SYNOPSIS

use Proc::Reliable;

Create a new process object

$myproc = Proc::Reliable->new();

Run a subprocess and collect its output

   $output = $myproc->run("/bin/ls -l");

Check for problems

   if($myproc->status()) {
     print("problem!\n");
   }

Run another subprocess, keeping stdout and stderr separated. Also, send the subprocess some data on stdin.

   $msg = "Hello World\n");
   $p->want_single_list(0);
   $stdout = $p->run("/usr/bin/fastmail - [email protected]", $msg);
   if($p->status()) {
     print("problem: ", $p->stderr(), "\n");
   }

Another way to get output

   ($stdout, $stderr, $status, $msg) = $p->run("/bin/ls -l");

OPTIONS

Run Modes

 $p->run("shell-command-line");  # Launch a shell process
 $p->run("cmdline", "data");     # Launch a shell process with stdin data
 $p->run(["cmd", "arg1", ...]);  # Bypass shell processing of arguments
 $p->run(sub { ... });           # Launch a perl subroutine
 $p->run(\&subroutine);          # Launch a perl subroutine

Option settings below represent defaults

 $p->num_tries(1);           # execute the program only once
 $p->time_per_try(60);       # time per try 60 sec
 $p->maxtime(60);            # set overall timeout
 $p->time_btw_tries(5);      # time between tries 5 sec
 $p->want_single_list();     # return STDOUT and STDERR together
 $p->accept_no_error();      # Re-try if any STDERR output
 $p->pattern_stdout($pat);   # require STDOUT to match regex $pat
 $p->pattern_stderr($pat);   # require STDERR to match regex $pat
 $p->allow_shell(1);         # allowed to use shell for operation
 $p->child_exit_time(1.0);   # timeout for child to exit after it closes stdout
 $p->sigterm_exit_time(0.5); # timeout for child to exit after sigterm
 $p->sigkill_exit_time(0.5); # timeout for child to exit after sigkill
 $p->input_chunking(0);      # feed stdin data line-by-line to subprocess
 $p->stdin_error_ok(0);      # ok if child exits without reading all stdin
 $p->stdout_cb(undef);       # callback function for line-by-line stdout
 $p->stderr_cb(undef);       # callback function for line-by-line stderr

Getting output

 $out = $p->stdout();        # stdout produced by last run()
 $err = $p->stderr();        # stderr produced by last run()
 $stat = $p->status();       # exit code produced by last run()
 $msg = $p->msg();           # module messages produced by last run()

Debug

Proc::Reliable::debug($level); # Turn debug on

OVERVIEW

Proc::Reliable is a class for simple, reliable and configurable subprocess execution in perl. In particular, it is especially useful for managing the execution of 'problem' programs which are likely to fail, hang, or otherwise behave in an unruly manner.

Proc::Reliable includes all the functionality of the backticks operator and system() functions, plus many common uses of fork() and exec(), open2() and open3(). Proc::Reliable incorporates a number of options, including sending data to the subprocess on STDIN, collecting STDOUT and STDERR separately or together, killing hung processes, timouts and automatic retries.

DESCRIPTION

A new process object is created by

   $myproc = Proc::Reliable->new();

The default will run a subprocess only once with a 60-second timeout. Either shell-like command lines or references to perl subroutines can be specified for launching a process in background. A simple list process, for example, can be started via the shell as

   $out = $myproc->run("ls");

To separate stdout, stderr, and exit status:

   ($out, $err, $status, $msg) = $myproc->run("ls");

The output data is also stored within the $myproc object for later retrieval. You can also run a perl subroutine in a subprocess, with

   $myproc->run(sub { return <*>; });

The run Method will try to run the named process. If the process times out (after time_per_try seconds) or has an error defined as unacceptable and you would like to re-run it, you can use the num_tries option. Use the time_btw_tries option to set the number of seconds between runs. This can repeat until maxtime seconds have elapsed.

When using num_tries, the user can specify what constitutes an unacceptable error of STDOUT or STDERR output --- i.e. demanding a retry. One common shorthand is to have the run method retry if there is any return from STDERR.

   $myproc->accept_no_error();    # Re-try if any STDERR
   $myproc->pattern_stdout($pat); # require STDOUT to match regex $pat
   $myproc->pattern_stderr($pat); # require STDERR to match regex $pat

Subprocess completion is detected when the process closes all filehandles. The process must then exit before child_exit_time expires, or it will be killed. If the subprocess does not exit, it is sent a TERM signal unless sigterm_exit_time is 0. then if it does not exit before sigterm_exit_time expires, it is sent a KILL signal unless sigkill_exit_time is 0. then if it does not exit before sigkill_exit_time expires an error is generated. waiting is done in 0.01 second increments.

Proc::Reliable is not MT-Safe due to signals usage.

