A framework to implement robust multiprocess servers like Unicorn
Ruby
Latest commit c9df44c Jul 10, 2018

README.md

ServerEngine

ServerEngine is a framework to implement robust multiprocess servers like Unicorn.

Main features:

                  Heartbeat via pipe
                      & auto-restart
                 /                \               ---+
+------------+  /   +----------+   \  +--------+     |
| Supervisor |------|  Server  |------| Worker |     |
+------------+      +----------+\     +--------+     | Multi-process
                        /         \                  | or multi-thread
                       /            \ +--------+     |
      Dynamic reconfiguration         | Worker |     |
     and live restart support         +--------+     |
                                                  ---+

ServerEngine also provides useful options and utilities such as logging, signal handlers, changing process names shown by ps command, chuser, stacktrace and heap dump on signal.

Examples

Simplest server

What you need to implement at least is a worker module which has run and stop methods.

require 'serverengine'

module MyWorker
  def run
    until @stop
      logger.info "Awesome work!"
      sleep 1
    end
  end

  def stop
    @stop = true
  end
end

se = ServerEngine.create(nil, MyWorker, {
  daemonize: true,
  log: 'myserver.log',
  pid_path: 'myserver.pid',
})
se.run

Send TERM signal (or KILL on Windows) to kill the daemon. See also Signals section bellow for details.

Multiprocess server

Simply set worker_type: "process" or worker_type: "thread" parameter, and set number of workers to workers parameter.

se = ServerEngine.create(nil, MyWorker, {
  daemonize: true,
  log: 'myserver.log',
  pid_path: 'myserver.pid',
  worker_type: 'process',
  workers: 4,
})
se.run

See also Worker types section bellow.

Multiprocess TCP server

One of the typical implementation styles of TCP servers is that a parent process listens socket and child processes accept connections from clients.

ServerEngine allows you to optionally implement a server module to control the parent process:

# Server module controls the parent process
module MyServer
  def before_run
    @sock = TCPServer.new(config[:bind], config[:port])
  end

  attr_reader :sock
end

# Worker module controls child processes
module MyWorker
  def run
    until @stop
      # you should use Cool.io or EventMachine actually
      c = server.sock.accept
      c.write "Awesome work!"
      c.close
    end
  end

  def stop
    @stop = true
  end
end

se = ServerEngine.create(MyServer, MyWorker, {
  daemonize: true,
  log: 'myserver.log',
  pid_path: 'myserver.pid',
  worker_type: 'process',
  workers: 4,
  bind: '0.0.0.0',
  port: 9071,
})
se.run

Multiprocess server on Windows and JRuby platforms

Above worker_type: "process" depends on fork system call, which doesn't work on Windows or JRuby platform. ServerEngine provides worker_type: "spawn" for those platforms (This is still EXPERIMENTAL). However, unfortunately, you need to implement different worker module because worker_type: "spawn" is not compatible with worker_type: "process" in terms of API.

What you need to implement at least to use worker_type: "spawn" is def spawn(process_manager) method. You will call process_manager.spawn at the method, where spawn is same with Process.spawn excepting return value.

In addition, Windows does not support signals. ServerEngine provides command_sender: "pipe" for Windows (and for other platforms, if you want to use it). When using command_sender: "pipe", the child process have to handle commands sent from parent process via STDIN. On Windows, command_sender: "pipe" is default.

You can call Server#stop(stop_graceful) and Server#restart(stop_graceful) instead of sending signals.

module MyWorker
  def spawn(process_manager)
    env = {
      'SERVER_ENGINE_CONFIG' => config.to_json
    }
    script = %[
      require 'serverengine'
      require 'json'

      conf = JSON.parse(ENV['SERVER_ENGINE_CONFIG'], symbolize_names: true)
      logger = ServerEngine::DaemonLogger.new(conf[:log] || STDOUT, conf)

