Drop memory constraint on low zoom render threads

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

diff --git a/cookbooks/tile/recipes/default.rb b/cookbooks/tile/recipes/default.rb
index b62c89d1dccaa655ddf4725f3b773bf831e05f6e..6cfe88d45f31ff69540737cbd908a0a365292b25 100644 (file)
--- a/cookbooks/tile/recipes/default.rb
+++ b/cookbooks/tile/recipes/default.rb
@@ -29,6 +29,7 @@ include_recipe "ruby"
 include_recipe "tools"
 
 blocks = data_bag_item("tile", "blocks")
 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"
 web_passwords = data_bag_item("web", "passwords")
 
 apache_module "alias"


@@ -59,6 +60,14 @@ end
 
 fastlyips = JSON.parse(IO.read("#{Chef::Config[:file_cache_path]}/fastly-ip-list.json"))
 
 
 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
 apache_site "default" do
   action :disable
 end


@@ -69,7 +78,9 @@ end
 
 apache_site "tile.openstreetmap.org" do
   template "apache.erb"
 
 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
 end
 
 template "/etc/logrotate.d/apache2" do


@@ -85,38 +96,30 @@ directory "/srv/tile.openstreetmap.org" do
   mode "755"
 end
 
   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"
 
 systemd_service "renderd" do
   dropin "chef"
   after "postgresql.service"
   wants "postgresql.service"

+  environment "LD_PRELOAD" => "libtcmalloc.so.4"

   limit_nofile 4096
   limit_nofile 4096

-  private_tmp true
-  private_devices true
-  private_network true
-  protect_system "full"
-  protect_home true
-  no_new_privileges true
+  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
 
   restart "on-failure"
 end
 

-systemd_service "renderd" do
-  action :delete
-end
-

 service "renderd" do
   action [:enable, :start]
   subscribes :restart, "systemd_service[renderd]"
 service "renderd" do
   action [:enable, :start]
   subscribes :restart, "systemd_service[renderd]"


@@ -186,11 +189,22 @@ template "/srv/tile.openstreetmap.org/cgi-bin/debug" do
   mode "755"
 end
 
   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
 end
 
 directory "/srv/tile.openstreetmap.org/data" do


@@ -273,22 +287,20 @@ end
 
 nodejs_package "carto"
 
 
 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"
   runtime_directory "update-lowzoom-%i"
 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"
   runtime_directory "update-lowzoom-%i"

-  private_tmp true
-  private_devices true
-  private_network true
-  protect_system "strict"
-  protect_home true
+  sandbox true
+  restrict_address_families "AF_UNIX"

   read_write_paths [
     "/srv/tile.openstreetmap.org/tiles/%i",
     "/var/log/tile"
   ]
   read_write_paths [
     "/srv/tile.openstreetmap.org/tiles/%i",
     "/var/log/tile"
   ]

-  no_new_privileges true

   restart "on-failure"
 end
 
   restart "on-failure"
 end
 


@@ -307,7 +319,7 @@ node[:tile][:styles].each do |name, details|
     owner "root"
     group "root"
     mode "755"
     owner "root"
     group "root"
     mode "755"

-    variables :style => name
+    variables :style => name, :threads => lowzoom_threads

   end
 
   service "update-lowzoom@#{name}" do
   end
 
   service "update-lowzoom@#{name}" do


@@ -435,7 +447,7 @@ postgresql_extension "hstore" do
   only_if { node[:tile][:database][:hstore] }
 end
 
   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"
   postgresql_table table do
     cluster node[:tile][:database][:cluster]
     database "gis"


@@ -444,7 +456,7 @@ end
   end
 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"
   postgresql_table table do
     cluster node[:tile][:database][:cluster]
     database "gis"


@@ -506,61 +518,14 @@ package %w[
   osm2pgsql
   osmium-tool
   pyosmium
   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 "/var/lib/replicate" do
+  owner "tile"
+  group "tile"

   mode "755"
 end
 
   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
-  private_tmp true
-  private_devices true
-  private_network true
-  protect_system "strict"
-  protect_home true
-  read_write_paths "/srv/tile.openstreetmap.org/conf"
-  no_new_privileges true
-  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"
 template "/usr/local/bin/expire-tiles" do
   source "expire-tiles.erb"
   owner "root"


