Esteemed Monks,

I have a series of Perl CGI scripts that intermittently are very slow. The same scripts have been in use for a long time on the same system with no problem, but this has started only recently -- and system traffic hasn't increased recently.

In investigating the problem I wrote the following test CGI to figure out which part of the CGI is taking the most time, and found that:

• When the scripts are being slow, the entire lag occurs at the end of the BEGIN block ($time4 - $time3). The lag is between 30-120 seconds
• If, during a slow period, I run a similar script that doesn't have a BEGIN block, it runs very quickly. Complicated scripts without BEGIN blocks also run quickly.
• During a not-slow period, everything runs normally, and all timed sections run in 0-1 seconds each.
• The site isn't receiving unreasonable amounts of traffic during these slow periods, and our performance monitoring tools don't show that the CPU, IO, or memory is maxed-out.

We're using Perl 5.00503 under Apache, without mod_perl.

What I'm looking for is some insight/speculation into why the interpreter might lag so much at the end of a BEGIN block.

BEGIN {
  $time1 = time;
  [ DO MISCELLANEOUS TASKS ]
  $time2 = time;
  unshift @INC, "$home/custom/lib";
  $time3 = time;
}
$time4 = time;

print "Content-type:text/html\n\n";
   
print "Begin Block: " . ($time3 - $time1) . "\n";
print "    Miscellaneous Tasks: " . ($time2 - $time1) . "\n";
print "    unshift inc: " . ($time3 - $time2) . "\n";
print "End begin block/begin main body: " . ($time4 - $time3) . "\n";
[download]