      @stop = false
      command_pipe = STDIN.dup
      STDIN.reopen(File::NULL)
      Thread.new do
        until @stop
          case command_pipe.gets.chomp
          when "GRACEFUL_STOP"
            @stop = true
          when "IMMEDIATE_STOP"
            @stop = true
          when "GRACEFUL_RESTART", "IMMEDIATE_RESTART"
            # do something...
          end
        end
      end

      until @stop
        logger.info 'Awesome work!'
        sleep 1
      end
    ]
    process_manager.spawn(env, "ruby", "-e", script)
  end
end

se = ServerEngine.create(nil, MyWorker, {
  worker_type: 'spawn',
  command_sender: 'pipe',
  log: 'myserver.log',
})
se.run

Logging

ServerEngine logger rotates logs by 1MB and keeps 5 generations by default.

se = ServerEngine.create(MyServer, MyWorker, {
  log: 'myserver.log',
  log_level: 'debug',
  log_rotate_age: 5,
  log_rotate_size: 1*1024*1024,
})
se.run

ServerEngine's default logger extends from Ruby's standard Logger library to:

  • support multiprocess aware log rotation
  • support reopening of log file
  • support 'trace' level, which is lower level than 'debug'

See also Configuration section bellow.

Supervisor auto restart

Server programs running 24x7 hours need to survive even if a process stalled because of unexpected memory swapping or network errors.

Supervisor process runs as the parent process of the server process and monitor it to restart automatically. You can enable supervisor process by setting supervisor: true parameter:

se = ServerEngine.create(nil, MyWorker, {
  daemonize: true,
  pid_path: 'myserver.pid',
  supervisor: true,  # enables supervisor process
})
se.run

This auto restart reature will be suppressed for workers which exits with exit code specified by unrecoverable_exit_codes. At this case, whole process will exit without error (exit code 0).

Live restart

You can restart a server process without waiting for completion of all workers using INT signal (supervisor: true and enable_detach: true parameters must be enabled). This feature allows you to minimize downtime where workers take long time to complete a task.

# 1. starts server
+------------+    +----------+    +-----------+
| Supervisor |----|  Server  |----| Worker(s) |
+------------+    +----------+    +-----------+

# 2. receives SIGINT and waits for shutdown of the server for server_detach_wait
+------------+    +----------+    +-----------+
| Supervisor |    |  Server  |----| Worker(s) |
+------------+    +----------+    +-----------+

# 3. starts new server if the server doesn't exit in server_detach_wait time
+------------+    +----------+    +-----------+
| Supervisor |\   |  Server  |----| Worker(s) |
+------------+ |  +----------+    +-----------+
               |
               |  +----------+    +-----------+
               \--|  Server  |----| Worker(s) |
                  +----------+    +-----------+

# 4. old server exits eventually
+------------+
| Supervisor |\
+------------+ |
               |
               |  +----------+    +-----------+
               \--|  Server  |----| Worker(s) |
                  +----------+    +-----------+

Note that network servers (which listen sockets) shouldn't use live restart because it causes "Address already in use" error at the server process. Instead, simply use worker_type: "process" configuration and send USR1 to restart workers instead of the server. It restarts a worker without waiting for shutdown of the other workers. This way doesn't cause downtime because server process doesn't close listening sockets and keeps accepting new clients (See also restart_server_process parameter if necessary).

Dynamic config reloading

Robust servers should not restart only to update configuration parameters.

module MyWorker
  def initialize
    reload
  end

  def reload
    @message = config[:message] || "Awesome work!"
    @sleep = config[:sleep] || 1
  end

  def run
    until @stop
      logger.info @message
      sleep @sleep
    end
  end

  def stop
    @stop = true
  end
end

se = ServerEngine.create(nil, MyWorker) do
  YAML.load_file("config.yml").merge({
    daemonize: true,
    worker_type: 'process',
  })
end
se.run

Send USR2 signal to reload configuration file.

