X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/aeaa46f3071de018b5c5e0f5678cb5b4082e3ddc..4b03c18aff06d5f34a587fd90af9af0f528e4c61:/cookbooks/tile/recipes/default.rb?ds=sidebyside

diff --git a/cookbooks/tile/recipes/default.rb b/cookbooks/tile/recipes/default.rb
index 5a7a33af8..eab19d75d 100644
--- a/cookbooks/tile/recipes/default.rb
+++ b/cookbooks/tile/recipes/default.rb
@@ -20,13 +20,15 @@
 include_recipe "accounts"
 include_recipe "apache"
 include_recipe "git"
-include_recipe "munin"
 include_recipe "nodejs"
 include_recipe "postgresql"
+include_recipe "prometheus"
 include_recipe "python"
+include_recipe "ruby"
 include_recipe "tools"
 
 blocks = data_bag_item("tile", "blocks")
+admins = data_bag_item("apache", "admins")
 web_passwords = data_bag_item("web", "passwords")
 
 apache_module "alias"
@@ -40,20 +42,44 @@ apache_module "tile" do
   conf "tile.conf.erb"
 end
 
+apache_conf "renderd" do
+  action :disable
+end
+
 ssl_certificate node[:fqdn] do
   domains [node[:fqdn], "tile.openstreetmap.org", "render.openstreetmap.org"]
   notifies :reload, "service[apache2]"
 end
 
-tilecaches = search(:node, "roles:tilecache").sort_by { |n| n[:hostname] }
+remote_file "#{Chef::Config[:file_cache_path]}/fastly-ip-list.json" do
+  source "https://api.fastly.com/public-ip-list"
+  compile_time true
+  ignore_failure true
+end
+
+fastlyips = JSON.parse(IO.read("#{Chef::Config[:file_cache_path]}/fastly-ip-list.json"))
+
+remote_file "#{Chef::Config[:file_cache_path]}/statuscake-locations.json" do
+  source "https://app.statuscake.com/Workfloor/Locations.php?format=json"
+  compile_time true
+  ignore_failure true
+end
+
+statuscakelocations = JSON.parse(IO.read("#{Chef::Config[:file_cache_path]}/statuscake-locations.json"))
 
 apache_site "default" do
-  action [:disable]
+  action :disable
+end
+
+apache_site "tileserver_site" do
+  action :disable
 end
 
 apache_site "tile.openstreetmap.org" do
   template "apache.erb"
-  variables :caches => tilecaches
+  variables :fastly => fastlyips["addresses"] + fastlyips["ipv6_addresses"],
+            :statuscake => statuscakelocations.flat_map { |_, v| [v["ip"], v["ipv6"]] },
+            :admins => admins["hosts"]
 end
 
 template "/etc/logrotate.d/apache2" do
@@ -69,22 +95,27 @@ directory "/srv/tile.openstreetmap.org" do
   mode "755"
 end
 
-package "renderd"
+tile_directories = node[:tile][:styles].collect do |_, style|
+  style[:tile_directories].collect { |directory| directory[:name] }
+end.flatten.sort.uniq
+
+package %w[
+  renderd
+  libgoogle-perftools4
+]
 
 systemd_service "renderd" do
-  description "Mapnik rendering daemon"
+  dropin "chef"
   after "postgresql.service"
   wants "postgresql.service"
-  user "www-data"
-  exec_start "/usr/bin/renderd -f"
-  runtime_directory "renderd"
-  standard_error "null"
-  private_tmp true
-  private_devices true
-  private_network true
-  protect_system "full"
-  protect_home true
-  no_new_privileges true
+  environment "LD_PRELOAD" => "libtcmalloc.so.4"
+  limit_nofile 4096
+  memory_high "80%"
+  memory_max "90%"
+  sandbox true
+  restrict_address_families "AF_UNIX"
+  read_write_paths tile_directories
+  system_call_filter ["@system-service", "mincore"]
   restart "on-failure"
 end
 
@@ -115,7 +146,7 @@ remote_directory "/srv/tile.openstreetmap.org/html" do
   mode "755"
   files_owner "tile"
   files_group "tile"
-  files_mode 0o644
+  files_mode "644"
 end
 
 template "/srv/tile.openstreetmap.org/html/index.html" do
@@ -128,6 +159,7 @@ end
 package %w[
   python3-cairo
   python3-mapnik
+  python3-pyproj
   python3-setuptools
 ]
 
