Non-buffered processor in Perl

Lets say that you have the following problem: you want to write a script which processes the output of a program and writes out the modified somewere, with as little buffering as possible. One concrete example (for which I needed the script) is log rotation: you want to save the output of a program (which doesn’t support log rotation by itself) to a logfile which gets rotate at midnight (because it includes the date in the name). Also, an other constraint is that you would like to “time-out” the read attempt to do some maintenance work (for example you would like to rotate your logs – create the files with the different dates – even when no data is written to it).

One possibility would have been to use IO::Select, however it doesn’t support filehandles on Windows (not that Windows wouldn’t have the API to do so, it’s just that nobody was implemented it in Perl core). Fortunately we can have something very similar to it:

use strict;
use warnings;
use IO::Handle;

binmode STDIN;
binmode STDOUT;

my $BUFFLEN = 4096;
while (1) {
  my $buffer;
  my $read_count = sysread(STDIN, $buffer, $BUFFLEN);
  if (not defined($read_count)) {
    # nothing to read, pause
    sleep 0.1;
  if (0 == $read_count) {
    # EOF condition
    exit 0;
  syswrite(STDOUT, $buffer);

The magic is done here by STDIN->blocking(0); which sets the filehandle into a non-blocking mode, returning “undef” is there is nothing to read. Whenever this happens (ie. there is no data on the input) it pauses for a brief moment (1/10 of a second) and then retries.

Some other remarks about the code:

  • the input is read and the output is written as binary. This means that no processing is done which could screw up the flow (for example trying to convert data between character sets and screwing up Unicode characters)
  • care is taken to introduce minimal buffering. Output is produced as soon as the input arrives. For more intricacies of Linux buffering see this nice article at pixelbeat.
  • the code is very performant. I’ve measured throughputs up to 1.4 Gb/sec and can certainly handle anything the disk can (if we consider it in the context of log rotator)
  • the code has been tested and works on both Windows (Strawberry Perl 5.12.1) and Linux. It should work mostly anywhere since it uses Core Perl.

Leave a Reply

Your email address will not be published. Required fields are marked *