Signals

  • TERM: graceful shutdown
  • QUIT: immediate shutdown (available only when worker_type is "process")
  • USR1: graceful restart
  • HUP: immediate restart (available only when worker_type is "process")
  • USR2: reload config file and reopen log file
  • INT: detach process for live restarting (available only when supervisor and enable_detach parameters are true. otherwise graceful shutdown)
  • CONT: dump stacktrace and memory information to /tmp/sigdump-.log file. This can be disabled by including disable_sigdump: true in the configuration.

Immediate shutdown and restart send SIGQUIT signal to worker processes which kills the processes. Graceful shutdown and restart call Worker#stop method and wait for completion of Worker#run method.

Note that signals are not supported on Windows. You have to use piped command instead of signals on Windows. See also Multiprocess server on Windows and JRuby platforms section.

Utilities

BlockingFlag

ServerEngine::BlockingFlag is recommended to stop workers because stop method is called by a different thread from the run thread.

module MyWorker
  def initialize
    @stop_flag = ServerEngine::BlockingFlag.new
  end

  def run
    until @stop_flag.wait_for_set(1.0)  # or @stop_flag.set?
      logger.info @message
    end
  end

  def stop
    @stop_flag.set!
  end
end

se = ServerEngine.create(nil, MyWorker) do
  YAML.load_file(config).merge({
    daemonize: true,
    worker_type: 'process'
  })
end
se.run

SocketManager

ServerEngine::SocketManager is a powerful library to listen on the same port across multiple worker processes dynamically.

module MyServer
  def before_run
    @socket_manager_path = ServerEngine::SocketManager::Server.generate_path
    @socket_manager_server = ServerEngine::SocketManager::Server.open(@socket_manager_path)
  end

  def after_run
    @socket_manager_server.close
  end

  attr_reader :socket_manager_path
end

module MyWorker
  def initialize
    @stop_flag = ServerEngine::BlockingFlag.new
    @socket_manager = ServerEngine::SocketManager::Client.new(server.socket_manager_path)
  end

  def run
    lsock = @socket_manager.listen_tcp('0.0.0.0', 12345)
    until @stop
      c = lsock.accept
      c.write "Awesome work!"
      c.close
    end
  end

  def stop
    @stop = true
  end
end

se = ServerEngine.create(MyServer, MyWorker, {
  daemonize: true,
  log: 'myserver.log',
  pid_path: 'myserver.pid',
  worker_type: 'process',
  workers: 4,
  bind: '0.0.0.0',
  port: 9071,
})
se.run

See also examples.

Module API

Available methods are different depending on worker_type. ServerEngine supports 3 worker types:

  • embedded: uses a thread to run worker module (default). This type doesn't support immediate shutdown or immediate restart.
  • thread: uses threads to run worker modules. This type doesn't support immediate shutdown or immediate restart.
  • process: uses processes to run worker modules. This type doesn't work on Windows or JRuby platform.
  • spawn: uses processes to run worker modules. This type works on Windows and JRuby platform but available interface of worker module is limited (See also Worker module section).

Worker module

  • interface
    • initialize is called in the parent process (or thread) in contrast to the other methods
    • before_fork is called before fork for each worker process [worker_type = "thread", "process"]
    • run is the required method for worker_type = "embedded", "thread", "process"
    • spawn(process_manager) is the required method for worker_type = "spawn". Should call process_manager.spawn([env,] command... [,options]).
    • stop is called when TERM signal is received [worker_type = "embedded", "thread", "process"]
    • reload is called when USR2 signal is received [worker_type = "embedded", "thread", "process"]
    • after_start is called after starting the worker process in the parent process (or thread) [worker_type = "thread", "process", "spawn"]
  • api
    • server server instance
    • config configuration
    • logger logger
    • worker_id serial id of workers beginning from 0

Server module

  • interface
    • initialize is called in the parent process in contrast to the other methods
    • before_run is called before starting workers
    • after_run is called before shutting down
    • after_start is called after starting the server process in the parent process (available if supervisor parameter is true)
  • hook points (call super in these methods)
    • reload_config
    • stop(stop_graceful)
    • restart(stop_graceful)
  • api
    • config configuration
    • logger logger