METHODS

The following methods are available:

new (Constructor)

Create a new instance of this class by writing either

    $proc = new Proc::Reliable;   or   $proc = Proc::Reliable->new();

The new method accepts any valid configuration options:

    $proc = Proc::Reliable->new('maxtime' => 200, 'num_tries' => 3);

run

Run a new process and collect the standard output and standard error via separate pipes.

  $out = $proc->run("program-name");
 ($out, $err, $status, $msg) = $proc->run("program-name");

by default with a single return value, stdout and stderr are combined to a single stream and returned. with 4 return values, stdout and stderr are separated, and the program exit status is also returned. $msg contains messages from Proc::Reliable when errors occur. Set want_single_list(1) to force stdout and stderr to be combined, and want_single_list(0) to force them separated. The results from run() are stored as member data also:

  $proc->want_single_list(0);
  $proc->run("program");
  if($proc->status) {
    print($proc->stderr);
    exit;
  }
  else {
    print($proc->stdout);
  }

Program exit status is returned in the same format as exec(): bits 0-7 set if program exited from a signal, bits 8-15 are the exit status on a normal program exit.

You can also set up callbacks to run a function of your choice as each line of stdout and stderr is produced by the child process using the stdout_cb and stderr_cb options.

There are a number of other options. You can also feed the forked program data on stdin via a second argument to run():

 $myinput = "hello\ntest\n";
 $output = $proc->run("program-name", $myinput);

The first option to run() supports three forms: 1) string containing command string to execute. this incurs shell parsing. 2) arrayref containing split command string to execute. this bypasses shell parsing. 3) coderef to perl function. The first two options are executed via exec(), so the specifics of incurring shell parsing are the same.

The second option to run() supports two forms: 1) string containing data to feed on stdin 2) stringref pointing to data to feed on stdin

You can start execution of an independent Perl function (like ``eval'' except with timeout, retries, etc.). Simply provide the function reference like

 $output = $proc->run(\&perl_function);

or supply an unnamed subroutine:

 $output = $proc->run( sub { sleep(1) } );

The run Method returns after the the function finishes, one way or another.

debug

Switches debug messages on and off --- Proc::Reliable::debug(1) switches them on, Proc::Reliable::debug(0) keeps Proc::Reliable quiet.

maxtime

Return or set the maximum time in seconds per run method call. Default is 300 seconds (i.e. 5 minutes).

num_tries

Return or set the maximum number of tries the run method will attempt an operation if there are unallowed errors. Default is 5.

time_per_try

Return or set the maximum time in seconds for each attempt which run makes of an operation. Multiple tries in case of error can go longer than this. Default is 30 seconds.

time_btw_tries

Return or set the time in seconds between attempted operations in case of unacceptable error. Default is 5 seconds.

child_exit_time

When the subprocess closes stdout, it is assumed to have completed normal operation. It is expected to exit within the amount of time specified. If it does not exit, it will be killed (with SIGTERM). This option can be disabled by setting to '0'. Values are in seconds, with a resolution of 0.01.

sigterm_exit_time

If the time_per_try or max_time has been exceeded, or if child_exit_time action has not succeeded, the subprocess will be killed with SIGTERM. This option specifies the amount of time to allow the process to exit after closing stdout. This option can be disabled by setting to '0'. Values are in seconds, with a resolution of 0.01.

sigkill_exit_time

Similar to sigterm_exit_time, but a SIGKILL is sent instead of a SIGTERM. When both options are enabled, the SIGTERM is sent first and SIGKILL is then sent after the specified time only if the subprocess is still alive. This option can be disabled by setting to '0'. Values are in seconds, with a resolution of 0.01.

input_chunking

If data is being written to the subprocess on stdin, this option will cause the module to split() the input data at linefeeds, and only feed the subprocess a line at a time. This option typically would be used when the subprocess is an application with a command prompt and does not work properly when all the data is fed on stdin at once. The module will feed the subprocess one line of data on stdin, and will then wait until some data is produced by the subprocess on stdout or stderr. It will then feed the next line of data on stdin.

stdout_cb

Set up a callback function to get stdout data from the child line-by-line. The function you supply will be called whenever the child prints a line onto stdout. This is the only way to get output from the child while it is still running, the normal method will give you all the output at once after the child exits.

stderr_cb

Similar to stdout_cb for stderr data.

REQUIREMENTS

I recommend using at least perl 5.003.

AUTHORS

Proc::Reliable by Dan Goldwater <dgold at zblob dot com>

Based on Proc::Short, written by John Hanju Kim <[email protected]>.

Contributions by Stephen Cope and Jason Robertson.

COPYRIGHT

Copyright 2001 by Dan Goldwater, all rights reserved. Copyright 1999 by John Hanju Kim, all rights reserved.

This program is free software, you can redistribute it and/or modify it under the same terms as Perl itself.