target "postgres_checkpoints"
     conf "munin.erb"
     conf_variables :port => details[:port]
   end
 
-  munin_plugin "postgres_connections_db_#{suffix}" do
+  munin_plugin "postgres_connections_db_#{munin_suffix}" do
     target "postgres_connections_db"
     conf "munin.erb"
     conf_variables :port => details[:port]
   end
 
-  munin_plugin "postgres_users_#{suffix}" do
+  munin_plugin "postgres_users_#{munin_suffix}" do
     target "postgres_users"
     conf "munin.erb"
     conf_variables :port => details[:port]
   end
 
-  munin_plugin "postgres_xlog_#{suffix}" do
+  munin_plugin "postgres_xlog_#{munin_suffix}" do
     target "postgres_xlog"
     conf "munin.erb"
     conf_variables :port => details[:port]
@@ -170,34 +172,17 @@ clusters.each do |name, details|
 
   next unless File.exist?("/var/lib/postgresql/#{details[:version]}/main/recovery.conf")
 
-  munin_plugin "postgres_replication_#{suffix}" do
+  munin_plugin "postgres_replication_#{munin_suffix}" do
     target "postgres_replication"
     conf "munin.erb"
     conf_variables :port => details[:port]
   end
 end
 
-uris = clusters.collect do |_, details|
-  "postgres@:#{details[:port]}/postgres?host=/run/postgresql"
-end
-
-template "/etc/prometheus/exporters/postgres_queries.yml" do
-  source "postgres_queries.yml.erb"
-  owner "root"
-  group "root"
-  mode "644"
+service "prometheus-postgres-exporter" do
+  action [:stop, :disable]
 end
 
-prometheus_exporter "postgres" do
-  port 9187
-  scrape_interval "1m"
-  scrape_timeout "1m"
-  user "postgres"
-  options "--extend.query-path=/etc/prometheus/exporters/postgres_queries.yml"
-  environment "DATA_SOURCE_URI" => uris.sort.uniq.first,
-              "PG_EXPORTER_AUTO_DISCOVER_DATABASES" => "true",
-              "PG_EXPORTER_EXCLUDE_DATABASES" => "postgres,template0,template1"
-  restrict_address_families "AF_UNIX"
-  remove_ipc false
-  subscribes :restart, "template[/etc/prometheus/exporters/postgres_queries.yml]"
+systemd_service "prometheus-postgres-exporter" do
+  action :delete
 end