@@ -135,24 +167,6 @@ python_package "pyotp" do
   python_version "3"
 end
 
-package %w[
-  fonts-noto-cjk
-  fonts-noto-hinted
-  fonts-noto-unhinted
-  fonts-hanazono
-  ttf-unifont
-]
-
-["NotoSansArabicUI-Regular.ttf", "NotoSansArabicUI-Bold.ttf"].each do |font|
-  remote_file "/usr/share/fonts/truetype/noto/#{font}" do
-    action :create_if_missing
-    source "https://github.com/googlei18n/noto-fonts/raw/master/hinted/#{font}"
-    owner "root"
-    group "root"
-    mode "644"
-  end
-end
-
 directory "/srv/tile.openstreetmap.org/cgi-bin" do
   owner "tile"
   group "tile"
@@ -174,11 +188,22 @@ template "/srv/tile.openstreetmap.org/cgi-bin/debug" do
   mode "755"
 end
 
-template "/etc/cron.hourly/export" do
-  source "export.cron.erb"
-  owner "root"
-  group "root"
-  mode "755"
+systemd_service "export-cleanup" do
+  description "Cleanup stale export temporary files"
+  joins_namespace_of "apache2.service"
+  exec_start "find /tmp -ignore_readdir_race -name 'export??????' -mmin +60 -delete"
+  user "www-data"
+  sandbox true
+end
+
+systemd_timer "export-cleanup" do
+  description "Cleanup stale export temporary files"
+  on_boot_sec "60m"
+  on_unit_inactive_sec "60m"
+end
+
+service "export-cleanup.timer" do
+  action [:enable, :start]
 end
 
 directory "/srv/tile.openstreetmap.org/data" do
@@ -187,7 +212,11 @@ directory "/srv/tile.openstreetmap.org/data" do
   mode "755"
 end
 
-package "mapnik-utils"
+package %w[
+  mapnik-utils
+  tar
+  unzip
+]
 
 node[:tile][:data].each_value do |data|
   url = data[:url]
@@ -206,8 +235,6 @@ node[:tile][:data].each_value do |data|
   end
 
   if file =~ /\.tgz$/
-    package "tar"
-
     execute file do
       action :nothing
       command "tar -zxf #{file} -C #{directory}"
@@ -215,8 +242,6 @@ node[:tile][:data].each_value do |data|
       group "tile"
     end
   elsif file =~ /\.tar\.bz2$/
-    package "tar"
-
     execute file do
       action :nothing
       command "tar -jxf #{file} -C #{directory}"
@@ -224,8 +249,6 @@ node[:tile][:data].each_value do |data|
       group "tile"
     end
   elsif file =~ /\.zip$/
-    package "unzip"
-
     execute file do
       action :nothing
       command "unzip -qq -o #{file} -d #{directory}"
@@ -263,16 +286,20 @@ end
 
 nodejs_package "carto"
 
+lowzoom_threads = node.cpu_cores - 1
+
 systemd_service "update-lowzoom@" do
   description "Low zoom tile update service for %i layer"
   user "tile"
+  exec_start_pre "+/bin/systemctl stop render-lowzoom.service"
   exec_start "/bin/bash /usr/local/bin/update-lowzoom-%i"
-  private_tmp true
-  private_devices true
-  private_network true
-  protect_system "full"
-  protect_home true
-  no_new_privileges true
+  runtime_directory "update-lowzoom-%i"
+  sandbox true
+  restrict_address_families "AF_UNIX"
+  read_write_paths [
+    "/srv/tile.openstreetmap.org/tiles/%i",
+    "/var/log/tile"
+  ]
   restart "on-failure"
 end
 
@@ -291,7 +318,7 @@ node[:tile][:styles].each do |name, details|
     owner "root"
     group "root"
     mode "755"
-    variables :style => name
+    variables :style => name, :threads => lowzoom_threads
   end
 
   service "update-lowzoom@#{name}" do
@@ -307,15 +334,15 @@ node[:tile][:styles].each do |name, details|
 
   details[:tile_directories].each do |directory|
     directory directory[:name] do
-      owner "www-data"
-      group "www-data"
+      owner "_renderd"
+      group "_renderd"
       mode "755"
     end
 
     directory[:min_zoom].upto(directory[:max_zoom]) do |zoom|
       directory "#{directory[:name]}/#{zoom}" do
-        owner "www-data"
-        group "www-data"
+        owner "_renderd"
+        group "_renderd"
         mode "755"
       end
 
@@ -348,9 +375,20 @@ node[:tile][:styles].each do |name, details|
     group "tile"
   end
 
