SpeedyCGI - Speed up perl scripts by running them persistently.


     ### Your Script Here.  For example:
     print "Content-type: text/html\n\nHello World!\n";

     ## Optionally, use the CGI::SpeedyCGI module for various things

     # Create a SpeedyCGI object
     use CGI::SpeedyCGI;
     my $sp = CGI::SpeedyCGI->new;

     # See if we are running under SpeedyCGI or not.
     print "Running under speedy=", $sp->i_am_speedy ? 'yes' : 'no', "\n";

     # Register a shutdown handler
     $sp->add_shutdown_handler(sub { do something here });

     # Register a cleanup handler
     $sp->register_cleanup(sub { do something here });

     # Set/get some SpeedyCGI options
     $sp->setopt('timeout', 30);
     print "maxruns=", $sp->getopt('maxruns'), "\n";

    SpeedyCGI is a way to run perl scripts persistently, which can make them
    run much more quickly. A script can be made to to run persistently by
    changing the interpreter line at the top of the script from:




    After the script is initially run, instead of exiting, the perl
    interpreter is kept running. During subsequent runs, this interpreter is
    used to handle new executions instead of starting a new perl interpreter
    each time. A very fast frontend program, written in C, is executed for
    each request. This fast frontend then contacts the persistent Perl
    process, which is usually already running, to do the work and return the

    By default each perl script runs in its own Unix process, so one perl
    script can't interfere with another. Command line options can also be
    used to deal with programs that have memory leaks or other problems that
    might keep them from otherwise running persistently.

    SpeedyCGI can be used to speed up perl CGI scripts. It conforms to the
    CGI specification, and does not run perl code inside the web server.
    Since the perl interpreter runs outside the web server, it can't cause
    problems for the web server itself.

    SpeedyCGI also provides an Apache module so that under the Apache web
    server, scripts can be run without the overhead of doing a fork/exec for
    each request. With this module a small amount of frontend code is run
    within the web server - the perl interpreters still run outside the

    SpeedyCGI and PersistentPerl are currently both names for the same code.
    SpeedyCGI was the original name, but because people weren't sure what it
    did, the name PersistentPerl was picked as an alias. At some point
    SpeedyCGI will be replaced by PersistentPerl, or become a sub-class of
    PersistentPerl to avoid always having two distributions.

  Setting Option Values

    SpeedyCGI options can be set in several ways:

    Command Line
        The speedy command line is the same as for regular perl, with the
        exception that SpeedyCGI specific options can be passed in after a

        For example the line:

                #!/usr/bin/speedy -w -- -t300

        at the top of your script will set the perl option "`-w'" and will
        pass the "`-t'" option to SpeedyCGI, setting the Timeout value to
        300 seconds.

        Environment variables can be used to pass in options. This can only
        be done before the initial execution, not from within the script
        itself. The name of the environment variable is always SPEEDY_
        followed by the option name in upper-case. For example to set the
        speedy Timeout option, use the environment variable named

        The CGI::SpeedyCGI module provides the setopt method to set options
        from within the perl script at runtime. There is also a getopt
        method to retrieve the current options. See the section on "METHODS"

        If you are using the optional Apache module, SpeedyCGI options can
        be set in the httpd.conf file. The name of the apache directive will
        always be Speedy followed by the option name. For example to set the
        Timeout option, use the apache directive SpeedyTimeout.


    Not all options below are available in all contexts. The context for
    which each option is valid is listed on the "Context" line in the
    section below. There are three contexts:

        The command-line "speedy" program, used normally with #! at the top
        of your script or from a shell prompt.

        The optional Apache mod_speedycgi module.

        During perl execution via the CGI::SpeedyCGI module's getopt/setopt

  Options Available

            Command Line    : -p<string>
            Default Value   : "/usr/bin/speedy_backend"
            Context         : mod_speedycgi, speedy


                Path to the speedy backend program.

            Command Line    : -B<number>
            Default Value   : 131072
            Context         : speedy


                Use <number> bytes as the maximum size for the buffer that
                receives data from the perl backend.

