# -*- encoding: binary -*- # :enddoc: require 'rainbows' Rainbows.forked = true module Zbatery VERSION = "3.3.0" class << self # runs the Zbatery HttpServer with +app+ and +options+ and does # not return until the server has exited. def run(app, options = {}) Rainbows::HttpServer.new(app, options).start.join end end Rainbows::Const::RACK_DEFAULTS["SERVER_SOFTWARE"] = "Zbatery #{VERSION}" # we don't actually fork workers, but allow using the # {before,after}_fork hooks found in Unicorn/Rainbows! # config files... FORK_HOOK = lambda { |_,_| } end # :stopdoc: # override stuff we don't need or can't use portably module Rainbows module Base # master == worker in our case def init_worker_process(worker) after_fork.call(self, worker) worker.user(*user) if user.kind_of?(Array) && ! worker.switched build_app! unless preload_app Rainbows::Response.setup(self.class) Rainbows::MaxBody.setup Rainbows::ProcessClient.const_set(:APP, @app) logger.info "Zbatery #@use worker_connections=#@worker_connections" end end # we can't/don't need to do the fchmod heartbeat Unicorn/Rainbows! does def self.tick alive end class HttpServer # this class is only used to avoid breaking Unicorn user switching class DeadIO def chown(*args); end alias fcntl chown end # only used if no concurrency model is specified def worker_loop(worker) init_worker_process(worker) begin ret = IO.select(LISTENERS, nil, nil, nil) and ret[0].each do |sock| io = sock.kgio_tryaccept and process_client(io) end rescue Errno::EINTR rescue Errno::EBADF, TypeError break rescue => e Rainbows::Error.listen_loop(e) end while Rainbows.alive end # no-op def maintain_worker_count; end def init_self_pipe!; end # can't just do a graceful exit if reopening logs fails, so we just # continue on... def reopen_logs logger.info "reopening logs" Unicorn::Util.reopen_logs logger.info "done reopening logs" rescue => e logger.error "failed reopening logs #{e.message}" end def trap_deferred(sig) # nothing end def join trap(:INT) { stop(false) } trap(:TERM) { stop(false) } trap(:QUIT) { stop } trap(:USR1) { reopen_logs } trap(:USR2) { reexec } trap(:HUP) { reexec; stop } # technically feasible in some cases, just not sanely supportable: %w(TTIN TTOU WINCH).each do |sig| trap(sig) { logger.info "SIG#{sig} is not handled by Zbatery" } end if ready_pipe ready_pipe.syswrite($$.to_s) ready_pipe.close self.ready_pipe = nil end extend(Rainbows.const_get(@use)) worker = Worker.new(0, DeadIO.new) before_fork.call(self, worker) worker_loop(worker) # runs forever end def stop(graceful = true) Rainbows.quit! exit!(0) unless graceful end def before_fork hook = super hook == Zbatery::FORK_HOOK or logger.warn "calling before_fork without forking" hook end def after_fork hook = super hook == Zbatery::FORK_HOOK or logger.warn "calling after_fork without having forked" hook end end end Unicorn::Configurator::DEFAULTS[:before_fork] = Unicorn::Configurator::DEFAULTS[:after_fork] = Zbatery::FORK_HOOK