Module EventMachine
In: lib/eventmachine.rb

$Id: callback.rb 321 2007-05-22 22:15:04Z blackhedd $

Author:Francis Cianfrocca (gmail: blackhedd)
Date:19 May 2006

See EventMachine and EventMachine::Connection for documentation and usage examples.


Classes and Modules

Module EventMachine::Deferrable
Module EventMachine::Eventable
Module EventMachine::Protocols
Module EventMachine::UuidGenerator
Class EventMachine::Connection
Class EventMachine::DatagramObject
Class EventMachine::DefaultDeferrable
Class EventMachine::EM
Class EventMachine::Error
Class EventMachine::EvmaTCPClient
Class EventMachine::EvmaTCPServer
Class EventMachine::EvmaUDPSocket
Class EventMachine::FileStreamer
Class EventMachine::LoopbreakReader
Class EventMachine::PeriodicTimer
Class EventMachine::Reactor
Class EventMachine::Selectable
Class EventMachine::StreamObject
Class EventMachine::Timer


TimerFired = 100
ConnectionData = 101
ConnectionUnbound = 102
ConnectionAccepted = 103
ConnectionCompleted = 104
LoopbreakSignalled = 105
VERSION = "0.8.0"
TimerFired = 100   TODO: These event numbers are defined in way too many places. DRY them up.
ConnectionData = 101
ConnectionUnbound = 102
ConnectionAccepted = 103
ConnectionCompleted = 104
LoopbreakSignalled = 105

Public Class methods

EventMachine#add_periodic_timer adds a periodic timer to the event loop. It takes the same parameters as the one-shot timer method, EventMachine#add_timer. This method schedules execution of the given block repeatedly, at intervals of time at least as great as the number of seconds given in the first parameter to the call.

Usage example

The following sample program will write a dollar-sign to stderr every five seconds. (Of course if the program defined network clients and/or servers, they would be doing their work while the periodic timer is counting off.)

 EventMachine::run {
   EventMachine::add_periodic_timer( 5 ) { $stderr.write "$" }

EventMachine#add_timer adds a one-shot timer to the event loop. Call it with one or two parameters. The first parameters is a delay-time expressed in seconds (not milliseconds). The second parameter, if present, must be a proc object. If a proc object is not given, then you can also simply pass a block to the method call.

EventMachine#add_timer may be called from the block passed to EventMachine#run or from any callback method. It schedules execution of the proc or block passed to add_timer, after the passage of an interval of time equal to at least the number of seconds specified in the first parameter to the call.

EventMachine#add_timer is a non-blocking call. Callbacks can and will be called during the interval of time that the timer is in effect. There is no built-in limit to the number of timers that can be outstanding at any given time.

Usage example

This example shows how easy timers are to use. Observe that two timers are initiated simultaneously. Also, notice that the event loop will continue to run even after the second timer event is processed, since there was no call to EventMachine#stop_event_loop. There will be no activity, of course, since no network clients or servers are defined. Stop the program with Ctrl-C.

 require 'rubygems'
 require 'eventmachine'

 EventMachine::run {
   puts "Starting the run now: #{}"
   EventMachine::add_timer 5, proc { puts "Executing timer event: #{}" }
   EventMachine::add_timer( 10 ) { puts "Executing timer event: #{}" }

close_connection The extension version does NOT raise any kind of an error if an attempt is made to close a non-existent connection. Not sure whether we should. For now, we‘ll raise an error here in that case.

EventMachine#connect initiates a TCP connection to a remote server and sets up event-handling for the connection. You can call EventMachine#connect in the block supplied to EventMachine#run or in any callback method.

EventMachine#connect takes the IP address (or hostname) and port of the remote server you want to connect to. It also takes an optional handler Module which you must define, that contains the callbacks that will be invoked by the event loop on behalf of the connection.

See the description of EventMachine#start_server for a discussion of the handler Module. All of the details given in that description apply for connections created with EventMachine#connect.

