puma 3.12.6 → 5.3.2

data/lib/puma/runner.rb

@@ -14,10 +14,7 @@ module Puma
       @options = cli.options
       @app = nil
       @control = nil
-    end
-
-    def daemon?
-      @options[:daemon]
+      @started_at = Time.now
     end
 
     def development?
@@ -32,7 +29,8 @@ module Puma
       @events.log str
     end
 
-    def before_restart
+    # @version 5.0.0
+    def stop_control
       @control.stop(true) if @control
     end
 
@@ -50,36 +48,27 @@ module Puma
 
       require 'puma/app/status'
 
-      uri = URI.parse str
-
-      app = Puma::App::Status.new @launcher
-
       if token = @options[:control_auth_token]
-        app.auth_token = token unless token.empty? or token == :none
+        token = nil if token.empty? || token == 'none'
       end
 
-      control = Puma::Server.new app, @launcher.events
-      control.min_threads = 0
-      control.max_threads = 1
+      app = Puma::App::Status.new @launcher, token
 
-      case uri.scheme
-      when "tcp"
-        log "* Starting control server on #{str}"
-        control.add_tcp_listener uri.host, uri.port
-      when "unix"
-        log "* Starting control server on #{str}"
-        path = "#{uri.host}#{uri.path}"
-        mask = @options[:control_url_umask]
+      control = Puma::Server.new app, @launcher.events,
+        { min_threads: 0, max_threads: 1, queue_requests: false }
 
-        control.add_unix_listener path, mask
-      else
-        error "Invalid control URI: #{str}"
-      end
+      control.binder.parse [str], self, 'Starting control server'
 
-      control.run
+      control.run thread_name: 'control'
       @control = control
     end
 
+    # @version 5.0.0
+    def close_control_listeners
+      @control.binder.close_listeners if @control
+    end
+
+    # @!attribute [r] ruby_engine
     def ruby_engine
       if !defined?(RUBY_ENGINE) || RUBY_ENGINE == "ruby"
         "ruby #{RUBY_VERSION}-p#{RUBY_PATCHLEVEL}"
@@ -97,12 +86,15 @@ module Puma
       max_t = @options[:max_threads]
 
       log "Puma starting in #{mode} mode..."
-      log "* Version #{Puma::Const::PUMA_VERSION} (#{ruby_engine}), codename: #{Puma::Const::CODE_NAME}"
-      log "* Min threads: #{min_t}, max threads: #{max_t}"
-      log "* Environment: #{ENV['RACK_ENV']}"
+      log "* Puma version: #{Puma::Const::PUMA_VERSION} (#{ruby_engine}) (\"#{Puma::Const::CODE_NAME}\")"
+      log "*  Min threads: #{min_t}"
+      log "*  Max threads: #{max_t}"
+      log "*  Environment: #{ENV['RACK_ENV']}"
 
-      if @options[:mode] == :tcp
-        log "* Mode: Lopez Express (tcp)"
+      if mode == "cluster"
+        log "*   Master PID: #{Process.pid}"
+      else
+        log "*          PID: #{Process.pid}"
       end
     end
 
@@ -121,8 +113,8 @@ module Puma
         end
 
         STDOUT.reopen stdout, (append ? "a" : "w")
-        STDOUT.sync = true
         STDOUT.puts "=== puma startup: #{Time.now} ==="
+        STDOUT.flush unless STDOUT.sync
       end
 
       if stderr
@@ -131,8 +123,13 @@ module Puma
         end
 
         STDERR.reopen stderr, (append ? "a" : "w")
-        STDERR.sync = true
         STDERR.puts "=== puma startup: #{Time.now} ==="
+        STDERR.flush unless STDERR.sync
+      end
+
+      if @options[:mutate_stdout_and_stderr_to_sync_on_write]
+        STDOUT.sync = true
+        STDERR.sync = true
       end
     end
 
@@ -142,7 +139,6 @@ module Puma
         exit 1
       end
 
-      # Load the app before we daemonize.
       begin
         @app = @launcher.config.app
       rescue Exception => e
@@ -153,31 +149,14 @@ module Puma
       @launcher.binder.parse @options[:binds], self
     end
 
+    # @!attribute [r] app
     def app
       @app ||= @launcher.config.app
     end
 
     def start_server
-      min_t = @options[:min_threads]
-      max_t = @options[:max_threads]
-
       server = Puma::Server.new app, @launcher.events, @options
-      server.min_threads = min_t
-      server.max_threads = max_t
       server.inherit_binder @launcher.binder
-
-      if @options[:mode] == :tcp
-        server.tcp_mode!
-      end
-
-      if @options[:early_hints]
-        server.early_hints = true
-      end
-
-      unless development? || test?
-        server.leak_stack_on_error = false
-      end
-
       server
     end
   end

data/lib/puma/server.rb

@@ -6,21 +6,15 @@ require 'puma/thread_pool'
 require 'puma/const'
 require 'puma/events'
 require 'puma/null_io'
-require 'puma/compat'
 require 'puma/reactor'
 require 'puma/client'
 require 'puma/binder'
-require 'puma/delegation'
-require 'puma/accept_nonblock'
 require 'puma/util'
-
-require 'puma/puma_http11'
-
-unless Puma.const_defined? "IOBuffer"
-  require 'puma/io_buffer'
-end
+require 'puma/io_buffer'
+require 'puma/request'
 
 require 'socket'
+require 'forwardable'
 