            Command Line    : -b<number>
            Default Value   : 131072
            Context         : speedy


                Use <number> bytes as the maximum size for the buffer that
                sends data to the perl backend.

            Command Line    : -g<string>
            Default Value   : "none"
            Context         : mod_speedycgi, speedy


                Allow a single perl interpreter to run multiple scripts.
                All scripts that are run with the same group name and by
                the same user will be run by the same group of perl
                interpreters. If the group name is "none" then grouping is
                disabled and each interpreter will run one script.
                Different group names allow scripts to be separated into
                different groups. Name is case-sensitive, and only the
                first 12-characters are significant. Specifying an empty
                group name is the same as specifying the group name
                "default" - this allows just specifying "-g" on the command
                line to turn on grouping.

            Command Line    : -M<number>
            Default Value   : 0 (no max)
            Context         : mod_speedycgi, speedy


                If non-zero, limits the number of speedy backends running
                for this perl script to <number>.

            Command Line    : -r<number>
            Default Value   : 500
            Context         : mod_speedycgi, module, speedy


                Once the perl interpreter has run <number> times, re-exec
                the backend process.  Zero indicates no maximum.  This
                option is useful for processes that tend to consume
                resources over time.

            Command Line    : N/A
            Default Value   : ""
            Context         : mod_speedycgi


                Command-line options to pass to the perl interpreter.

            Command Line    : -t<number>
            Default Value   : 3600 (one hour)
            Context         : mod_speedycgi, module, speedy


                If no new requests have been received after <number>
                seconds, exit the persistent perl interpreter.  Zero
                indicates no timeout.

            Command Line    : -T<string>
            Default Value   : "/tmp/speedy"
            Context         : mod_speedycgi, speedy


                Use the given prefix for creating temporary files.  This
                must be a filename prefix, not a directory name.

            Command Line    : -v
            Context         : speedy


                Print the SpeedyCGI version and exit.

    The following methods are available in the CGI::SpeedyCGI module.

    new Create a new CGI::SpeedyCGI object.

            my $sp = CGI::SpeedyCGI->new;

        Register a function that will be called at the end of each request,
        after your script finishes running, but before STDOUT and STDERR are
        closed. Multiple functions can be added by calling the method more
        than once. At the end of the request, each function will be called
        in the order in which it was registered.


        Add a function to the list of functions that will be called right
        before the perl interpreter exits. This is not at the end of each
        request, it is when the perl interpreter decides to exit completely
        due to a Timeout or reaching MaxRuns.

            $sp->add_shutdown_handler(sub {$dbh->logout});

        Deprecated. Similar to `add_shutdown_handler', but only allows for a
        single function to be registered.

            $sp->set_shutdown_handler(sub {$dbh->logout});

        Returns a boolean telling whether this script is running under
        SpeedyCGI or not. A perl script can run under regular perl, or under
        SpeedyCGI. This method allows the script to tell which environment
        it is in.


        To make your script as portable as possible, you can use the
        following test to make sure both the SpeedyCGI module is available
        and you are running under SpeedyCGI:

            if (eval {require CGI::SpeedyCGI} && CGI::SpeedyCGI->i_am_speedy) {
                Do something SpeedyCGI specific here...

        To increase the speed of this check you can also test whether the
        following variable is defined instead of going through the object


    setopt($optname, $value)
        Set one of the SpeedyCGI options given in the section on "Options
        Available". Returns the option's previous value. $optname is case-

            $sp->setopt('TIMEOUT', 300);

        Return the current value of one of the SpeedyCGI options. $optname
        is case-insensitive.


        Shut down the perl interpreter right away. This function does not


        Shut down the perl interpreter as soon as this request is done.


    To install SpeedyCGI you will need to either download a binary package
    for your OS, or compile SpeedyCGI from source code. See the section on
    "DOWNLOADING" for information on where to obtain the source code and

  Binary Installation

    Once you have downloaded the binary package for your OS, you'll need to
    install it using the normal package tools for your OS. The commands to
    do that are:

         rpm -i <filename>

         gunzip <filename>.gz
         pkgadd -d <filename>

    BSD pkg_add <filename>

    If you are also installing the apache module you will have to configure
    Apache as documented in the section on "Apache Configuration".