Usage Example

Here‘s a program which connects to a web server, sends a naive request, parses the HTTP header of the response, and then (antisocially) ends the event loop, which automatically drops the connection (and incidentally calls the connection‘s unbind method).

 require 'rubygems'
 require 'eventmachine'

 module DumbHttpClient

   def post_init
     send_data "GET / HTTP/1.1\r\nHost: _\r\n\r\n"
     @data = ""

   def receive_data data
     @data << data
     if  @data =~ /[\n][\r]*[\n]/m
       $`.each {|line| puts ">>> #{line}" }

       puts "Now we'll terminate the loop, which will also close the connection"

   def unbind
     puts "A connection has terminated"

 end # DumbHttpClient

 EventMachine::run {
   EventMachine::connect "", 80, DumbHttpClient
 puts "The event loop has ended"

There are times when it‘s more convenient to define a protocol handler as a Class rather than a Module. Here‘s how to do this:

 class MyProtocolHandler < EventMachine::Connection
   def initialize *args
     # whatever else you want to do here

   #.......your other class code
 end # class MyProtocolHandler

If you do this, then an instance of your class will be instantiated to handle every network connection created by your code or accepted by servers that you create. If you redefine post_init in your protocol-handler class, your post_init method will be called inside the call to super that you will make in your initialize method (if you provide one).

connect_server. Return a connection descriptor to the caller. TODO, what do we return here if we can‘t connect?

Make a connection to a Unix-domain socket. This is not implemented on Windows platforms. The parameter socketname is a String which identifies the Unix-domain socket you want to connect to. socketname is the name of a file on your local system, and in most cases is a fully-qualified path name. Make sure that your process has enough local permissions to open the Unix-domain socket. See also the documentation for connect_server. This method behaves like connect_server in all respects except for the fact that it connects to a local Unix-domain socket rather than a TCP socket. NOTE: this functionality will soon be subsumed into the connect method. This method will still be supported as an alias.

defer is for integrating blocking operations into EventMachine‘s control flow. Call defer with one or two blocks, as shown below (the second block is optional):

 operation = proc {
   # perform a long-running operation here, such as a database query.
   "result" # as usual, the last expression evaluated in the block will be the return value.
 callback = proc {|result|
   # do something with result here, such as send it back to a network client.

 EventMachine.defer( operation, callback )

The action of defer is to take the block specified in the first parameter (the "operation") and schedule it for asynchronous execution on an internal thread pool maintained by EventMachine. When the operation completes, it will pass the result computed by the block (if any) back to the EventMachine reactor. Then, EventMachine calls the block specified in the second parameter to defer (the "callback"), as part of its normal, synchronous event handling loop. The result computed by the operation block is passed as a parameter to the callback. You may omit the callback parameter if you don‘t need to execute any code after the operation completes.

Caveats: This is a provisional implementation and is subject to change. Note carefully that the code in your deferred operation will be executed on a separate thread from the main EventMachine processing and all other Ruby threads that may exist in your program. Also, multiple deferred operations may be running at once! Therefore, you are responsible for ensuring that your operation code is threadsafe. [Need more explanation and examples.] Don‘t write a deferred operation that will block forever. If so, the current implementation will not detect the problem, and the thread will never be returned to the pool. EventMachine limits the number of threads in its pool, so if you do this enough times, your subsequent deferred operations won‘t get a chance to run. [We might put in a timer to detect this problem.]

This is mostly useful for automated tests. Return a distinctive symbol so the caller knows whether he‘s dealing with an extension or with a pure-Ruby library.

Schedules a proc for execution immediately after the next "turn" through the reactor core. An advanced technique, this can be useful for improving memory management and/or application responsiveness, especially when scheduling large amounts of data for writing to a network connection. TODO, we need a FAQ entry on this subject.

next_tick takes either a single argument (which must be a Proc) or a block. And I‘m taking suggestions for a better name for this method.

EventMachine#open_datagram_socket is for support of UDP-based protocols. Its usage is similar to that of EventMachine#start_server. It takes three parameters: an IP address (which must be valid on the machine which executes the method), a port number, and an optional Module name which will handle the data. This method will create a new UDP (datagram) socket and bind it to the address and port that you specify. The normal callbacks (see EventMachine#start_server) will be called as events of interest occur on the newly-created socket, but there are some differences in how they behave.

Connection#receive_data will be called when a datagram packet is received on the socket, but unlike TCP sockets, the message boundaries of the received data will be respected. In other words, if the remote peer sent you a datagram of a particular size, you may rely on Connection#receive_data to give you the exact data in the packet, with the original data length. Also observe that Connection#receive_data may be called with a zero-length data payload, since empty datagrams are permitted in UDP.