 module Puma
 
@@ -28,7 +22,7 @@ module Puma
   #
   # This class is used by the `Puma::Single` and `Puma::Cluster` classes
   # to generate one or more `Puma::Server` instances capable of handling requests.
-  # Each Puma process will contain one `Puma::Server` instacne.
+  # Each Puma process will contain one `Puma::Server` instance.
   #
   # The `Puma::Server` instance pulls requests from the socket, adds them to a
   # `Puma::Reactor` where they get eventually passed to a `Puma::ThreadPool`.
@@ -37,18 +31,31 @@ module Puma
   class Server
 
     include Puma::Const
-    extend  Puma::Delegation
+    include Request
+    extend Forwardable
 
     attr_reader :thread
     attr_reader :events
+    attr_reader :min_threads, :max_threads  # for #stats
+    attr_reader :requests_count             # @version 5.0.0
+
+    # @todo the following may be deprecated in the future
+    attr_reader :auto_trim_time, :early_hints, :first_data_timeout,
+      :leak_stack_on_error,
+      :persistent_timeout, :reaping_time
+
+    # @deprecated v6.0.0
+    attr_writer :auto_trim_time, :early_hints, :first_data_timeout,
+      :leak_stack_on_error, :min_threads, :max_threads,
+      :persistent_timeout, :reaping_time
+
     attr_accessor :app
+    attr_accessor :binder
+
+    def_delegators :@binder, :add_tcp_listener, :add_ssl_listener,
+      :add_unix_listener, :connected_ports
 
-    attr_accessor :min_threads
-    attr_accessor :max_threads
-    attr_accessor :persistent_timeout
-    attr_accessor :auto_trim_time
-    attr_accessor :reaping_time
-    attr_accessor :first_data_timeout
+    ThreadLocalKey = :puma_server
 
     # Create a server for the rack app +app+.
     #
@@ -58,88 +65,115 @@ module Puma
     # Server#run returns a thread that you can join on to wait for the server
     # to do its work.
     #
+    # @note Several instance variables exist so they are available for testing,
+    #   and have default values set via +fetch+.  Normally the values are set via
+    #   `::Puma::Configuration.puma_default_options`.
+    #
     def initialize(app, events=Events.stdio, options={})
       @app = app
       @events = events
 
-      @check, @notify = Puma::Util.pipe
-
+      @check, @notify = nil
       @status = :stop
 
-      @min_threads = 0
-      @max_threads = 16
       @auto_trim_time = 30
       @reaping_time = 1
 
       @thread = nil
       @thread_pool = nil
-      @early_hints = nil
 
-      @persistent_timeout = options.fetch(:persistent_timeout, PERSISTENT_TIMEOUT)
-      @first_data_timeout = options.fetch(:first_data_timeout, FIRST_DATA_TIMEOUT)
+      @options = options
 
-      @binder = Binder.new(events)
-      @own_binder = true
+      @early_hints         = options.fetch :early_hints, nil
+      @first_data_timeout  = options.fetch :first_data_timeout, FIRST_DATA_TIMEOUT
+      @min_threads         = options.fetch :min_threads, 0
+      @max_threads         = options.fetch :max_threads , (Puma.mri? ? 5 : 16)
+      @persistent_timeout  = options.fetch :persistent_timeout, PERSISTENT_TIMEOUT
+      @queue_requests      = options.fetch :queue_requests, true
+      @max_fast_inline     = options.fetch :max_fast_inline, MAX_FAST_INLINE
+      @io_selector_backend = options.fetch :io_selector_backend, :auto
 
-      @leak_stack_on_error = true
+      temp = !!(@options[:environment] =~ /\A(development|test)\z/)
+      @leak_stack_on_error = @options[:environment] ? temp : true
 
-      @options = options
-      @queue_requests = options[:queue_requests].nil? ? true : options[:queue_requests]
+      @binder = Binder.new(events)
 
       ENV['RACK_ENV'] ||= "development"
 
       @mode = :http
 
       @precheck_closing = true
-    end
 
-    attr_accessor :binder, :leak_stack_on_error, :early_hints
-
-    forward :add_tcp_listener,  :@binder
-    forward :add_ssl_listener,  :@binder
-    forward :add_unix_listener, :@binder
-    forward :connected_port,    :@binder
+      @requests_count = 0
+    end
 
     def inherit_binder(bind)
       @binder = bind
-      @own_binder = false
     end
 
-    def tcp_mode!
-      @mode = :tcp
+    class << self
+      # @!attribute [r] current
+      def current
+        Thread.current[ThreadLocalKey]
+      end
+
+      # :nodoc:
+      # @version 5.0.0
+      def tcp_cork_supported?
+        Socket.const_defined?(:TCP_CORK) && Socket.const_defined?(:IPPROTO_TCP)
+      end
+
+      # :nodoc:
+      # @version 5.0.0
+      def closed_socket_supported?
+        Socket.const_defined?(:TCP_INFO) && Socket.const_defined?(:IPPROTO_TCP)
+      end
+      private :tcp_cork_supported?
+      private :closed_socket_supported?
     end
 
