Add worker_check_interval configuration option (puma#2759)

Co-authored-by: Olivier Bellone <olivier@bellone.fr>

lib/puma/cluster.rb

@@ -135,7 +135,7 @@ def all_workers_booted?
     def check_workers
       return if @next_check >= Time.now
 
-      @next_check = Time.now + Const::WORKER_CHECK_INTERVAL
+      @next_check = Time.now + @options[:worker_check_interval]
 
       timeout_workers
       wait_workers

lib/puma/cluster/worker.rb

@@ -130,7 +130,7 @@ def run
                 Puma::Util.purge_interrupt_queue
                 break
               end
-              sleep Const::WORKER_CHECK_INTERVAL
+              sleep @options[:worker_check_interval]
             end
           end
           server_thread.join

lib/puma/configuration.rb

@@ -11,6 +11,7 @@ module ConfigDefault
 
     DefaultTCPHost = "0.0.0.0"
     DefaultTCPPort = 9292
+    DefaultWorkerCheckInterval = 5
     DefaultWorkerTimeout = 60
     DefaultWorkerShutdownTimeout = 30
   end
@@ -195,6 +196,7 @@ def puma_default_options
         :workers => Integer(ENV['WEB_CONCURRENCY'] || 0),
         :silence_single_worker_warning => false,
         :mode => :http,
+        :worker_check_interval => DefaultWorkerCheckInterval,
         :worker_timeout => DefaultWorkerTimeout,
         :worker_boot_timeout => DefaultWorkerTimeout,
         :worker_shutdown_timeout => DefaultWorkerShutdownTimeout,

lib/puma/const.rb

@@ -235,9 +235,6 @@ module Const
 
     EARLY_HINTS = "rack.early_hints".freeze
 
-    # Minimum interval to checks worker health
-    WORKER_CHECK_INTERVAL = 5
-
     # Illegal character in the key or value of response header
     DQUOTE = "\"".freeze
     HTTP_HEADER_DELIMITER = Regexp.escape("(),/:;<=>?@[]{}\\").freeze

lib/puma/dsl.rb

@@ -736,6 +736,19 @@ def tag(string)
       @options[:tag] = string.to_s
     end
 
+    # Change the default interval for checking workers.
+    #
+    # The default value is 5 seconds.
+    #
+    # @note Cluster mode only.
+    # @example
+    #   worker_check_interval 5
+    # @see Puma::Cluster#check_workers
+    #
+    def worker_check_interval(interval)
+      @options[:worker_check_interval] = Integer(interval)
+    end
+
     # Verifies that all workers have checked in to the master process within
     # the given timeout. If not the worker process will be restarted. This is
     # not a request timeout, it is to protect against a hung or dead process.
@@ -750,7 +763,7 @@ def tag(string)
     #
     def worker_timeout(timeout)
       timeout = Integer(timeout)
-      min = Const::WORKER_CHECK_INTERVAL
+      min = @options.fetch(:worker_check_interval, Puma::ConfigDefault::DefaultWorkerCheckInterval)
 
       if timeout <= min
         raise "The minimum worker_timeout must be greater than the worker reporting interval (#{min})"

test/helpers/integration.rb

@@ -1,6 +1,7 @@
 # frozen_string_literal: true
 
 require "puma/control_cli"
+require "json"
 require "open3"
 require "io/wait"
 require_relative 'tmp_path'
@@ -255,6 +256,11 @@ def cli_pumactl(argv, unix: false)
     r
   end
 
+  def get_stats
+    read_pipe = cli_pumactl "stats"
+    JSON.parse(read_pipe.readlines.last)
+  end
+
   def hot_restart_does_not_drop_connections(num_threads: 1, total_requests: 500)
     skipped = true
     skip_if :jruby, suffix: <<-MSG

test/test_integration_cluster.rb

@@ -1,6 +1,8 @@
 require_relative "helper"
 require_relative "helpers/integration"
 
+require "time"
+
 class TestIntegrationCluster < TestIntegration
   parallelize_me! if ::Puma.mri?
 
@@ -147,14 +149,29 @@ def test_stuck_phased_restart
     worker_respawn { |phase0_worker_pids| Process.kill :USR1, @pid }
   end
 
+  def test_worker_check_interval
+    @control_tcp_port = UniquePort.call
+    worker_check_interval = 1
+
+    cli_server "-w 1 -t 1:1 --control-url tcp://#{HOST}:#{@control_tcp_port} --control-token #{TOKEN} test/rackup/hello.ru", config: "worker_check_interval #{worker_check_interval}"
+
+    sleep worker_check_interval + 1
+    last_checkin_1 = Time.parse(get_stats["worker_status"].first["last_checkin"])
+
+    sleep worker_check_interval + 1
+    last_checkin_2 = Time.parse(get_stats["worker_status"].first["last_checkin"])
+
+    assert(last_checkin_2 > last_checkin_1)
+  end
+
   def test_worker_boot_timeout
     timeout = 1
     worker_timeout(timeout, 2, "worker failed to boot within \\\d+ seconds", "worker_boot_timeout #{timeout}; on_worker_boot { sleep #{timeout + 1} }")
   end
 
   def test_worker_timeout
     skip 'Thread#name not available' unless Thread.current.respond_to?(:name)
-    timeout = Puma::Const::WORKER_CHECK_INTERVAL + 1
+    timeout = Puma::ConfigDefault::DefaultWorkerCheckInterval + 1
     worker_timeout(timeout, 1, "worker failed to check in within \\\d+ seconds", <<RUBY)
 worker_timeout #{timeout}
 on_worker_boot do

test/test_launcher.rb

@@ -148,7 +148,7 @@ def test_puma_stats_clustered
     end
     launcher = launcher(conf)
     Thread.new do
-      sleep Puma::Const::WORKER_CHECK_INTERVAL + 1
+      sleep Puma::ConfigDefault::DefaultWorkerCheckInterval + 1
       status = Puma.stats_hash[:worker_status].first[:last_status]
       Puma::Server::STAT_METHODS.each do |stat|
         assert_includes status, stat