Connection#send_data is available with UDP packets as with TCP, but there is an important difference. Because UDP communications are connectionless, there is no implicit recipient for the packets you send. Ordinarily you must specify the recipient for each packet you send. However, EventMachine provides for the typical pattern of receiving a UDP datagram from a remote peer, performing some operation, and then sending one or more packets in response to the same remote peer. To support this model easily, just use Connection#send_data in the code that you supply for Connection:receive_data. EventMachine will provide an implicit return address for any messages sent to Connection#send_data within the context of a Connection#receive_data callback, and your response will automatically go to the correct remote peer. (TODO: Example-code needed!)

Observe that the port number that you supply to EventMachine#open_datagram_socket may be zero. In this case, EventMachine will create a UDP socket that is bound to an ephemeral (not well-known) port. This is not appropriate for servers that must publish a well-known port to which remote peers may send datagrams. But it can be useful for clients that send datagrams to other servers. If you do this, you will receive any responses from the remote servers through the normal Connection#receive_data callback. Observe that you will probably have issues with firewalls blocking the ephemeral port numbers, so this technique is most appropriate for LANs. (TODO: Need an example!)

If you wish to send datagrams to arbitrary remote peers (not necessarily ones that have sent data to which you are responding), then see Connection#send_datagram.

DO NOT call send_data from a datagram socket outside of a receive_data method. Use send_datagram. If you do use send_data outside of a receive_data method, you‘ll get a confusing error because there is no "peer," as send_data requires. (Inside of receive_data, send_data "fakes" the peer as described above.)

TODO, must document popen. At this moment, it‘s only available on Unix. This limitation is expected to go away.

EventMachine::run initializes and runs an event loop. This method only returns if user-callback code calls stop_event_loop. Use the supplied block to define your clients and servers. The block is called by EventMachine::run immediately after initializing its internal event loop but before running the loop. Therefore this block is the right place to call start_server if you want to accept connections from remote clients.

For programs that are structured as servers, it‘s usually appropriate to start an event loop by calling EventMachine::run, and let it run forever. It‘s also possible to use EventMachine::run to make a single client-connection to a remote server, process the data flow from that single connection, and then call stop_event_loop to force EventMachine::run to return. Your program will then continue from the point immediately following the call to EventMachine::run.

You can of course do both client and servers simultaneously in the same program. One of the strengths of the event-driven programming model is that the handling of network events on many different connections will be interleaved, and scheduled according to the actual events themselves. This maximizes efficiency.

Server usage example

See the text at the top of this file for an example of an echo server.

Client usage example

See the description of stop_event_loop for an extremely simple client example.

send_datagram. This is currently only for UDP! We need to make it work with unix-domain sockets as well.

Sets the maximum number of file or socket descriptors that your process may open. You can pass this method an integer specifying the new size of the descriptor table. Returns the new descriptor-table size, which may be less than the number you requested. If you call this method with no arguments, it will simply return the current size of the descriptor table without attempting to change it.

The new limit on open descriptors ONLY applies to sockets and other descriptors that belong to EventMachine. It has NO EFFECT on the number of descriptors you can create in ordinary Ruby code.

Not available on all platforms. Increasing the number of descriptors beyond its default limit usually requires superuser privileges. (See set_effective_user for a way to drop superuser privileges while your program is running.)

A wrapper over the setuid system call. Particularly useful when opening a network server on a privileged port because you can use this call to drop privileges after opening the port. Also very useful after a call to set_descriptor_table_size, which generally requires that you start your process with root privileges.

This method has no effective implementation on Windows or in the pure-Ruby implementation of EventMachine. Call set_effective_user by passing it a string containing the effective name of the user whose privilege-level your process should attain. This method is intended for use in enforcing security requirements, consequently it will throw a fatal error and end your program if it fails.

For advanced users. This function sets the default timer granularity, which by default is slightly smaller than 100 milliseconds. Call this function to set a higher or lower granularity. The function affects the behavior of add_timer and add_periodic_timer. Most applications will not need to call this function.

The argument is a number of milliseconds. Avoid setting the quantum to very low values because that may reduce performance under some extreme conditions. We recommend that you not set a quantum lower than 10.

You may only call this function while an EventMachine loop is running (that is, after a call to EventMachine#run and before a subsequent call to EventMachine#stop).

set_timer_quantum in milliseconds. The underlying Reactor function wants a (possibly fractional) number of seconds.