     # On Linux, use TCP_CORK to better control how the TCP stack
     # packetizes our stream. This improves both latency and throughput.
+    # socket parameter may be an MiniSSL::Socket, so use to_io
     #
-    if RUBY_PLATFORM =~ /linux/
-      UNPACK_TCP_STATE_FROM_TCP_INFO = "C".freeze
-
+    if tcp_cork_supported?
       # 6 == Socket::IPPROTO_TCP
       # 3 == TCP_CORK
       # 1/0 == turn on/off
       def cork_socket(socket)
+        skt = socket.to_io
         begin
-          socket.setsockopt(6, 3, 1) if socket.kind_of? TCPSocket
+          skt.setsockopt(Socket::IPPROTO_TCP, Socket::TCP_CORK, 1) if skt.kind_of? TCPSocket
         rescue IOError, SystemCallError
           Thread.current.purge_interrupt_queue if Thread.current.respond_to? :purge_interrupt_queue
         end
       end
 
       def uncork_socket(socket)
+        skt = socket.to_io
         begin
-          socket.setsockopt(6, 3, 0) if socket.kind_of? TCPSocket
+          skt.setsockopt(Socket::IPPROTO_TCP, Socket::TCP_CORK, 0) if skt.kind_of? TCPSocket
         rescue IOError, SystemCallError
           Thread.current.purge_interrupt_queue if Thread.current.respond_to? :purge_interrupt_queue
         end
       end
+    else
+      def cork_socket(socket)
+      end
+
+      def uncork_socket(socket)
+      end
+    end
+
+    if closed_socket_supported?
+      UNPACK_TCP_STATE_FROM_TCP_INFO = "C".freeze
 
       def closed_socket?(socket)
-        return false unless socket.kind_of? TCPSocket
-        return false unless @precheck_closing
+        skt = socket.to_io
+        return false unless skt.kind_of?(TCPSocket) && @precheck_closing
 
         begin
-          tcp_info = socket.getsockopt(Socket::SOL_TCP, Socket::TCP_INFO)
+          tcp_info = skt.getsockopt(Socket::IPPROTO_TCP, Socket::TCP_INFO)
         rescue IOError, SystemCallError
           Thread.current.purge_interrupt_queue if Thread.current.respond_to? :purge_interrupt_queue
           @precheck_closing = false
@@ -151,21 +185,17 @@ module Puma
         end
       end
     else
-      def cork_socket(socket)
-      end
-
-      def uncork_socket(socket)
-      end
-
       def closed_socket?(socket)
         false
       end
     end
 
+    # @!attribute [r] backlog
     def backlog
       @thread_pool and @thread_pool.backlog
     end
 
+    # @!attribute [r] running
     def running
       @thread_pool and @thread_pool.spawned
     end
@@ -178,172 +208,37 @@ module Puma
     # there are 5 threads sitting idle ready to take
     # a request. If one request comes in, then the
     # value would be 4 until it finishes processing.
+    # @!attribute [r] pool_capacity
     def pool_capacity
       @thread_pool and @thread_pool.pool_capacity
     end
 
-    # Lopez Mode == raw tcp apps
-
-    def run_lopez_mode(background=true)
-      @thread_pool = ThreadPool.new(@min_threads,
-                                    @max_threads,
-                                    Hash) do |client, tl|
-
-        io = client.to_io
-        addr = io.peeraddr.last
-
-        if addr.empty?
-          # Set unix socket addrs to localhost
-          addr = "127.0.0.1:0"
-        else
-          addr = "#{addr}:#{io.peeraddr[1]}"
-        end
-
-        env = { 'thread' => tl, REMOTE_ADDR => addr }
-
-        begin
-          @app.call env, client.to_io
-        rescue Object => e
-          STDERR.puts "! Detected exception at toplevel: #{e.message} (#{e.class})"
-          STDERR.puts e.backtrace
-        end
-
-        client.close unless env['detach']
-      end
-
-      @events.fire :state, :running
-
-      if background
-        @thread = Thread.new { handle_servers_lopez_mode }
-        return @thread
-      else
-        handle_servers_lopez_mode
-      end
-    end
-
-    def handle_servers_lopez_mode
-      begin
-        check = @check
-        sockets = [check] + @binder.ios
-        pool = @thread_pool
-
-        while @status == :run
-          begin
-            ios = IO.select sockets
-            ios.first.each do |sock|
-              if sock == check
-                break if handle_check
-              else
-                begin
-                  if io = sock.accept_nonblock
-                    client = Client.new io, nil
-                    pool << client
-                  end
-                rescue SystemCallError
-                  # nothing
-                rescue Errno::ECONNABORTED
-                  # client closed the socket even before accept
-                  begin
-                    io.close
-                  rescue
-                    Thread.current.purge_interrupt_queue if Thread.current.respond_to? :purge_interrupt_queue
-                  end
-                end
-              end
-            end
-          rescue Object => e
-            @events.unknown_error self, e, "Listen loop"
-          end
-        end
-
-        @events.fire :state, @status
-
-        graceful_shutdown if @status == :stop || @status == :restart
-
-      rescue Exception => e
-        STDERR.puts "Exception handling servers: #{e.message} (#{e.class})"
-        STDERR.puts e.backtrace
-      ensure
-        begin
-          @check.close
-        rescue
-          Thread.current.purge_interrupt_queue if Thread.current.respond_to? :purge_interrupt_queue
-        end
-
-        @notify.close
-
-        if @status != :restart and @own_binder
-          @binder.close
-        end
-      end
-
-      @events.fire :state, :done
-    end
     # Runs the server.
     #
     # If +background+ is true (the default) then a thread is spun
     # up in the background to handle requests. Otherwise requests
     # are handled synchronously.
     #
-    def run(background=true)
+    def run(background=true, thread_name: 'server')
       BasicSocket.do_not_reverse_lookup = true
 
