Experiment: Fork from existing worker process to optimize copy-on-write performance

History.md

@@ -10,6 +10,7 @@
   * Increases maximum URI path length from 2048 to 8196 bytes (#2167)
   * Force shutdown responses can be overridden by using the `lowlevel_error_handler` config (#2203)
   * Faster phased restart and worker timeout (#2121)
+  * Add `fork_worker` option and `refork` command for improved copy-on-write performance (#2099)
 
 * Deprecations, Removals and Breaking API Changes
   * `Puma.stats` now returns a Hash instead of a JSON string (#2086)

docs/fork_worker.md

@@ -0,0 +1,31 @@
+# Fork-Worker Cluster Mode [Experimental]
+
+Puma 5 introduces an experimental new cluster-mode configuration option, `fork_worker` (`--fork-worker` from the CLI). This mode causes Puma to fork additional workers from worker 0, instead of directly from the master process:
+
+```
+10000   \_ puma 4.3.3 (tcp://0.0.0.0:9292) [puma]
+10001       \_ puma: cluster worker 0: 10000 [puma]
+10002           \_ puma: cluster worker 1: 10000 [puma]
+10003           \_ puma: cluster worker 2: 10000 [puma]
+10004           \_ puma: cluster worker 3: 10000 [puma]
+```
+
+Similar to the `preload_app!` option, the `fork_worker` option allows your application to be initialized only once for copy-on-write memory savings, and it has two additional advantages:
+
+1. **Compatible with phased restart.** Because the master process itself doesn't preload the application, this mode works with phased restart (`SIGUSR1` or `pumactl phased-restart`). When worker 0 reloads as part of a phased restart, it initializes a new copy of your application first, then the other workers reload by forking from this new worker already containing the new preloaded application.
+
+   This allows a phased restart to complete as quickly as a hot restart (`SIGUSR2` or `pumactl restart`), while still minimizing downtime by staggering the restart across cluster workers.
+
+2. **'Refork' for additional copy-on-write improvements in running applications.** Fork-worker mode introduces a new `refork` command that re-loads all nonzero workers by re-forking them from worker 0.
+
+   This command can potentially improve memory utilization in large or complex applications that don't fully pre-initialize on startup, because the re-forked workers can share copy-on-write memory with a worker that has been running for a while and serving requests.
+
+   You can trigger a refork by sending the cluster the `SIGURG` signal or running the `pumactl refork` command at any time. A refork will also automatically trigger once, after a certain number of requests have been processed by worker 0 (default 1000). To configure the number of requests before the auto-refork, pass a positive integer argument to `fork_workers` (e.g., `fork_workers 1000`), or `0` to disable.
+
+### Limitations
+
+- This mode is still very experimental so there may be bugs or edge-cases, particularly around expected behavior of existing hooks. Please open a [bug report](https://github.com/puma/puma/issues/new?template=bug_report.md) if you encounter any issues.
+
+- In order to fork new workers cleanly, worker 0 shuts down its server and stops serving requests so there are no open file descriptors or other kinds of shared global state between processes, and to maximize copy-on-write efficiency across the newly-forked workers. This may temporarily reduce total capacity of the cluster during a phased restart / refork.
+
+  In a cluster with `n` workers, a normal phased restart stops and restarts workers one by one while the application is loaded in each process, so `n-1` workers are available serving requests during the restart. In a phased restart in fork-worker mode, the application is first loaded in worker 0 while `n-1` workers are available, then worker 0 remains stopped while the rest of the workers are reloaded one by one, leaving only `n-2` workers to be available for a brief period of time. Reloading the rest of the workers should be quick because the application is preloaded at that point, but there may be situations where it can take longer (slow clients, long-running application code, slow worker-fork hooks, etc).

docs/signals.md

@@ -41,6 +41,7 @@ Puma cluster responds to these signals:
 - `HUP`  reopen log files defined in stdout_redirect configuration parameter. If there is no stdout_redirect option provided it will behave like `INT`
 - `INT` equivalent of sending Ctrl-C to cluster. Will attempt to finish then exit.
 - `CHLD`
+- `URG` refork workers in phases from worker 0, if `fork_workers` option is enabled.
 
 ## Callbacks order in case of different signals
 

lib/puma/app/status.rb

@@ -63,6 +63,11 @@ def call(env)
           end
 
           rack_response(200, backtraces.to_json)
+
+        when /\/refork$/
+          Process.kill "SIGURG", $$
+          rack_response(200, OK_STATUS)
+
         else
           rack_response 404, "Unsupported action", 'text/plain'
         end

lib/puma/cli.rb

@@ -130,6 +130,12 @@ def setup_options
             user_config.environment arg
           end
 
