NAME SpeedyCGI - Speed up perl scripts by running them persistently SYNOPSIS #!/usr/bin/speedy ### Your Script Here 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"; DESCRIPTION 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: #!/usr/bin/perl to #!/usr/bin/speedy 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 results. 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 server. OPTIONS 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 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 SPEEDY_TIMEOUT. Module 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" below. Apache 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. Context 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: speedy The command-line "speedy" program, used normally with #! at the top of your script or from a shell prompt. mod_speedycgi The optional Apache mod_speedycgi module. module During perl execution via the CGI::SpeedyCGI module's getopt/setopt methods. Options Available BackendProg Command Line : -p Default Value : "/usr/bin/speedy_backend" Context : mod_speedycgi, speedy Description: Path to the speedy backend program. BufsizGet Command Line : -B Default Value : 131072 Context : speedy Description: Use bytes as the maximum size for the buffer that receives data from the perl backend. BufsizPost Command Line : -b Default Value : 131072 Context : speedy Description: Use bytes as the maximum size for the buffer that sends data to the perl backend. Group Command Line : -g Default Value : "none" Context : mod_speedycgi, speedy Description: 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. MaxBackends Command Line : -M Default Value : 0 (no max) Context : mod_speedycgi, speedy Description: If non-zero, limits the number of speedy backends running for this perl script to . MaxRuns Command Line : -r Default Value : 500 Context : mod_speedycgi, module, speedy Description: Once the perl interpreter has run times, re-exec the backend process. Zero indicates no maximum. This option is useful for processes that tend to consume resources over time. PerlArgs Command Line : N/A Default Value : "" Context : mod_speedycgi Description: Command-line options to pass to the perl interpreter. Timeout Command Line : -t Default Value : 3600 (one hour) Context : mod_speedycgi, module, speedy Description: If no new requests have been received after seconds, exit the persistent perl interpreter. Zero indicates no timeout. TmpBase Command Line : -T Default Value : "/tmp/speedy" Context : mod_speedycgi, speedy Description: Use the given prefix for creating temporary files. This must be a filename prefix, not a directory name. Version Command Line : -v Context : speedy Description: Print the SpeedyCGI version and exit. METHODS The following methods are available in the CGI::SpeedyCGI module. new Create a new CGI::SpeedyCGI object. my $sp = CGI::SpeedyCGI->new; register_cleanup($function_ref) 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. $sp->register_cleanup(\&cleanup_func); add_shutdown_handler($function_ref) 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}); set_shutdown_handler($function_ref) Deprecated. Similar to `add_shutdown_handler', but only allows for a single function to be registered. $sp->set_shutdown_handler(sub {$dbh->logout}); i_am_speedy 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. $sp->i_am_speedy; 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 interface: $CGI::SpeedyCGI::i_am_speedy setopt($optname, $value) Set one of the SpeedyCGI options given in the section on "Options Available". Returns the option's previous value. $optname is case- insensitive. $sp->setopt('TIMEOUT', 300); getopt($optname) Return the current value of one of the SpeedyCGI options. $optname is case-insensitive. $sp->getopt('TIMEOUT'); shutdown_now Shut down the perl interpreter right away. This function does not return. $sp->shutdown_now shutdown_next_time Shut down the perl interpreter as soon as this request is done. $sp->shutdown_next_time INSTALLATION 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 binaries. 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: Linux rpm -i Solaris gunzip .gz pkgadd -d BSD pkg_add 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 5.004 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 make test make install This will install the speedy and speedy_backend binaries in the same directory where perl was installed, and the SpeedyCGI.pm 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. 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 mod_speedycgi.so 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 see. LoadModule speedycgi_module modules/mod_speedycgi.so 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 /speedy/script-name. Alias /speedy/ /home/httpd/cgi-bin/ 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 Options ExecCGI FREQUENTLY ASKED QUESTIONS 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 altogether. Here's a good explanation of the problem - it's for mod_perl, but the same thing applies to speedycgi: http://perl.apache.org/faq/mod_perl_cgi.html#Variables_retain_their_value_fro 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- used. 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. USING GROUPS The group feature in SpeedyCGI can be used to help reduce the amount of memory used by the perl interpreters. When groups are not used (ie when group name is "none"), each perl script is given its own set of perl interpreters, separate from the perl interpreters used for other scripts. In SpeedyCGI each perl interpreter is also a separate unix process. When grouping is used, perl interpreters are put into a group. All perl interpreters in that group can run perl scripts in that 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. Group names are created by the person running SpeedyCGI based on their needs. There are two special group names "none" and "default". All other group names are created by the user of SpeedyCGI using the Group option described in the section on "OPTIONS". If you want to use the maximum amount of grouping possible (ie all scripts in the same interpreter), then you should always use the group name "default". When you do this, you will get the fewest number of perl interpreters possible. Each 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 cannot 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 is compiled and run out of package main, and always given its own interpreter. Other 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 group name "pariah") to keep them away from other scripts that 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 your other scripts. * To pass different perl or SpeedyCGI parameters to different scripts. The first script to start up in a group sets the perl and SpeedyCGI parameters used from then on for all scripts in that group. You may want to use separate groups to create separate policies for different scripts. Say you 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 user a MaxRuns setting of 5 for all of these scripts. You then want all your other scripts to run in a separate group with a normal MaxRuns policy. What you can do is edit the ten email scripts, and at the top, put in 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 the normal MaxRuns setting. DOWNLOADING Binaries Binaries for many OSes can be found at: http://daemoninc.com/SpeedyCGI/download.html Source Code The standard source code distribution can be retrieved from any CPAN mirror or from: http://daemoninc.com/SpeedyCGI/download.html http://www.cpan.org/modules/by-authors/id/H/HO/HORROCKS/ The latest development code can be obtained from the SourceForge CVS repository using the following commands: cvs -d:pserver:anonymous@cvs.SpeedyCGI.sourceforge.net:/cvsroot/speedycgi login cvs -z3 -d:pserver:anonymous@cvs.SpeedyCGI.sourceforge.net:/cvsroot/speedycgi co 2.x Press Enter when prompted for a password. AUTHOR Sam Horrocks http://daemoninc.com sam@daemoninc.com Contributors 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 SEE ALSO perl(1), httpd(8), apxs(8). MORE INFORMATION SpeedyCGI Home Page http://daemoninc.com/SpeedyCGI/ Mailing List * SpeedyCGI users mailing list - speedycgi-users@lists.sourceforge.net. Archives and subscription information are at http://lists.sourceforge.net/lists/listinfo/speedycgi-users * SpeedyCGI announcements mailing list - speedycgi- announce@lists.sourceforge.net. Archives and subscription information are at http://lists.sourceforge.net/lists/listinfo/speedycgi-announce 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 http://www.sourceforge.net/projects/speedycgi/. Go to the Bug Tracking menu and select the group "bug" for bugs, or the group "rfe" for the todo list. Japanese Translation http://member.nifty.ne.jp/hippo2000/perltips/CGI/SpeedyCGI.htm Benchmarks http://daemoninc.com/SpeedyCGI/benchmarks/ Success Stories http://daemoninc.com/SpeedyCGI/success_stories/ Revision History http://daemoninc.com/SpeedyCGI/CGI-SpeedyCGI/Changes YAPC 2001 Presentation I gave an Introduction to SpeedyCGI talk at YAPC 2001. It can be found at http://daemoninc.com/SpeedyCGI/yapc_2001/ COPYRIGHT Copyright (C) 2002 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 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General 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 (http://www.apache.org/).