       @events.fire :state, :booting
 
       @status = :run
 
-      if @mode == :tcp
-        return run_lopez_mode(background)
-      end
-
-      queue_requests = @queue_requests
-
-      @thread_pool = ThreadPool.new(@min_threads,
-                                    @max_threads,
-                                    IOBuffer) do |client, buffer|
-
-        # Advertise this server into the thread
-        Thread.current[ThreadLocalKey] = self
-
-        process_now = false
-
-        begin
-          if queue_requests
-            process_now = client.eagerly_finish
-          else
-            client.finish
-            process_now = true
-          end
-        rescue MiniSSL::SSLError => e
-          ssl_socket = client.io
-          addr = ssl_socket.peeraddr.last
-          cert = ssl_socket.peercert
-
-          client.close
-
-          @events.ssl_error self, addr, cert, e
-        rescue HttpParserError => e
-          client.write_400
-          client.close
-
-          @events.parse_error self, client.env, e
-        rescue ConnectionError, EOFError
-          client.close
-        else
-          if process_now
-            process_client client, buffer
-          else
-            client.set_timeout @first_data_timeout
-            @reactor.add client
-          end
-        end
-      end
+      @thread_pool = ThreadPool.new(
+        @min_threads,
+        @max_threads,
+        ::Puma::IOBuffer,
+        &method(:process_client)
+      )
 
+      @thread_pool.out_of_band_hook = @options[:out_of_band]
       @thread_pool.clean_thread_locals = @options[:clean_thread_locals]
 
-      if queue_requests
-        @reactor = Reactor.new self, @thread_pool
-        @reactor.run_in_thread
+      if @queue_requests
+        @reactor = Reactor.new(@io_selector_backend, &method(:reactor_wakeup))
+        @reactor.run
       end
 
       if @reaping_time
@@ -354,22 +249,69 @@ module Puma
         @thread_pool.auto_trim!(@auto_trim_time)
       end
 
+      @check, @notify = Puma::Util.pipe unless @notify
+
       @events.fire :state, :running
 
       if background
-        @thread = Thread.new { handle_servers }
+        @thread = Thread.new do
+          Puma.set_thread_name thread_name
+          handle_servers
+        end
         return @thread
       else
         handle_servers
       end
     end
 
+    # This method is called from the Reactor thread when a queued Client receives data,
+    # times out, or when the Reactor is shutting down.
+    #
+    # It is responsible for ensuring that a request has been completely received
+    # before it starts to be processed by the ThreadPool. This may be known as read buffering.
+    # If read buffering is not done, and no other read buffering is performed (such as by an application server
+    # such as nginx) then the application would be subject to a slow client attack.
+    #
+    # For a graphical representation of how the request buffer works see [architecture.md](https://github.com/puma/puma/blob/master/docs/architecture.md#connection-pipeline).
+    #
+    # The method checks to see if it has the full header and body with
+    # the `Puma::Client#try_to_finish` method. If the full request has been sent,
+    # then the request is passed to the ThreadPool (`@thread_pool << client`)
+    # so that a "worker thread" can pick up the request and begin to execute application logic.
+    # The Client is then removed from the reactor (return `true`).
+    #
+    # If a client object times out, a 408 response is written, its connection is closed,
+    # and the object is removed from the reactor (return `true`).
+    #
+    # If the Reactor is shutting down, all Clients are either timed out or passed to the
+    # ThreadPool, depending on their current state (#can_close?).
+    #
+    # Otherwise, if the full request is not ready then the client will remain in the reactor
+    # (return `false`). When the client sends more data to the socket the `Puma::Client` object
+    # will wake up and again be checked to see if it's ready to be passed to the thread pool.
+    def reactor_wakeup(client)
+      shutdown = !@queue_requests
+      if client.try_to_finish || (shutdown && !client.can_close?)
+        @thread_pool << client
+      elsif shutdown || client.timeout == 0
+        client.timeout!
+      else
+        client.set_timeout(@first_data_timeout)
+        false
+      end
+    rescue StandardError => e
+      client_error(e, client)
+      client.close
+      true
+    end
+
     def handle_servers
       begin
         check = @check
         sockets = [check] + @binder.ios
         pool = @thread_pool
         queue_requests = @queue_requests
+        drain = @options[:drain_on_shutdown] ? 0 : nil
 
         remote_addr_value = nil
         remote_addr_header = nil
@@ -381,59 +323,58 @@ module Puma
           remote_addr_header = @options[:remote_address_header]
         end
 
-        while @status == :run
+        while @status == :run || (drain && shutting_down?)
           begin
-            ios = IO.select sockets
+            ios = IO.select sockets, nil, nil, (shutting_down? ? 0 : nil)
+            break unless ios
             ios.first.each do |sock|
               if sock == check
                 break if handle_check
               else
-                begin
-                  if io = sock.accept_nonblock
-                    client = Client.new io, @binder.env(sock)
-                    if remote_addr_value
-                      client.peerip = remote_addr_value
-                    elsif remote_addr_header
-                      client.remote_addr_header = remote_addr_header
-                    end
-
-                    pool << client
-                    pool.wait_until_not_full
-                  end
-                rescue SystemCallError
-                  # nothing
-                rescue Errno::ECONNABORTED
-                  # client closed the socket even before accept
-                  begin
-                    io.close
-                  rescue
-                    Thread.current.purge_interrupt_queue if Thread.current.respond_to? :purge_interrupt_queue
-                  end
+                pool.wait_until_not_full
+                pool.wait_for_less_busy_worker(@options[:wait_for_less_busy_worker])
+
+                io = begin
+                  sock.accept_nonblock
+                rescue IO::WaitReadable
+                  next
                 end
+                drain += 1 if shutting_down?
+                client = Client.new io, @binder.env(sock)
+                client.listener = sock
+                if remote_addr_value
+                  client.peerip = remote_addr_value
+                elsif remote_addr_header
+                  client.remote_addr_header = remote_addr_header
+                end
+                pool << client
               end
             end
           rescue Object => e
-            @events.unknown_error self, e, "Listen loop"
+            @events.unknown_error e, nil, "Listen loop"
           end
         end
 
