man LWP::UserAgent::Determined (3): a virtual browser that retries errors

SYNOPSIS

use strict;
use LWP::UserAgent::Determined;
my $browser = LWP::UserAgent::Determined->new;
my $response = $browser->get($url, headers... );

DESCRIPTION

This class works just like LWP::UserAgent (and is based on it, by being a subclass of it), except that when you use it to get a web page but run into a possibly-temporary error (like a DNS lookup timeout), it'll wait a few seconds and retry a few times.

It also adds some methods for controlling exactly what errors are considered retry-worthy and how many times to wait and for how many seconds, but normally you needn't bother about these, as the default settings are relatively sane.

METHODS

This module inherits all of LWP::UserAgent's methods, and adds the following.

$timing_string = $browser->timing();

$browser->timing( "10,30,90" )

The "timing" method gets or sets the string that controls how many times it should retry, and how long the pauses should be.

If you specify empty-string, this means not to retry at all.

If you specify a string consisting of a single number, like ``10'', that means that if the first request doesn't succeed, then "$browser->get(...)" (or any other method based on "request" or "simple_request") should wait 10 seconds and try again (and if that fails, then it's final).

If you specify a string with several numbers in it (like ``10,30,90''), then that means $browser can retry as that many times (i.e., one initial try, plus a maximum of the three retries, because three numbers there), and that it should wait first those numbers of seconds each time. So "$browser->timing( "10,30,90" )" basically means:

  try the request; return it unless it's a temporary-looking error;
  sleep 10;
  retry the request; return it unless it's a temporary-looking error;
  sleep 30;
  retry the request; return it unless it's a temporary-looking error;
  sleep 90  the request;
  return it;

The default value is ``1,3,15''.

$http_codes_hr = $browser->codes_to_determinate();

This returns the hash that is the set of HTTP codes that merit a retry (like 500 and 408, but unlike 404 or 200). You can delete or add entries like so;

  $http_codes_hr = $browser->codes_to_determinate();
  delete $http_codes_hr->{408};
  $http_codes_hr->{567} = 1;

(You can actually set a whole new hashset with "$browser->codes_to_determinate($new_hr)", but there's usually no benefit to that as opposed to the above.)

The current default is 408 (Timeout) plus some 5xx codes.

$browser->before_determined_callback()

$browser->before_determined_callback( \&some_routine );

$browser->after_determined_callback()

$browser->after_determined_callback( \&some_routine );

These read (first two) or set (second two) callbacks that are called before the actual HTTP/FTP/etc request is made. By default, these are set to undef, meaning nothing special is called. If you want to alter try requests, or inspect responses before any retrying is considered, you can set up these callbacks.

The arguments passed to these routines are:

0: the current $browser object
1: an arrayref to the list of timing pauses (based on $browser->timing)
2: the duration of the number of seconds we'll pause if this request fails this time, or undef if this is the last chance.
3: the value of $browser->codes_to_determinate
4: an arrayref of the arguments we pass to LWP::UserAgent::simple_request (the first of which is the request object)
(5): And, only for after_determined_callback, the response we just got.

Example use:

  $browser->before_determined_callback( sub {
    print "Trying ", $_[4][0]->uri, " ...\n";
  });

IMPLEMENTATION

This class works by overriding LWP::UserAgent's "simple_request" method with its own around-method that just loops. See the source of this module; it's straightforward. Relatively.

COPYRIGHT AND DISCLAIMER

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.

AUTHOR

Originally created by Sean M. Burke, "[email protected]"