X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/8f368b10a8cd0e097632874f720c49ce8eacea3a..066b78765de430ece0377b01bac14a9658cc301c:/cookbooks/tile/recipes/default.rb

diff --git a/cookbooks/tile/recipes/default.rb b/cookbooks/tile/recipes/default.rb
index 963c3dd2f..eab19d75d 100644
--- a/cookbooks/tile/recipes/default.rb
+++ b/cookbooks/tile/recipes/default.rb
@@ -20,7 +20,6 @@
 include_recipe "accounts"
 include_recipe "apache"
 include_recipe "git"
-include_recipe "munin"
 include_recipe "nodejs"
 include_recipe "postgresql"
 include_recipe "prometheus"
@@ -29,6 +28,7 @@ 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"
@@ -59,6 +59,14 @@ 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
 end
@@ -69,7 +77,9 @@ end
 
 apache_site "tile.openstreetmap.org" do
   template "apache.erb"
-  variables :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
@@ -85,35 +95,30 @@ 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"
+package %w[
+  renderd
+  libgoogle-perftools4
+]
 
 systemd_service "renderd" do
   dropin "chef"
   after "postgresql.service"
   wants "postgresql.service"
+  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 "/store/tiles"
+  read_write_paths tile_directories
+  system_call_filter ["@system-service", "mincore"]
   restart "on-failure"
 end
 
-systemd_service "renderd" do
-  action :delete
-end
-
 service "renderd" do
   action [:enable, :start]
   subscribes :restart, "systemd_service[renderd]"
@@ -183,11 +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
@@ -270,6 +286,8 @@ 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"
@@ -300,7 +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
@@ -428,7 +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"
@@ -437,7 +455,7 @@ 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"
@@ -471,11 +489,6 @@ if node[:tile][:database][:external_data_script]
   end
 end
 
-postgresql_munin "gis" do
-  cluster node[:tile][:database][:cluster]
-  database "gis"
-end
-
 directory File.dirname(node[:tile][:database][:node_file]) do
   owner "root"
   group "root"
@@ -499,54 +512,19 @@ package %w[
   osm2pgsql
   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"
-  group "root"
-  mode "755"
-  files_owner "root"
-  files_group "root"
-  files_mode "755"
-end
-
-template "/usr/local/bin/tile-ratelimit" do
-  source "tile-ratelimit.erb"
-  owner "root"
-  group "root"
+directory node[:tile][:replication][:directory] do
+  owner "tile"
+  group "tile"
   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]"
+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
@@ -556,12 +534,6 @@ 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 "_renderd"
@@ -573,7 +545,6 @@ 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
@@ -582,14 +553,11 @@ systemd_service "expire-tiles" do
   user "_renderd"
   exec_start "/usr/local/bin/expire-tiles"
   nice 10
-  standard_output "null"
   sandbox true
-  read_write_paths [
-    "/store/database/nodes",
-    "/store/tiles/%i",
-    "/var/lib/replicate/expire-queue",
-    "/var/log/tile"
-  ]
+  restrict_address_families "AF_UNIX"
+  read_write_paths tile_directories + [
+                     "/var/lib/replicate/expire-queue"
+                   ]
 end
 
 systemd_path "expire-tiles" do
@@ -602,6 +570,13 @@ 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"
@@ -612,8 +587,7 @@ systemd_service "replicate" do
   restrict_address_families "AF_UNIX"
   read_write_paths [
     "/store/database/nodes",
-    "/var/lib/replicate",
-    "/var/log/tile"
+    "/var/lib/replicate"
   ]
   restart "on-failure"
 end
@@ -624,18 +598,12 @@ 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
@@ -650,7 +618,7 @@ end
 
 systemd_timer "render-lowzoom" do
   description "Render low zoom tiles"
-  on_calendar "Fri *-*-* 23:00:00 UTC"
+  on_calendar "23:00 #{node[:timezone]}"
 end
 
 service "render-lowzoom.timer" do
@@ -667,34 +635,30 @@ 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 "_renderd"
-    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"
-
-munin_plugin "renderd_processed"
-munin_plugin "renderd_queue"
-munin_plugin "renderd_queue_time"
-munin_plugin "renderd_zoom"
-munin_plugin "renderd_zoom_time"
-
-munin_plugin "replication_delay"
-
 package "ruby-webrick"
 
 prometheus_exporter "modtile" do