+        @events.debug "Drained #{drain} additional connections." if drain
         @events.fire :state, @status
 
-        graceful_shutdown if @status == :stop || @status == :restart
         if queue_requests
-          @reactor.clear!
+          @queue_requests = false
           @reactor.shutdown
         end
+        graceful_shutdown if @status == :stop || @status == :restart
       rescue Exception => e
-        STDERR.puts "Exception handling servers: #{e.message} (#{e.class})"
-        STDERR.puts e.backtrace
+        @events.unknown_error e, nil, "Exception handling servers"
       ensure
-        @check.close
-        @notify.close
-
-        if @status != :restart and @own_binder
-          @binder.close
+        begin
+          @check.close unless @check.closed?
+        rescue Errno::EBADF, RuntimeError
+          # RuntimeError is Ruby 2.2 issue, can't modify frozen IOError
+          # Errno::EBADF is infrequently raised
         end
+        @notify.close
+        @notify = nil
+        @check = nil
       end
 
       @events.fire :state, :done
@@ -458,86 +399,83 @@ module Puma
       return false
     end
 
-    # Given a connection on +client+, handle the incoming requests.
+    # Given a connection on +client+, handle the incoming requests,
+    # or queue the connection in the Reactor if no request is available.
     #
-    # This method support HTTP Keep-Alive so it may, depending on if the client
+    # This method is called from a ThreadPool worker thread.
+    #
+    # This method supports HTTP Keep-Alive so it may, depending on if the client
     # indicates that it supports keep alive, wait for another request before
     # returning.
     #
+    # Return true if one or more requests were processed.
     def process_client(client, buffer)
+      # Advertise this server into the thread
+      Thread.current[ThreadLocalKey] = self
+
+      clean_thread_locals = @options[:clean_thread_locals]
+      close_socket = true
+
+      requests = 0
+
       begin
+        if @queue_requests &&
+          !client.eagerly_finish
 
-        clean_thread_locals = @options[:clean_thread_locals]
-        close_socket = true
+          client.set_timeout(@first_data_timeout)
+          if @reactor.add client
+            close_socket = false
+            return false
+          end
+        end
 
-        requests = 0
+        with_force_shutdown(client) do
+          client.finish(@first_data_timeout)
+        end
 
         while true
-          case handle_request(client, buffer)
+          @requests_count += 1
+          case handle_request(client, buffer, requests + 1)
           when false
-            return
+            break
           when :async
             close_socket = false
-            return
+            break
           when true
-            return unless @queue_requests
             buffer.reset
 
             ThreadPool.clean_thread_locals if clean_thread_locals
 
             requests += 1
 
-            check_for_more_data = @status == :run
+            # As an optimization, try to read the next request from the
+            # socket for a short time before returning to the reactor.
+            fast_check = @status == :run
+
+            # Always pass the client back to the reactor after a reasonable
+            # number of inline requests if there are other requests pending.
+            fast_check = false if requests >= @max_fast_inline &&
+              @thread_pool.backlog > 0
 
-            if requests >= MAX_FAST_INLINE
-              # This will mean that reset will only try to use the data it already
-              # has buffered and won't try to read more data. What this means is that
-              # every client, independent of their request speed, gets treated like a slow
-              # one once every MAX_FAST_INLINE requests.
-              check_for_more_data = false
+            next_request_ready = with_force_shutdown(client) do
+              client.reset(fast_check)
             end
 
-            unless client.reset(check_for_more_data)
-              close_socket = false
+            unless next_request_ready
+              break unless @queue_requests
               client.set_timeout @persistent_timeout
-              @reactor.add client
-              return
+              if @reactor.add client
+                close_socket = false
+                break
+              end
             end
           end
         end
-
-      # The client disconnected while we were reading data
-      rescue ConnectionError
-        # Swallow them. The ensure tries to close +client+ down
-
-      # SSL handshake error
-      rescue MiniSSL::SSLError => e
-        lowlevel_error(e, client.env)
-
-        ssl_socket = client.io
-        addr = ssl_socket.peeraddr.last
-        cert = ssl_socket.peercert
-
-        close_socket = true
-
-        @events.ssl_error self, addr, cert, e
-
-      # The client doesn't know HTTP well
-      rescue HttpParserError => e
-        lowlevel_error(e, client.env)
-
-        client.write_400
-
-        @events.parse_error self, client.env, e
-
-      # Server error
+        true
       rescue StandardError => e
-        lowlevel_error(e, client.env)
-
-        client.write_500
-
-        @events.unknown_error self, e, "Read"
-
+        client_error(e, client)
+        # The ensure tries to close +client+ down
+        requests > 0
       ensure
         buffer.reset
 
@@ -547,406 +485,57 @@ module Puma
           Thread.current.purge_interrupt_queue if Thread.current.respond_to? :purge_interrupt_queue
           # Already closed
         rescue StandardError => e
-          @events.unknown_error self, e, "Client"
+          @events.unknown_error e, nil, "Client"
         end
       end
     end
 
