I did write a pair of scripts - IIRC they wrote log files of the number incremented in a file and then I checked if there were any duplicates reported between the two logs or any numbers out of sequence. All seemed as expected.

So either I was not testing it in a way to break it, or it is no longer an issue.

Perrin I did read tilly's articles (here's one) which states flock does not work for Linux. Since I can't make it not work I was looking for a script to show me that it fails.

Let's see if I still have the script... ah, here was my test script:

#!/usr/bin/perl -w
use strict;
use Fcntl qw(:DEFAULT :flock);
use Time::HiRes 'usleep';
use Devel::Peek;

open LOG, ">$$.log" or die $!;

while ( 1 ) {
   open LOCK, "lock.file" or die "lock file $!";
   die "$$ failed to get lock" unless flock(LOCK,LOCK_EX);

   # perlfaq example
   sysopen(FH, "numfile", O_RDWR|O_CREAT)  
        or die "can't open numfile: $!";
   my $num = <FH> || 0;
   chomp $num;
   seek(FH, 0, 0)  or die "can't rewind numfile: $!";
   truncate(FH, 0) or die "can't truncate numfile: $!";
   $num++;
   (print FH $num, "\n") or die "can't write numfile: $!";
   close FH              or die "can't close numfile: $!";

   print LOG "$num\t$$\n";

   close LOCK;
   usleep( 100 );
   last if $num >= 100000;
}
[download]

I ran about four or five processes at the same time and then merged and sorted and made sure there were no duplicates or missing numbers in the logs.

Thanks,

Most of the time kernel locking works fine over NFS. I encountered failures on a mail server with several thousand users at a rate of about once a week. The mail server was Solaris and the clients included a variety of versions of free and commercial UNIXes.

My code to replicate the problem was much more aggressive than yours, spawning hundreds of child processes. I'd suggest modifying your code to take a command line argument and fork that many child processes. Run it with lots of children on both your NFS client and server hosts and see if you can get any failures that way.

It could be that your UNIX client and server have a working, compatible NFS locking implementation between them, but this test will not tell you if your code will work other clients or servers, so why not use a guaranteed to work method like atomically renaming files via link(), preferrably via a standard CPAN module?

Whether flock works or not may depend on the kernel and version of NFS in use.

I am sure that it was not working for tilly when he wrote that. But Linux does not stand still.