+  if details[:fonts_script]
+    execute details[:fonts_script] do
+      action :nothing
+      command details[:fonts_script]
+      cwd style_directory
+      user "tile"
+      group "tile"
+      subscribes :run, "git[#{style_directory}]"
+    end
+  end
+
   execute "#{style_directory}/project.mml" do
     action :nothing
-    command "carto -a 3.0.0 project.mml > project.xml"
+    command "carto -a 3.0.22 project.mml > project.xml"
     cwd style_directory
     user "tile"
     group "tile"
@@ -375,6 +413,11 @@ postgresql_user "tomh" do
   superuser true
 end
 
+postgresql_user "pnorman" do
+  cluster node[:tile][:database][:cluster]
+  superuser true
+end
+
 postgresql_user "tile" do
   cluster node[:tile][:database][:cluster]
 end
@@ -383,6 +426,10 @@ postgresql_user "www-data" do
   cluster node[:tile][:database][:cluster]
 end
 
+postgresql_user "_renderd" do
+  cluster node[:tile][:database][:cluster]
+end
+
 postgresql_database "gis" do
   cluster node[:tile][:database][:cluster]
   owner "tile"
@@ -399,7 +446,7 @@ postgresql_extension "hstore" do
   only_if { node[:tile][:database][:hstore] }
 end
 
-%w[geography_columns planet_osm_nodes planet_osm_rels planet_osm_ways raster_columns raster_overviews spatial_ref_sys].each do |table|
+%w[geography_columns planet_osm_nodes planet_osm_rels planet_osm_ways raster_columns raster_overviews].each do |table|
   postgresql_table table do
     cluster node[:tile][:database][:cluster]
     database "gis"
@@ -408,18 +455,38 @@ end
   end
 end
 
-%w[geometry_columns planet_osm_line planet_osm_point planet_osm_polygon planet_osm_roads].each do |table|
+%w[geometry_columns planet_osm_line planet_osm_point planet_osm_polygon planet_osm_roads spatial_ref_sys].each do |table|
   postgresql_table table do
     cluster node[:tile][:database][:cluster]
     database "gis"
     owner "tile"
-    permissions "tile" => :all, "www-data" => :select
+    permissions "tile" => :all, "www-data" => :select, "_renderd" => :select
   end
 end
 
-postgresql_munin "gis" do
-  cluster node[:tile][:database][:cluster]
-  database "gis"
+package %w[
+  gdal-bin
+  python3-yaml
+  python3-psycopg2
+]
+
+if node[:tile][:database][:external_data_script]
+  execute node[:tile][:database][:external_data_script] do
+    command "#{node[:tile][:database][:external_data_script]} -R _renderd"
+    cwd "/srv/tile.openstreetmap.org"
+    user "tile"
+    group "tile"
+    ignore_failure true
+  end
+
+  Array(node[:tile][:database][:external_data_tables]).each do |table|
+    postgresql_table table do
+      cluster node[:tile][:database][:cluster]
+      database "gis"
+      owner "tile"
+      permissions "tile" => :all, "www-data" => :select, "_renderd" => :select
+    end
+  end
 end
 
 directory File.dirname(node[:tile][:database][:node_file]) do
@@ -431,7 +498,7 @@ end
 
 file node[:tile][:database][:node_file] do
   owner "tile"
-  group "www-data"
+  group "_renderd"
   mode "660"
 end
 
@@ -443,20 +510,21 @@ end
 
 package %w[
   osm2pgsql
-  ruby
   osmium-tool
   pyosmium
-  python3-pyproj
 ]
 
-remote_directory "/usr/local/bin" do
-  source "bin"
-  owner "root"
-  group "root"
+directory node[:tile][:replication][:directory] do
+  owner "tile"
+  group "tile"
   mode "755"
-  files_owner "root"
-  files_group "root"
-  files_mode 0o755
+end
+
+link "/var/lib/replicate" do
+  to node[:tile][:replication][:directory]
+  owner "tile"
+  group "tile"
+  not_if { node[:tile][:replication][:directory] == "/var/lib/replicate" }
 end
 
 template "/usr/local/bin/expire-tiles" do
@@ -466,15 +534,9 @@ template "/usr/local/bin/expire-tiles" do
   mode "755"
 end
 
-directory "/var/lib/replicate" do
-  owner "tile"
-  group "tile"
-  mode "755"
-end
-
 directory "/var/lib/replicate/expire-queue" do
   owner "tile"
-  group "www-data"
+  group "_renderd"
   mode "775"
 end
 