EventMachine::start_server initiates a TCP server (socket acceptor) on the specified IP address and port. The IP address must be valid on the machine where the program runs, and the process must be privileged enough to listen on the specified port (on Unix-like systems, superuser privileges are usually required to listen on any port lower than 1024). Only one listener may be running on any given address/port combination. start_server will fail if the given address and port are already listening on the machine, either because of a prior call to start_server or some unrelated process running on the machine. If start_server succeeds, the new network listener becomes active immediately and starts accepting connections from remote peers, and these connections generate callback events that are processed by the code specified in the handler parameter to start_server.

The optional handler which is passed to start_server is the key to EventMachine‘s ability to handle particular network protocols. The handler parameter passed to start_server must be a Ruby Module that you must define. When the network server that is started by start_server accepts a new connection, it instantiates a new object of an anonymous class that is inherited from EventMachine::Connection, into which the methods from your handler have been mixed. Your handler module may redefine any of the methods in EventMachine::Connection in order to implement the specific behavior of the network protocol.

Callbacks invoked in response to network events always take place within the execution context of the object derived from EventMachine::Connection extended by your handler module. There is one object per connection, and all of the callbacks invoked for a particular connection take the form of instance methods called against the corresponding EventMachine::Connection object. Therefore, you are free to define whatever instance variables you wish, in order to contain the per-connection state required by the network protocol you are implementing.

start_server is often called inside the block passed to EventMachine::run, but it can be called from any EventMachine callback. start_server will fail unless the EventMachine event loop is currently running (which is why it‘s often called in the block suppled to EventMachine::run).

You may call start_server any number of times to start up network listeners on different address/port combinations. The servers will all run simultaneously. More interestingly, each individual call to start_server can specify a different handler module and thus implement a different network protocol from all the others.

Usage example

Here is an example of a server that counts lines of input from the remote peer and sends back the total number of lines received, after each line. Try the example with more than one client connection opened via telnet, and you will see that the line count increments independently on each of the client connections. Also very important to note, is that the handler for the receive_data function, which our handler redefines, may not assume that the data it receives observes any kind of message boundaries. Also, to use this example, be sure to change the server and port parameters to the start_server call to values appropriate for your environment.

 require 'rubygems'
 require 'eventmachine'

 module LineCounter

   MaxLinesPerConnection = 10

   def post_init
     puts "Received a new connection"
     @data_received = ""
     @line_count = 0

   def receive_data data
     @data_received << data
     while @data_received.slice!( /^[^\n]*[\n]/m )
       @line_count += 1
       send_data "received #{@line_count} lines so far\r\n"
       @line_count == MaxLinesPerConnection and close_connection_after_writing

 end # module LineCounter

 EventMachine::run {
   host,port = "", 8090
   EventMachine::start_server host, port, LineCounter
   puts "Now accepting connections on address #{host}, port #{port}..."
   EventMachine::add_periodic_timer( 10 ) { $stderr.write "*" }

stop_event_loop may called from within a callback method while EventMachine‘s processing loop is running. It causes the processing loop to stop executing, which will cause all open connections and accepting servers to be run down and closed. Callbacks for connection-termination will be called as part of the processing of stop_event_loop. (There currently is no option to panic-stop the loop without closing connections.) When all of this processing is complete, the call to EventMachine::run which started the processing loop will return and program flow will resume from the statement following EventMachine::run call.

Usage example

 require 'rubygems'
 require 'eventmachine'

 module Redmond

   def post_init
     puts "We're sending a dumb HTTP request to the remote peer."
     send_data "GET / HTTP/1.1\r\nHost:\r\n\r\n"

   def receive_data data
     puts "We received #{data.length} bytes from the remote peer."
     puts "We're going to stop the event loop now."

   def unbind
     puts "A connection has terminated."


 puts "We're starting the event loop now."
 EventMachine::run {
   EventMachine::connect "", 80, Redmond
 puts "The event loop has stopped."

This program will produce approximately the following output:

 We're starting the event loop now.
 We're sending a dumb HTTP request to the remote peer.
 We received 1440 bytes from the remote peer.
 We're going to stop the event loop now.
 A connection has terminated.
 The event loop has stopped.

Stop a TCP server socket that was started with EventMachine#start_server.