-    # Given a Hash +env+ for the request read from +client+, add
-    # and fixup keys to comply with Rack's env guidelines.
-    #
-    def normalize_env(env, client)
-      if host = env[HTTP_HOST]
-        if colon = host.index(":")
-          env[SERVER_NAME] = host[0, colon]
-          env[SERVER_PORT] = host[colon+1, host.bytesize]
-        else
-          env[SERVER_NAME] = host
-          env[SERVER_PORT] = default_server_port(env)
-        end
-      else
-        env[SERVER_NAME] = LOCALHOST
-        env[SERVER_PORT] = default_server_port(env)
-      end
-
-      unless env[REQUEST_PATH]
-        # it might be a dumbass full host request header
-        uri = URI.parse(env[REQUEST_URI])
-        env[REQUEST_PATH] = uri.path
-
-        raise "No REQUEST PATH" unless env[REQUEST_PATH]
-
-        # A nil env value will cause a LintError (and fatal errors elsewhere),
-        # so only set the env value if there actually is a value.
-        env[QUERY_STRING] = uri.query if uri.query
-      end
-
-      env[PATH_INFO] = env[REQUEST_PATH]
-
-      # From http://www.ietf.org/rfc/rfc3875 :
-      # "Script authors should be aware that the REMOTE_ADDR and
-      # REMOTE_HOST meta-variables (see sections 4.1.8 and 4.1.9)
-      # may not identify the ultimate source of the request.
-      # They identify the client for the immediate request to the
-      # server; that client may be a proxy, gateway, or other
-      # intermediary acting on behalf of the actual source client."
-      #
-
-      unless env.key?(REMOTE_ADDR)
-        begin
-          addr = client.peerip
-        rescue Errno::ENOTCONN
-          # Client disconnects can result in an inability to get the
-          # peeraddr from the socket; default to localhost.
-          addr = LOCALHOST_IP
-        end
-
-        # Set unix socket addrs to localhost
-        addr = LOCALHOST_IP if addr.empty?
-
-        env[REMOTE_ADDR] = addr
-      end
-    end
-
-    def default_server_port(env)
-      return PORT_443 if env[HTTPS_KEY] == 'on' || env[HTTPS_KEY] == 'https'
-      env['HTTP_X_FORWARDED_PROTO'] == 'https' ? PORT_443 : PORT_80
-    end
-
-    # Given the request +env+ from +client+ and a partial request body
-    # in +body+, finish reading the body if there is one and invoke
-    # the rack app. Then construct the response and write it back to
-    # +client+
-    #
-    # +cl+ is the previously fetched Content-Length header if there
-    # was one. This is an optimization to keep from having to look
-    # it up again.
-    #
-    def handle_request(req, lines)
-      env = req.env
-      client = req.io
-
-      return false if closed_socket?(client)
-
-      normalize_env env, req
-
-      env[PUMA_SOCKET] = client
-
-      if env[HTTPS_KEY] && client.peercert
-        env[PUMA_PEERCERT] = client.peercert
-      end
-
-      env[HIJACK_P] = true
-      env[HIJACK] = req
-
-      body = req.body
-
-      head = env[REQUEST_METHOD] == HEAD
-
-      env[RACK_INPUT] = body
-      env[RACK_URL_SCHEME] =  env[HTTPS_KEY] ? HTTPS : HTTP
-
-      if @early_hints
-        env[EARLY_HINTS] = lambda { |headers|
-          fast_write client, "HTTP/1.1 103 Early Hints\r\n".freeze
-
-          headers.each_pair do |k, vs|
-            if vs.respond_to?(:to_s) && !vs.to_s.empty?
-              vs.to_s.split(NEWLINE).each do |v|
-                next if possible_header_injection?(v)
-                fast_write client, "#{k}: #{v}\r\n"
-              end
-            else
-              fast_write client, "#{k}: #{vs}\r\n"
-            end
-          end
-
-          fast_write client, "\r\n".freeze
-        }
-      end
-
-      # Fixup any headers with , in the name to have _ now. We emit
-      # headers with , in them during the parse phase to avoid ambiguity
-      # with the - to _ conversion for critical headers. But here for
-      # compatibility, we'll convert them back. This code is written to
-      # avoid allocation in the common case (ie there are no headers
-      # with , in their names), that's why it has the extra conditionals.
-
-      to_delete = nil
-      to_add = nil
-
-      env.each do |k,v|
-        if k.start_with?("HTTP_") and k.include?(",") and k != "HTTP_TRANSFER,ENCODING"
-          if to_delete
-            to_delete << k
-          else
-            to_delete = [k]
-          end
-
-          unless to_add
-            to_add = {}
-          end
-
-          to_add[k.gsub(",", "_")] = v
-        end
-      end
-
-      if to_delete
-        to_delete.each { |k| env.delete(k) }
-        env.merge! to_add
-      end
-
-      # A rack extension. If the app writes #call'ables to this
-      # array, we will invoke them when the request is done.
-      #
-      after_reply = env[RACK_AFTER_REPLY] = []
-
-      begin
-        begin
-          status, headers, res_body = @app.call(env)
-
-          return :async if req.hijacked
-
-          status = status.to_i
-
-          if status == -1
-            unless headers.empty? and res_body == []