@@ -483,20 +545,19 @@ template "/usr/local/bin/replicate" do
   owner "root"
   group "root"
   mode "755"
-  variables :postgresql_version => postgresql_version.to_f
 end
 
 systemd_service "expire-tiles" do
   description "Tile dirtying service"
   type "simple"
-  user "www-data"
+  user "_renderd"
   exec_start "/usr/local/bin/expire-tiles"
-  standard_output "null"
-  private_tmp true
-  private_devices true
-  protect_system "full"
-  protect_home true
-  no_new_privileges true
+  nice 10
+  sandbox true
+  restrict_address_families "AF_UNIX"
+  read_write_paths tile_directories + [
+                     "/var/lib/replicate/expire-queue"
+                   ]
 end
 
 systemd_path "expire-tiles" do
@@ -509,17 +570,25 @@ service "expire-tiles.path" do
   subscribes :restart, "systemd_path[expire-tiles]"
 end
 
+template "/usr/local/bin/replicate-post" do
+  source "replicate-post.erb"
+  owner "root"
+  group "root"
+  mode "755"
+end
+
 systemd_service "replicate" do
   description "Rendering database replication service"
   after "postgresql.service"
   wants "postgresql.service"
   user "tile"
   exec_start "/usr/local/bin/replicate"
-  private_tmp true
-  private_devices true
-  protect_system "full"
-  protect_home true
-  no_new_privileges true
+  sandbox :enable_network => true
+  restrict_address_families "AF_UNIX"
+  read_write_paths [
+    "/store/database/nodes",
+    "/var/lib/replicate"
+  ]
   restart "on-failure"
 end
 
@@ -529,35 +598,27 @@ service "replicate" do
   subscribes :restart, "systemd_service[replicate]"
 end
 
-template "/etc/logrotate.d/replicate" do
-  source "replicate.logrotate.erb"
-  owner "root"
-  group "root"
-  mode "644"
-end
-
 template "/usr/local/bin/render-lowzoom" do
   source "render-lowzoom.erb"
   owner "root"
   group "root"
   mode "755"
+  variables :threads => lowzoom_threads
 end
 
 systemd_service "render-lowzoom" do
   description "Render low zoom tiles"
+  condition_path_exists_glob "!/run/update-lowzoom-*"
   user "tile"
   exec_start "/usr/local/bin/render-lowzoom"
-  private_tmp true
-  private_devices true
-  private_network true
-  protect_system "full"
-  protect_home true
-  no_new_privileges true
+  sandbox true
+  restrict_address_families "AF_UNIX"
+  read_write_paths "/var/log/tile"
 end
 
 systemd_timer "render-lowzoom" do
   description "Render low zoom tiles"
-  on_calendar "Sun *-*~07/1 01:00:00"
+  on_calendar "23:00 #{node[:timezone]}"
 end
 
 service "render-lowzoom.timer" do
@@ -574,30 +635,36 @@ template "/usr/local/bin/cleanup-tiles" do
   mode "755"
 end
 
-tile_directories = node[:tile][:styles].collect do |_, style|
-  style[:tile_directories].collect { |directory| directory[:name] }
-end.flatten.sort.uniq
+systemd_service "cleanup-tiles@" do
+  description "Cleanup old tiles for /%I"
+  exec_start "/usr/local/bin/cleanup-tiles /%I"
+  user "_renderd"
+  io_scheduling_class "idle"
+  sandbox true
+  read_write_paths "/%I"
+end
+
+systemd_timer "cleanup-tiles@" do
+  description "Cleanup old tiles for /%I"
+  on_boot_sec "30m"
+  on_unit_inactive_sec "60m"
+  randomized_delay_sec "10m"
+end
 
 tile_directories.each do |directory|
-  label = directory.gsub("/", "-")
+  label = directory[1..].gsub("/", "-")
 
-  cron_d "cleanup-tiles#{label}" do
-    minute "0"
-    user "www-data"
-    command "ionice -c 3 /usr/local/bin/cleanup-tiles #{directory}"
-    mailto "admins@openstreetmap.org"
+  service "cleanup-tiles@#{label}.timer" do
+    action [:enable, :start]
   end
 end
 
-munin_plugin "mod_tile_fresh"
-munin_plugin "mod_tile_latency"
-munin_plugin "mod_tile_response"
-munin_plugin "mod_tile_zoom"
+package "ruby-webrick"
 
-munin_plugin "renderd_processed"
-munin_plugin "renderd_queue"
-munin_plugin "renderd_queue_time"
-munin_plugin "renderd_zoom"
-munin_plugin "renderd_zoom_time"
+prometheus_exporter "modtile" do
+  port 9494
+end
 
-munin_plugin "replication_delay"
+prometheus_exporter "renderd" do
+  port 9393
+end