  Source Code Installation

    To compile SpeedyCGI you will need perl 5.005_03 or later, and a C
    compiler, preferably the same one that your perl distribution was
    compiled with. SpeedyCGI is known to work under Solaris, Redhat Linux,
    FreeBSD and OpenBSD. There may be problems with other OSes or earlier
    versions of Perl. SpeedyCGI may not work with threaded perl -- as of
    release 2.10, Linux and Solaris seem to work OK with threaded perl, but
    FreeBSD does not.

  Standard Install

    To do a standard install from source code, execute the following:

        perl Makefile.PL
        make test
        make install

    This will install the speedy and speedy_backend binaries in the same
    directory where perl was installed, and the module in the
    standard perl lib directory. It will also attempt to install the
    mod_speedycgi module if you have the command apxs in your path.

  Install in a Different Directory

    If you don't have permission to install into the standard perl
    directory, or if you want to install elsewhere, the easiest way is to
    compile and install your own copy of perl in another location, then use
    your new version of perl when you run "perl Makefile.PL". The SpeedyCGI
    binaries and module will then be installed in the same location as the
    new version of perl.

    If you can't install your own perl, you can take the following steps:

    *   Edit src/optdefs and change the default value for BackendProg to the
        location where speedy_backend will be installed.

    *   Compile as above, then manually copy the speedy and speedy_backend
        binaries to where you want to install them.

    *   If you want to use the CGI::SpeedyCGI module in your code (it's not
        required), you will have to use "use lib" so it can be located.

  Setuid Install

    SpeedyCGI has limited support for running setuid - installing this way
    may compromise the security of your system. To install setuid do the

    *   Run "perl Makefile.PL"

    *   Edit speedy/Makefile and add "-DIAMSUID" to the end of the "DEFINE = "

    *   Run make

    *   Take the resulting "speedy" binary and install it suid-root as

    *   Change your setuid scripts to use /usr/bin/speedy_suid as the

    This has been know to work in Linux and FreeBSD. Solaris will work as
    long as the Group option is set to "none".

  Apache Installation

    To compile the optional apache mod_speedycgi module you must have the
    apxs command in your path. Redhat includes this command with the
    "apache-devel" RPM, though it may not work properly for installation.

    If the apache installation fails:

    *   Copy the from the mod_speedycgi directory, or from the
        mod_speedycgi2/.libs directory, to wherever your apache modules are
        stored (try /usr/lib/apache)

    *   Edit your httpd.conf (try /etc/httpd/conf/httpd.conf) and add the
        following lines. The path at the end of the LoadModule directive may
        be different in your installation -- look at other LoadModules to

            LoadModule speedycgi_module modules/

        If you are using Apache-1, also add:

            AddModule mod_speedycgi.c

  Apache Configuration

    Once mod_speedycgi is installed, it has to be configured to be used for
    your perl scripts. There are two methods.

    Warning! The instructions below may compromise the security of your web
    site. The security risks associated with SpeedyCGI are similar to those
    of regular CGI. If you don't understand the security implications of the
    changes below then don't make them.

    1. Path Configuration
        This is similar to the way /cgi-bin works - everything under this
        path is handled by SpeedyCGI. Add the following lines near the top
        of your httpd.conf - this will cause all scripts in your cgi-bin
        directory to be handled by SpeedyCGI when they are accessed as

            Alias /speedy/ /home/httpd/cgi-bin/
            <Location /speedy>
                SetHandler speedycgi-script
                Options ExecCGI
                allow from all

    2. File Extension Configuration
        This will make SpeedyCGI handle all files with a certain extension,
        similar to the way .cgi files work. Add the following lines near the
        top of your httpd.conf file - this will set up the file extension
        ".speedy" to be handled by SpeedyCGI.

            AddHandler speedycgi-script .speedy
            <Location />
                Options ExecCGI

    How does the speedy front end connect to the back end process?
        Via a Unix socket in /tmp. A queue is kept in a shared file in /tmp
        that holds an entry for each process. In that queue are the pids of
        the perl processes waiting for connections. The frontend pulls a
        process out of this queue, connects to its socket, sends over the
        environment and argv, and then uses this socket for stdin/stdout to
        the perl process.