+          o.on "-f", "--fork-worker=[REQUESTS]", OptionParser::DecimalInteger,
+            "Fork new workers from existing worker. Cluster mode only",
+            "Auto-refork after REQUESTS (default 1000)" do |*args|
+            user_config.fork_worker *args.compact
+          end
+
           o.on "-I", "--include PATH", "Specify $LOAD_PATH directories" do |arg|
             $LOAD_PATH.unshift(*arg.split(':'))
           end

lib/puma/cluster.rb

@@ -37,7 +37,7 @@ def stop_workers
       begin
         loop do
           wait_workers
-          break if @workers.empty?
+          break if @workers.reject {|w| w.pid.nil?}.empty?
           sleep 0.2
         end
       rescue Interrupt
@@ -78,6 +78,7 @@ def initialize(idx, pid, phase, options)
       end
 
       attr_reader :index, :pid, :phase, :signal, :last_checkin, :last_status, :started_at
+      attr_writer :pid, :phase
 
       def booted?
         @stage == :booted
@@ -113,7 +114,7 @@ def term
             @term ||= true
             @first_term_sent ||= Time.now
           end
-          Process.kill @signal, @pid
+          Process.kill @signal, @pid if @pid
         rescue Errno::ESRCH
         end
       end
@@ -134,23 +135,43 @@ def spawn_workers
       return if diff < 1
 
       master = Process.pid
+      if @options[:fork_worker]
+        @fork_writer << "-1\n"
+      end
 
       diff.times do
         idx = next_worker_index
-        @launcher.config.run_hooks :before_worker_fork, idx, @launcher.events
 
-        pid = fork { worker(idx, master) }
-        if !pid
-          log "! Complete inability to spawn new workers detected"
-          log "! Seppuku is the only choice."
-          exit! 1
+        if @options[:fork_worker] && idx != 0
+          @fork_writer << "#{idx}\n"
+          pid = nil
+        else
+          pid = spawn_worker(idx, master)
         end
 
         debug "Spawned worker: #{pid}"
         @workers << Worker.new(idx, pid, @phase, @options)
+      end
 
-        @launcher.config.run_hooks :after_worker_fork, idx, @launcher.events
+      if @options[:fork_worker] &&
+        @workers.all? {|x| x.phase == @phase}
+
+        @fork_writer << "0\n"
+      end
+    end
+
+    def spawn_worker(idx, master)
+      @launcher.config.run_hooks :before_worker_fork, idx, @launcher.events
+
+      pid = fork { worker(idx, master) }
+      if !pid
+        log "! Complete inability to spawn new workers detected"
+        log "! Seppuku is the only choice."
+        exit! 1
       end
+
+      @launcher.config.run_hooks :after_worker_fork, idx, @launcher.events
+      pid
     end
 
     def cull_workers
@@ -213,7 +234,6 @@ def check_workers
 
     def wakeup!
       return unless @wakeup
-      @next_check = Time.now
 
       begin
         @wakeup.write "!" unless @wakeup.closed?
@@ -229,9 +249,14 @@ def worker(index, master)
 
       Signal.trap "SIGINT", "IGNORE"
 
+      fork_worker = @options[:fork_worker] && index == 0
+
       @workers = []
-      @master_read.close
-      @suicide_pipe.close
+      if !@options[:fork_worker] || fork_worker
+        @master_read.close
+        @suicide_pipe.close
+        @fork_writer.close
+      end
 
       Thread.new do
         Puma.set_thread_name "worker check pipe"
@@ -254,15 +279,45 @@ def worker(index, master)
       # things in shape before booting the app.
       @launcher.config.run_hooks :before_worker_boot, index, @launcher.events
 
-      server = start_server
+      server = @server ||= start_server
+      restart_server = Queue.new << true << false
+
+      if fork_worker
+        restart_server.clear
+        Signal.trap "SIGCHLD" do
+          Process.wait(-1, Process::WNOHANG) rescue nil
+          wakeup!
+        end
+
+        Thread.new do
+          Puma.set_thread_name "worker fork pipe"
+          while (idx = @fork_pipe.gets)
+            idx = idx.to_i
+            if idx == -1 # stop server
+              if restart_server.length > 0
+                restart_server.clear
+                server.begin_restart(true)
+                @launcher.config.run_hooks :before_refork, nil, @launcher.events
+                GC.compact if GC.respond_to?(:compact)
+              end
+            elsif idx == 0 # restart server
+              restart_server << true << false
+            else # fork worker
+              pid = spawn_worker(idx, master)
+              @worker_write << "f#{pid}:#{idx}\n" rescue nil
+            end
+          end
+        end
+      end
 
       Signal.trap "SIGTERM" do
         @worker_write << "e#{Process.pid}\n" rescue nil
         server.stop
+        restart_server << false
       end
 
       begin
-        @worker_write << "b#{Process.pid}\n"
+        @worker_write << "b#{Process.pid}:#{index}\n"
       rescue SystemCallError, IOError
         Thread.current.purge_interrupt_queue if Thread.current.respond_to? :purge_interrupt_queue
         STDERR.puts "Master seems to have exited, exiting."
@@ -283,7 +338,7 @@ def worker(index, master)
         end
       end
 
-      server.run.join
+      server.run.join while restart_server.pop
 
       # Invoke any worker shutdown hooks so they can prevent the worker
       # exiting until any background operations are completed
@@ -360,9 +415,31 @@ def preload?
       @options[:preload_app]
     end
 
