X-Git-Url: https://git.openstreetmap.org/chef.git/blobdiff_plain/5bd9046d1c65355ebd8d56eca3183ee00e9e03b3..68e068818d559ef35bdf2a138a93596337828ef9:/cookbooks/tile/recipes/default.rb

diff --git a/cookbooks/tile/recipes/default.rb b/cookbooks/tile/recipes/default.rb
index 9b46ec20a..d1a394b2f 100644
--- a/cookbooks/tile/recipes/default.rb
+++ b/cookbooks/tile/recipes/default.rb
@@ -25,9 +25,11 @@ 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"
@@ -41,6 +43,10 @@ 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]"
@@ -52,16 +58,29 @@ remote_file "#{Chef::Config[:file_cache_path]}/fastly-ip-list.json" do
   ignore_failure true
 end
 
-tilecaches = search(:node, "roles:tilecache").sort_by { |n| n[:hostname] }
 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, :fastly => fastlyips["addresses"]
+  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
@@ -77,24 +96,32 @@ directory "/srv/tile.openstreetmap.org" do
   mode "755"
 end
 
+directory "/srv/tile.openstreetmap.org/conf" do
+  owner "tile"
+  group "tile"
+  mode "755"
+end
+
+file "/srv/tile.openstreetmap.org/conf/ip.map" do
+  owner "tile"
+  group "adm"
+  mode "644"
+end
+
+tile_directories = node[:tile][:styles].collect do |_, style|
+  style[:tile_directories].collect { |directory| directory[:name] }
+end.flatten.sort.uniq
+
 package "renderd"
 
 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"
   limit_nofile 4096
-  memory_max "32G"
-  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 tile_directories
   restart "on-failure"
 end
 
@@ -138,6 +165,7 @@ end
 package %w[
   python3-cairo
   python3-mapnik
+  python3-pyproj
   python3-setuptools
 ]
 
@@ -145,24 +173,6 @@ 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"
@@ -184,11 +194,22 @@ 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
@@ -197,7 +218,11 @@ 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]
@@ -216,8 +241,6 @@ node[:tile][:data].each_value do |data|
   end
 
   if file =~ /\.tgz$/
-    package "tar"
-
     execute file do
       action :nothing
       command "tar -zxf #{file} -C #{directory}"
@@ -225,8 +248,6 @@ node[:tile][:data].each_value do |data|
       group "tile"
     end
   elsif file =~ /\.tar\.bz2$/
-    package "tar"
-
     execute file do
       action :nothing
       command "tar -jxf #{file} -C #{directory}"
@@ -234,8 +255,6 @@ 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}"
@@ -275,16 +294,16 @@ nodejs_package "carto"
 
 systemd_service "update-lowzoom@" do
   description "Low zoom tile update service for %i layer"
-  conflicts "render-lowzoom.service"
   user "tile"
+  exec_start_pre "+/bin/systemctl stop render-lowzoom.service"
   exec_start "/bin/bash /usr/local/bin/update-lowzoom-%i"
   runtime_directory "update-lowzoom-%i"
-  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 [
+    "/srv/tile.openstreetmap.org/tiles/%i",
+    "/var/log/tile"
+  ]
   restart "on-failure"
 end
 
@@ -319,15 +338,15 @@ 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
 
@@ -360,9 +379,20 @@ 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"
@@ -387,6 +417,11 @@ 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
@@ -395,6 +430,10 @@ 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"
@@ -425,7 +464,7 @@ end
     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
 
@@ -437,10 +476,20 @@ package %w[
 
 if node[:tile][:database][:external_data_script]
   execute node[:tile][:database][:external_data_script] do
-    command "#{node[:tile][:database][:external_data_script]} -R www-data"
+    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
 
@@ -458,7 +507,7 @@ end
 
 file node[:tile][:database][:node_file] do
   owner "tile"
-  group "www-data"
+  group "_renderd"
   mode "660"
 end
 
@@ -470,12 +519,23 @@ end
 
 package %w[
   osm2pgsql
-  ruby
   osmium-tool
   pyosmium
   python3-pyproj
 ]
 
+gem_package "apachelogregex" do
+  gem_binary node[:ruby][:gem]
+end
+
+gem_package "file-tail" do
+  gem_binary node[:ruby][:gem]
+end
+
+gem_package "lru_redux" do
+  gem_binary node[:ruby][:gem]
+end
+
 remote_directory "/usr/local/bin" do
   source "bin"
   owner "root"
@@ -486,6 +546,31 @@ remote_directory "/usr/local/bin" do
   files_mode "755"
 end
 
+template "/usr/local/bin/tile-ratelimit" do
+  source "tile-ratelimit.erb"
+  owner "root"
+  group "root"
+  mode "755"
+end
+
+systemd_service "tile-ratelimit" do
+  description "Monitor tile requests and enforce rate limits"
+  after "apache2.service"
+  user "tile"
+  group "adm"
+  exec_start "/usr/local/bin/tile-ratelimit"
+  nice 10
+  sandbox true
+  read_write_paths "/srv/tile.openstreetmap.org/conf"
+  restart "on-failure"
+end
+
+service "tile-ratelimit" do
+  action [:enable, :start]
+  subscribes :restart, "file[/usr/local/bin/tile-ratelimit]"
+  subscribes :restart, "systemd_service[tile-ratelimit]"
+end
+
 template "/usr/local/bin/expire-tiles" do
   source "expire-tiles.erb"
   owner "root"
@@ -501,7 +586,7 @@ end
 
 directory "/var/lib/replicate/expire-queue" do
   owner "tile"
-  group "www-data"
+  group "_renderd"
   mode "775"
 end
 
@@ -516,14 +601,16 @@ end
 systemd_service "expire-tiles" do
   description "Tile dirtying service"
   type "simple"
-  user "www-data"
+  user "_renderd"
   exec_start "/usr/local/bin/expire-tiles"
+  nice 10
   standard_output "null"
-  private_tmp true
-  private_devices true
-  protect_system "full"
-  protect_home true
-  no_new_privileges true
+  sandbox true
+  read_write_paths tile_directories + [
+    "/store/database/nodes",
+    "/var/lib/replicate/expire-queue",
+    "/var/log/tile"
+  ]
 end
 
 systemd_path "expire-tiles" do
@@ -542,11 +629,13 @@ systemd_service "replicate" do
   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",
+    "/var/log/tile"
+  ]
   restart "on-failure"
 end
 
@@ -575,17 +664,14 @@ systemd_service "render-lowzoom" do
   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 "Fri *-*-* 23:00:00 UTC"
 end
 
 service "render-lowzoom.timer" do
@@ -602,18 +688,27 @@ 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
 
@@ -630,6 +725,8 @@ munin_plugin "renderd_zoom_time"
 
 munin_plugin "replication_delay"
 
+package "ruby-webrick"
+
 prometheus_exporter "modtile" do
   port 9494
 end