994f003b5f
The callbacks aren't used in cluster mode, and puma prints a warning about that. This is fine, we don't need the callbacks in single-mode, but can still keep it, in case somebody switches on cluster mode. The problem is, the current puma version has a bug, where pumactl crashes when trying to print these warnings, so lets just silence the warnings. People running in single mode also don't need to care about the warnings anyway.
54 lines
1.9 KiB
Ruby
54 lines
1.9 KiB
Ruby
# frozen_string_literal: true
|
|
|
|
require_relative "load_config"
|
|
|
|
pidfile AppConfig.server.pid.get
|
|
bind AppConfig.server.listen.get
|
|
|
|
worker_timeout AppConfig.server.web_timeout.to_i
|
|
|
|
if AppConfig.server.stdout_log? || AppConfig.server.stderr_log?
|
|
stdout_redirect AppConfig.server.stdout_log? ? AppConfig.server.stdout_log.get : "/dev/null",
|
|
AppConfig.server.stderr_log? ? AppConfig.server.stderr_log.get : "/dev/null"
|
|
end
|
|
|
|
# In general, running Puma in cluster-mode is one of those very rare setups
|
|
# that's only relevant in *huge* scale. However, starting 1 worker runs Puma in
|
|
# cluster mode, with a single worker. This means you get to pay all the memory
|
|
# overhead of spawning in "cluster mode", but you don't get any performance
|
|
# benefits. This makes no sense. Setting "workers = 0" explicitly turns off
|
|
# cluster mode.
|
|
#
|
|
# For more details and further references, see
|
|
# https://github.com/puma/puma/commit/81d26e91b777ab120e8f52d45385f0e018438ba4
|
|
workers 0
|
|
|
|
preload_app!
|
|
|
|
# Below callbacks (before_fork and on_worker_boot) are only used when running
|
|
# in cluster mode. Since we're not running in cluster mode by default (see
|
|
# workers above), a warning is written on startup about them not being called
|
|
# in single mode. So lets just silence that warnings, but still keep the
|
|
# callbacks in place in case somebody switches on the cluster mode.
|
|
silence_fork_callback_warning
|
|
|
|
before_fork do
|
|
# we're preloading app in production, so force-reconenct the DB
|
|
ActiveRecord::Base.connection_pool.disconnect!
|
|
|
|
# drop the Redis connection
|
|
Sidekiq.redis {|redis| redis.client.disconnect }
|
|
end
|
|
|
|
on_worker_boot do
|
|
# reopen logfiles to obtain a new file descriptor
|
|
Logging.reopen
|
|
|
|
ActiveSupport.on_load(:active_record) do
|
|
# we're preloading app in production, so reconnect to DB
|
|
ActiveRecord::Base.establish_connection
|
|
end
|
|
|
|
# We don't generate uuids in the frontend, but let's be on the safe side
|
|
UUID.generator.next_sequence
|
|
end
|