    If another request comes in while SpeedyCGI script is running, does the client
    have to wait or is another process started?  Is there a way to set a limit
    on how many processes get started?
        If another request comes while all the perl processes are busy, then
        another perl process is started. Just like in regular perl there is
        normally no limit on how many processes get started. But, the
        processes are only started when the load is so high that they're
        necessary. If the load goes down, the processes will die off due to
        inactivity, unless you disable the timeout.

        Starting in version 1.8.3 an option was added to limit the number of
        perl backends running. See MaxBackends in the section on "Options
        Available" above.

    How much of perl's state is kept when speedy starts another request?
    Do globals keep their values?  Are destructors run after the request?
        Globals keep their values. Nothing is destroyed after the request.
        STDIN/STDOUT/STDERR are closed -- other files are not. `%ENV' and
        `@ARGV' are the only globals changed between requests.

    How can I make sure speedy restarts when I edit a perl library used
    by the CGI?
        Do a touch on the main cgi file that is executed. The mtime on the
        main file is checked each time the front-end runs.

    Do I need to be root to install and/or run SpeedyCGI?
        No, root is not required.

    How can I determine if my perl app needs to be changed to work with
    speedy?  Or is there no modification necessary?
        You may have to make modifications.

        Globals retain their values between runs, which can be good for
        keeping persistent database handles for example, or bad if your code
        assumes they're undefined.

        Also, if you create global variables with "my", you shouldn't try to
        reference those variables from within a subroutine - you should pass
        them into the subroutine instead. Or better yet just declare global
        variables with "use vars" instead of "my" to avoid the problem

        Here's a good explanation of the problem - it's for mod_perl, but
        the same thing applies to speedycgi:


        If all else fails you can disable persistence by setting MaxRuns to
        1. The only benefit of this over normal perl is that speedy will
        pre-compile your script.

    How do I keep a persistent connection to a database?
        Since globals retain their values between runs, the best way to do
        this is to store the connection in a global variable, then check on
        each run to see if that variable is already defined.

        For example, if your code has an "open_db_connection" subroutine
        that returns a database connection handle, you can use the code
        below to keep a persistent connection:

            use vars qw($dbh);
            unless (defined($dbh)) {
                $dbh = &open_db_connection;

        This code will store a persistent database connection handle in the
        global variable "$dbh" and only initialize it the first time the
        code is run. During subsequent runs, the existing connection is re-

        You may also want to check the connection each time before using it,
        in case it is not working for some reason. So, assuming you have a
        subroutine named "db_connection_ok" that returns true if the db
        connection is working, you can use code like this:

            use vars qw($dbh);
            unless (defined($dbh) && &db_connection_ok($dbh)) {
                $dbh = &open_db_connection;

    Why do scripts with persistent Oracle database connections hang?
        When using an IPC connection to Oracle, an oracle process is fork'ed
        and exec'ed and keeps the stdout connection open, so that the web
        server never gets an EOF. To fix the problem, either switch to using
        a TCP connection to the database, or add the following perl code
        somewhere before the DBI->connect statement:

            use Fcntl;
            fcntl(STDOUT, F_SETFD, FD_CLOEXEC);

        This will set the close-on-exec flag on standard out so it is closed
        when oracle is exec'ed.

    The group feature in SpeedyCGI can be used to help reduce the amount of
    memory used by the perl interpreters. By default groups are not used
    (group name is "none"), and each perl script is given its own set of
    perl interpreters. Each perl interpreter is also a separate system

    When grouping is used, perl interpreters and perl scripts are put in a
    group. All perl interpreters in a group can run perl scripts in the same
    group. What this means is that by putting all your scripts into one
    group, there could be one perl interpreter running all the perl scripts
    on your system. This can greatly reduce your memory needs when running
    lots of different perl scripts.