+    def fork_worker!
+      if (worker = @workers.find { |w| w.index == 0 })
+        worker.phase += 1
+      end
+      phased_restart
+    end
+
     # We do this in a separate method to keep the lambda scope
     # of the signals handlers as small as possible.
     def setup_signals
+      if @options[:fork_worker]
+        Signal.trap "SIGURG" do
+          fork_worker!
+        end
+
+        # Auto-fork after the specified number of requests.
+        if (fork_requests = @options[:fork_worker].to_i) > 0
+          @launcher.events.register(:ping!) do |w|
+            fork_worker! if w.index == 0 &&
+              w.phase == 0 &&
+              w.last_status[:requests_count] >= fork_requests
+          end
+        end
+      end
+
       Signal.trap "SIGCHLD" do
         wakeup!
       end
@@ -446,6 +523,10 @@ def run
       #
       @check_pipe, @suicide_pipe = Puma::Util.pipe
 
+      # Separate pipe used by worker 0 to receive commands to
+      # fork new worker processes.
+      @fork_pipe, @fork_writer = Puma::Util.pipe
+
       log "Use Ctrl-C to stop"
 
       redirect_io
@@ -484,11 +565,18 @@ def run
             if res
               req = read.read_nonblock(1)
 
+              @next_check = Time.now if req == "!"
               next if !req || req == "!"
 
               result = read.gets
               pid = result.to_i
 
+              if req == "b" || req == "f"
+                pid, idx = result.split(':').map(&:to_i)
+                w = @workers.find {|x| x.index == idx}
+                w.pid = pid if w.pid.nil?
+              end
+
               if w = @workers.find { |x| x.pid == pid }
                 case req
                 when "b"
@@ -502,6 +590,7 @@ def run
                   w.term unless w.term?
                 when "p"
                   w.ping!(result.sub(/^\d+/,'').chomp)
+                  @launcher.events.fire(:ping!, w)
                 end
               else
                 log "! Out-of-sync worker list, no #{pid} worker"
@@ -528,6 +617,7 @@ def run
     # `#term` if needed
     def wait_workers
       @workers.reject! do |w|
+        next false if w.pid.nil?
         begin
           if Process.wait(w.pid, Process::WNOHANG)
             true
@@ -536,7 +626,12 @@ def wait_workers
             nil
           end
         rescue Errno::ECHILD
-          true # child is already terminated
+          begin
+            Process.kill(0, w.pid)
+            false # child still alive, but has another parent
+          rescue Errno::ESRCH, Errno::EPERM
+            true # child is already terminated
+          end
         end
       end
     end

lib/puma/control_cli.rb

@@ -11,7 +11,7 @@
 module Puma
   class ControlCLI
 
-    COMMANDS = %w{halt restart phased-restart start stats status stop reload-worker-directory gc gc-stats thread-backtraces}
+    COMMANDS = %w{halt restart phased-restart start stats status stop reload-worker-directory gc gc-stats thread-backtraces refork}
     PRINTABLE_COMMANDS = %w{gc-stats stats thread-backtraces}
 
     def initialize(argv, stdout=STDOUT, stderr=STDERR)
@@ -239,6 +239,9 @@ def send_signal
 
           return
 
+        when "refork"
+          Process.kill "SIGURG", @pid
+
         else
           return
         end

lib/puma/dsl.rb

@@ -492,6 +492,28 @@ def after_worker_fork(&block)
 
     alias_method :after_worker_boot, :after_worker_fork
 
+    # When `fork_worker` is enabled, code to run in Worker 0
+    # before all other workers are re-forked from this process,
+    # after the server has temporarily stopped serving requests
+    # (once per complete refork cycle).
+    #
+    # This can be used to trigger extra garbage-collection to maximize
+    # copy-on-write efficiency, or close any connections to remote servers
+    # (database, Redis, ...) that were opened while the server was running.
+    #
+    # This can be called multiple times to add several hooks.
+    #
+    # @note Cluster mode with `fork_worker` enabled only.
+    # @example
+    #   on_refork do
+    #     3.times {GC.start}
+    #   end
+
+    def on_refork(&block)
+      @options[:before_refork] ||= []
+      @options[:before_refork] << block
+    end
+
     # Code to run out-of-band when the worker is idle.
     # These hooks run immediately after a request has finished
     # processing and there are no busy threads on the worker.
@@ -705,5 +727,19 @@ def set_remote_address(val=:socket)
       end
     end
 
+    # When enabled, workers will be forked from worker 0 instead of from the master process.
+    # This option is similar to `preload_app` because the app is preloaded before forking,
+    # but it is compatible with phased restart.
+    #
+    # This option also enables the `refork` command (SIGURG), which optimizes copy-on-write performance
+    # in a running app.
+    #
+    # A refork will automatically trigger once after the specified number of requests
+    # (default 1000), or pass 0 to disable auto refork.
+    #
+    # @note Cluster mode only.
+    def fork_worker(after_requests=1000)
+      @options[:fork_worker] = Integer(after_requests)
+    end
   end
 end