List of all configurations

  • Daemon
    • daemonize enables daemonize (default: false)
    • pid_path sets the path to pid file (default: don't create pid file)
    • supervisor enables supervisor if it's true (default: false)
    • disable_sigdump disables the handling of the SIGCONT signal and dumping of the thread (default: false)
    • daemon_process_name changes process name ($0) of server or supervisor process
    • chuser changes execution user
    • chgroup changes execution group
    • chumask changes umask
    • daemonize_error_exit_code exit code when daemonize, changing user or changing group fails (default: 1)
  • Supervisor: available only when supervisor parameters is true
    • server_process_name changes process name ($0) of server process
    • restart_server_process restarts server process when it receives USR1 or HUP signal. (default: false)
    • enable_detach enables INT signal (default: true)
    • exit_on_detach exits supervisor after detaching server process instead of restarting it (default: false)
    • disable_reload disables USR2 signal (default: false)
    • server_restart_wait sets wait time before restarting server after last restarting (default: 1.0) [dynamic reloadable]
    • server_detach_wait sets wait time before starting live restart (default: 10.0) [dynamic reloadable]
  • Multithread server and multiprocess server: available only when worker_type is thread or process
    • workers sets number of workers (default: 1) [dynamic reloadable]
    • start_worker_delay sets wait time before starting a new worker (default: 0) [dynamic reloadable]
    • start_worker_delay_rand randomizes start_worker_delay at this ratio (default: 0.2) [dynamic reloadable]
  • Multiprocess server: available only when worker_type is "process"
    • worker_process_name changes process name ($0) of workers [dynamic reloadable]
    • worker_heartbeat_interval sets interval of heartbeats in seconds (default: 1.0) [dynamic reloadable]
    • worker_heartbeat_timeout sets timeout of heartbeat in seconds (default: 180) [dynamic reloadable]
    • worker_graceful_kill_interval sets the first interval of TERM signals in seconds (default: 15) [dynamic reloadable]
    • worker_graceful_kill_interval_increment sets increment of TERM signal interval in seconds (default: 10) [dynamic reloadable]
    • worker_graceful_kill_timeout sets promotion timeout from TERM to QUIT signal in seconds. -1 means no timeout (default: 600) [dynamic reloadable]
    • worker_immediate_kill_interval sets the first interval of QUIT signals in seconds (default: 10) [dynamic reloadable]
    • worker_immediate_kill_interval_increment sets increment of QUIT signal interval in seconds (default: 10) [dynamic reloadable]
    • worker_immediate_kill_timeout sets promotion timeout from QUIT to KILL signal in seconds. -1 means no timeout (default: 600) [dynamic reloadable]
    • unrecoverable_exit_codes handles worker processes, which exits with code included in this value, as unrecoverable (not to restart) (default: [])
    • stop_immediately_at_unrecoverable_exit stops server immediately if a worker exits with unrecoverable exit status (default: false)
  • Multiprocess spawn server: available only when worker_type is "spawn"
    • all parameters of multiprocess server excepting worker_process_name
    • worker_reload_signal sets the signal to notice configuration reload to a spawned process. Set nil to disable (default: nil)
  • Logger
    • log sets path to log file. Set "-" for STDOUT (default: STDERR) [dynamic reloadable]
    • log_level log level: trace, debug, info, warn, error or fatal. (default: debug) [dynamic reloadable]
    • log_rotate_age generations to keep rotated log files (default: 5), 0 will disable log rotate.
    • log_rotate_size sets the size to rotate log files (default: 1048576), when log_rotate_age is 0, the log_rotate_size value will be ignored.
    • log_stdout hooks STDOUT to log file (default: true)
    • log_stderr hooks STDERR to log file (default: true)
    • logger_class class of the logger instance (default: ServerEngine::DaemonLogger)

Author:    Sadayuki Furuhashi
Copyright: Copyright (c) 2012-2013 Sadayuki Furuhashi
License:   Apache License, Version 2.0