@@ -568,12 +533,6 @@ template "/usr/local/bin/expire-tiles" do
   mode "755"
 end
 
   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"
 directory "/var/lib/replicate/expire-queue" do
   owner "tile"
   group "_renderd"


@@ -585,7 +544,6 @@ template "/usr/local/bin/replicate" do
   owner "root"
   group "root"
   mode "755"
   owner "root"
   group "root"
   mode "755"

-  variables :postgresql_version => postgresql_version.to_f

 end
 
 systemd_service "expire-tiles" do
 end
 
 systemd_service "expire-tiles" do


@@ -594,18 +552,11 @@ systemd_service "expire-tiles" do
   user "_renderd"
   exec_start "/usr/local/bin/expire-tiles"
   nice 10
   user "_renderd"
   exec_start "/usr/local/bin/expire-tiles"
   nice 10

-  standard_output "null"
-  private_tmp true
-  private_devices true
-  protect_system "strict"
-  protect_home true
-  read_write_paths [
-    "/store/database/nodes",
-    "/store/tiles/%i",
-    "/var/lib/replicate/expire-queue",
-    "/var/log/tile"
-  ]
-  no_new_privileges true
+  sandbox true
+  restrict_address_families "AF_UNIX"
+  read_write_paths tile_directories + [
+                     "/var/lib/replicate/expire-queue"
+                   ]

 end
 
 systemd_path "expire-tiles" do
 end
 
 systemd_path "expire-tiles" do


@@ -618,22 +569,36 @@ service "expire-tiles.path" do
   subscribes :restart, "systemd_path[expire-tiles]"
 end
 
   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
+
+osm2pgsql_arguments = %w[
+    --number-processes=1
+    --log-progress=false
+    --expire-tiles=13-19
+    --expire-output=/var/lib/replicate/dirty-tiles.txt
+  ]
+
+osm2pgsql_arguments.append("--multi-geometry") if node[:tile][:database][:multi_geometry]
+osm2pgsql_arguments.append("--hstore") if node[:tile][:database][:hstore]
+osm2pgsql_arguments.append("--tag-transform-script=#{node[:tile][:database][:tag_transform_script]}") if node[:tile][:database][:tag_transform_script]
+

 systemd_service "replicate" do
   description "Rendering database replication service"
   after "postgresql.service"
   wants "postgresql.service"
   user "tile"
   exec_start "/usr/local/bin/replicate"
 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 "strict"
-  protect_home true
+  sandbox :enable_network => true
+  restrict_address_families "AF_UNIX"

   read_write_paths [
     "/store/database/nodes",
   read_write_paths [
     "/store/database/nodes",

-    "/var/lib/replicate",
-    "/var/log/tile"
+    "/var/lib/replicate"

   ]
   ]

-  no_new_privileges true

   restart "on-failure"
 end
 
   restart "on-failure"
 end
 


@@ -643,18 +608,12 @@ service "replicate" do
   subscribes :restart, "systemd_service[replicate]"
 end
 
   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"
 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
 end
 
 systemd_service "render-lowzoom" do


@@ -662,18 +621,14 @@ systemd_service "render-lowzoom" do
   condition_path_exists_glob "!/run/update-lowzoom-*"
   user "tile"
   exec_start "/usr/local/bin/render-lowzoom"
   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 "strict"
-  protect_home true
+  sandbox true
+  restrict_address_families "AF_UNIX"

   read_write_paths "/var/log/tile"
   read_write_paths "/var/log/tile"

-  no_new_privileges true

 end
 
 systemd_timer "render-lowzoom" do
   description "Render low zoom tiles"
 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
 end
 
 service "render-lowzoom.timer" do


@@ -690,18 +645,27 @@ template "/usr/local/bin/cleanup-tiles" do
   mode "755"
 end
 
   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|
 
 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
 
   end
 end