-              raise "async response must have empty headers and body"
-            end
-
-            return :async
-          end
-        rescue ThreadPool::ForceShutdown => e
-          @events.log "Detected force shutdown of a thread, returning 503"
-          @events.unknown_error self, e, "Rack app"
-
-          status = 503
-          headers = {}
-          res_body = ["Request was internally terminated early\n"]
-
-        rescue Exception => e
-          @events.unknown_error self, e, "Rack app", env
-
-          status, headers, res_body = lowlevel_error(e, env)
-        end
-
-        content_length = nil
-        no_body = head
-
-        if res_body.kind_of? Array and res_body.size == 1
-          content_length = res_body[0].bytesize
-        end
-
-        cork_socket client
-
-        line_ending = LINE_END
-        colon = COLON
-
-        http_11 = if env[HTTP_VERSION] == HTTP_11
-          allow_chunked = true
-          keep_alive = env.fetch(HTTP_CONNECTION, "").downcase != CLOSE
-          include_keepalive_header = false
-
-          # An optimization. The most common response is 200, so we can
-          # reply with the proper 200 status without having to compute
-          # the response header.
-          #
-          if status == 200
-            lines << HTTP_11_200
-          else
-            lines.append "HTTP/1.1 ", status.to_s, " ",
-                         fetch_status_code(status), line_ending
-
-            no_body ||= status < 200 || STATUS_WITH_NO_ENTITY_BODY[status]
-          end
-          true
-        else
-          allow_chunked = false
-          keep_alive = env.fetch(HTTP_CONNECTION, "").downcase == KEEP_ALIVE
-          include_keepalive_header = keep_alive
-
-          # Same optimization as above for HTTP/1.1
-          #
-          if status == 200
-            lines << HTTP_10_200
-          else
-            lines.append "HTTP/1.0 ", status.to_s, " ",
-                         fetch_status_code(status), line_ending
-
-            no_body ||= status < 200 || STATUS_WITH_NO_ENTITY_BODY[status]
-          end
-          false
-        end
-
-        response_hijack = nil
-
-        headers.each do |k, vs|
-          case k.downcase
-          when CONTENT_LENGTH2
-            next if possible_header_injection?(vs)
-            content_length = vs
-            next
-          when TRANSFER_ENCODING
-            allow_chunked = false
-            content_length = nil
-          when HIJACK
-            response_hijack = vs
-            next
-          end
-
-          if vs.respond_to?(:to_s) && !vs.to_s.empty?
-            vs.to_s.split(NEWLINE).each do |v|
-              next if possible_header_injection?(v)
-              lines.append k, colon, v, line_ending
-            end
-          else
-            lines.append k, colon, line_ending
-          end
-        end
-
-        if include_keepalive_header
-          lines << CONNECTION_KEEP_ALIVE
-        elsif http_11 && !keep_alive
-          lines << CONNECTION_CLOSE
-        end
-
-        if no_body
-          if content_length and status != 204
-            lines.append CONTENT_LENGTH_S, content_length.to_s, line_ending
-          end
-
-          lines << line_ending
-          fast_write client, lines.to_s
-          return keep_alive
-        end
-
-        if content_length
-          lines.append CONTENT_LENGTH_S, content_length.to_s, line_ending
-          chunked = false
-        elsif !response_hijack and allow_chunked
-          lines << TRANSFER_ENCODING_CHUNKED
-          chunked = true
-        end
-
-        lines << line_ending
-
-        fast_write client, lines.to_s
-
-        if response_hijack
-          response_hijack.call client
-          return :async
-        end
-
-        begin
-          res_body.each do |part|
-            next if part.bytesize.zero?
-            if chunked
-              fast_write client, part.bytesize.to_s(16)
-              fast_write client, line_ending
-              fast_write client, part
-              fast_write client, line_ending
-            else
-              fast_write client, part
-            end
-
-            client.flush
-          end
-
-          if chunked
-            fast_write client, CLOSE_CHUNKED
-            client.flush
-          end
-        rescue SystemCallError, IOError
-          raise ConnectionError, "Connection error detected during write"
-        end
-
-      ensure
-        uncork_socket client
-
-        body.close
-        req.tempfile.unlink if req.tempfile
-        res_body.close if res_body.respond_to? :close
-
-        after_reply.each { |o| o.call }
-      end
-
-      return keep_alive
-    end
-
-    def fetch_status_code(status)
-      HTTP_STATUS_CODES.fetch(status) { 'CUSTOM' }
+    # Triggers a client timeout if the thread-pool shuts down
+    # during execution of the provided block.
+    def with_force_shutdown(client, &block)
+      @thread_pool.with_force_shutdown(&block)
+    rescue ThreadPool::ForceShutdown
+      client.timeout!
     end
-    private :fetch_status_code
 