    SpeedyCGI group names are entities unto themselves. They are not
    associated with Unix groups, or with the Group directive in Apache.
    Expect for the two special group names "none" and "default", all group
    names are created by the user of SpeedyCGI using the Group option
    described in the section on "OPTIONS"

    If you want the maximum amount of grouping possible then you should run
    all scripts with the group option set to "default". This the group name
    used if you just specify "-g" on the command line without an explicit
    group name. When you do this, you will get the fewest number of perl
    interpreters possible - any perl interpreter will be able to run any of
    your perl scripts.

    Although using group "default" for all scripts results in the most
    efficient use of resources, it's not always possible or desirable to do
    this. You may want to use other group names for the following reasons:

    * To isolate misbehaving scripts, or scripts that don't work in groups.
        Some scripts won't work in groups. When perl scripts are grouped
        together they are each given their own unique package name - they
        are not run out of the "main" package as they normally would be. So,
        for example, a script that explicitly uses "main" somewhere in its
        code to find its subroutines or variables probably won't work in
        groups. In this case, it's probably best to run such a script with
        group "none", so it's compiled and run out of package main, and
        always given its own interpreter.

        In other cases, scripts may make changes to included packages, etc,
        that may break other scripts running in the same interpreter. In
        this case such scripts can be given their own group name (like
        "pariah") to keep them away from scripts they are incompatible with.
        The rest of your scripts can then run out of group "default". This
        will ensure that the "pariah" scripts won't run within the same
        interpreter as the other scripts.

    * To pass different perl or SpeedyCGI parameters to different scripts.
        You may want to use separate groups to create different policies for
        different scripts.

        For example, you may have an email application that contains ten
        perl scripts, and since the common perl code used in this
        application has a bad memory leak, you want to use a MaxRuns setting
        of 5 for all of these scripts. You want to run all your other
        scripts with a normal MaxRuns setting. To accomplish this you can
        edit the ten email application scripts, and at the top use the line:

            #!/usr/bin/speedy -- -gmail -r5

        In the rest of your perl scripts you can use:

            #!/usr/bin/speedy -- -g

        What this will do is put the ten email scripts into a group of their
        own (named "mail") and give them all the default MaxRuns value of 5.
        All other scripts will be put into the group named "default", and
        this group will have a normal MaxRuns setting.


    Binaries for many OSes can be found at:

  Source Code

    The standard source code distribution can be retrieved from any CPAN
    mirror or from:

    The latest development code can be obtained from the SourceForge CVS
    repository using the following commands:

     cvs login 
     cvs -z3 co 2.x

    Press Enter when prompted for a password.

        Sam Horrocks


    A lot of people have helped out with code, patches, ideas, resources,
    etc. I'm sure I'm missing someone here - if so, please drop me an email.

    *   Gunther Birznieks

    *   Diana Eichert

    *   Takanori Kawai

    *   Robert Klep

    *   Marc Lehmann

    *   James McGregor

    *   Josh Rabinowitz

    *   Dave Parker

    *   Craig Sanders

    *   Joseph Wang

    perl(1), httpd(8), apxs(8).

  SpeedyCGI Home Page

  Mailing List

    *   SpeedyCGI users mailing list -
        Archives and subscription information are at

    *   SpeedyCGI announcements mailing list - speedycgi- Archives and subscription
        information are at

  Bugs and Todo List

    Please report any bugs or requests for changes to the mailing list.

    The current bugs / todo list can be found at Go to the Bug Tracking
    menu and select the group "bug" for bugs, or the group "rfe" for the
    todo list.

  Japanese Translation


  Success Stories

  Revision History

  YAPC 2001 Presentation

    I gave an Introduction to SpeedyCGI talk at YAPC 2001. It can be found

    Copyright (C) 2003 Sam Horrocks

    This program is free software; you can redistribute it and/or modify it
    under the terms of the GNU General Public License as published by the
    Free Software Foundation; either version 2 of the License, or (at your
    option) any later version.

    This program is distributed in the hope that it will be useful, but
    WITHOUT ANY WARRANTY; without even the implied warranty of
    Public License for more details.

    You should have received a copy of the GNU General Public License along
    with this program; if not, write to the Free Software Foundation, Inc.,
    59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.

    This product includes software developed by the Apache Software
    Foundation (