-    # Given the request +env+ from +client+ and the partial body +body+
-    # plus a potential Content-Length value +cl+, finish reading
-    # the body and return it.
-    #
-    # If the body is larger than MAX_BODY, a Tempfile object is used
-    # for the body, otherwise a StringIO is used.
-    #
-    def read_body(env, client, body, cl)
-      content_length = cl.to_i
+    # :nocov:
 
-      remain = content_length - body.bytesize
+    # Handle various error types thrown by Client I/O operations.
+    def client_error(e, client)
+      # Swallow, do not log
+      return if [ConnectionError, EOFError].include?(e.class)
 
-      return StringIO.new(body) if remain <= 0
-
-      # Use a Tempfile if there is a lot of data left
-      if remain > MAX_BODY
-        stream = Tempfile.new(Const::PUMA_TMP_BASE)
-        stream.binmode
+      lowlevel_error(e, client.env)
+      case e
+      when MiniSSL::SSLError
+        @events.ssl_error e, client.io
+      when HttpParserError
+        client.write_error(400)
+        @events.parse_error e, client
       else
-        # The body[0,0] trick is to get an empty string in the same
-        # encoding as body.
-        stream = StringIO.new body[0,0]
-      end
-
-      stream.write body
-
-      # Read an odd sized chunk so we can read even sized ones
-      # after this
-      chunk = client.readpartial(remain % CHUNK_SIZE)
-
-      # No chunk means a closed socket
-      unless chunk
-        stream.close
-        return nil
-      end
-
-      remain -= stream.write(chunk)
-
-      # Raed the rest of the chunks
-      while remain > 0
-        chunk = client.readpartial(CHUNK_SIZE)
-        unless chunk
-          stream.close
-          return nil
-        end
-
-        remain -= stream.write(chunk)
+        client.write_error(500)
+        @events.unknown_error e, nil, "Read"
       end
-
-      stream.rewind
-
-      return stream
     end
 
     # A fallback rack response if +@app+ raises as exception.
     #
-    def lowlevel_error(e, env)
+    def lowlevel_error(e, env, status=500)
       if handler = @options[:lowlevel_error_handler]
         if handler.arity == 1
           return handler.call(e)
-        else
+        elsif handler.arity == 2
           return handler.call(e, env)
+        else
+          return handler.call(e, env, status)
         end
       end
 
       if @leak_stack_on_error
-        [500, {}, ["Puma caught this error: #{e.message} (#{e.class})\n#{e.backtrace.join("\n")}"]]
+        backtrace = e.backtrace.nil? ? '<no backtrace available>' : e.backtrace.join("\n")
+        [status, {}, ["Puma caught this error: #{e.message} (#{e.class})\n#{backtrace}"]]
       else
-        [500, {}, ["An unhandled lowlevel error occurred. The application logs may have details.\n"]]
+        [status, {}, ["An unhandled lowlevel error occurred. The application logs may have details.\n"]]
       end
     end
 
@@ -968,31 +557,13 @@ module Puma
         $stdout.syswrite "#{pid}: === End thread backtrace dump ===\n"
       end
 
-      if @options[:drain_on_shutdown]
-        count = 0
-
-        while true
-          ios = IO.select @binder.ios, nil, nil, 0
-          break unless ios
-
-          ios.first.each do |sock|
-            begin
-              if io = sock.accept_nonblock
-                count += 1
-                client = Client.new io, @binder.env(sock)
-                @thread_pool << client
-              end
-            rescue SystemCallError
-            end
-          end
-        end
-
-        @events.debug "Drained #{count} additional connections."
+      if @status != :restart
+        @binder.close
       end
 
       if @thread_pool
         if timeout = @options[:force_shutdown_after]
-          @thread_pool.shutdown timeout.to_i
+          @thread_pool.shutdown timeout.to_f
         else
           @thread_pool.shutdown
         end
@@ -1000,18 +571,16 @@ module Puma
     end
 
     def notify_safely(message)
-      begin
-        @notify << message
-      rescue IOError
-         # The server, in another thread, is shutting down
+      @notify << message
+    rescue IOError, NoMethodError, Errno::EPIPE
+      # The server, in another thread, is shutting down
+      Thread.current.purge_interrupt_queue if Thread.current.respond_to? :purge_interrupt_queue
+    rescue RuntimeError => e
+      # Temporary workaround for https://bugs.ruby-lang.org/issues/13239
+      if e.message.include?('IOError')
         Thread.current.purge_interrupt_queue if Thread.current.respond_to? :purge_interrupt_queue
-      rescue RuntimeError => e
-        # Temporary workaround for https://bugs.ruby-lang.org/issues/13239
-        if e.message.include?('IOError')
-          Thread.current.purge_interrupt_queue if Thread.current.respond_to? :purge_interrupt_queue
-        else
-          raise e
-        end
+      else
+        raise e
       end
     end
     private :notify_safely
@@ -1029,44 +598,24 @@ module Puma
       @thread.join if @thread && sync
     end
 
-    def begin_restart
+    def begin_restart(sync=false)
       notify_safely(RESTART_COMMAND)
-    end
-
-    def fast_write(io, str)
-      n = 0
-      while true
-        begin
-          n = io.syswrite str
-        rescue Errno::EAGAIN, Errno::EWOULDBLOCK
-          if !IO.select(nil, [io], nil, WRITE_TIMEOUT)
-            raise ConnectionError, "Socket timeout writing data"
-          end
-
-          retry
-        rescue  Errno::EPIPE, SystemCallError, IOError
-          raise ConnectionError, "Socket timeout writing data"
-        end
-
-        return if n == str.bytesize
-        str = str.byteslice(n..-1)
-      end
-    end
-    private :fast_write
-
-    ThreadLocalKey = :puma_server
-
-    def self.current
-      Thread.current[ThreadLocalKey]
+      @thread.join if @thread && sync
     end
 
     def shutting_down?
       @status == :stop || @status == :restart
     end
 
-    def possible_header_injection?(header_value)
-      HTTP_INJECTION_REGEX =~ header_value.to_s
+    # List of methods invoked by #stats.
+    # @version 5.0.0
+    STAT_METHODS = [:backlog, :running, :pool_capacity, :max_threads, :requests_count].freeze
+
+    # Returns a hash of stats about the running server for reporting purposes.
+    # @version 5.0.0
+    # @!attribute [r] stats
+    def stats
+      STAT_METHODS.map {|name| [name, send(name) || 0]}.to_h
     end
-    private :possible_header_injection?
   end
 end