perl-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Stas Bekman <>
Subject [mp2] preventing (push|set)_handlers from registering phases that can't be run
Date Wed, 31 Aug 2005 00:54:19 GMT
At the moment you can say:

   $s->push_handlers(PerlChildInitHandler => 'foo');

after the child process has been spawned (e.g. in the response phase). Of 
course it won't do anything. I suggest that we prevent this from happening.

I've a real use for this case.

At the moment if before the child processes are spawned any perl code 
(e.g. is calling srand() or rand(), the child processes will 
all have the same IV for rand and all generate the same rand sequence 
which is not good. I wonder if we should try to fix that in mod_perl, or 
do you think someone will want this as a feature? The issue is 
perl-specific really:

% perl -le 'srand();  fork(); print "$$: " . rand();'
6326: 0.813502754391269
6325: 0.813502754391269

interestingly enough it doesn't happen with threads:

% perl -Mthreads -le 'srand(); threads->new("x")->join; x(); \
sub x { my $tid = threads->self->tid; print "$tid: " . rand()}'
1: 0.924164554876374
0: 0.306549259319098

at the moment under mp2 we are consistently giving the same sequence, be 
it prefork or worker.

In any case here, at, we try to solve this bug by making 
sure that ChildInit runs srand() to avoid this problem (so each process 
has its own seed), but it's so easy to forget to configure it, if you give 
this code to users. So we were wondering how to ensure this execution. I 
have suggested that if we ensure that

   $s->push_handlers(PerlChildInitHandler => 'foo');

can be only run before PerlChildInitHanlder phase (or during it), than we 
can easily ensure that the phase will be configured by doing this at the 
compile time:

  package foo;
  use Apache2::ServerUtil ();

  Apache2::ServerUtil->server->push_handlers(PerlChildInitHandler => 'foo');

now if the module is loaded at the server startup, it'll register the 
callback just fine. If not and it'll be attempted to load after the server 
startup, that push_handlers call will fail, saying:

   "It's too late to register PerlChildInitHandler"

What do you think?

Later we may extend this to other phases as well (e.g. prevent registering 
Auth handler, from response phase). of course assuming that we find an 
efficient way to do that.

I wonder if we should just srand() inside mod_perl, I should probably also 
ask this at p5p, since fork() and threads() are inconsistent in that 
aspect. since threads do have different seeds

Stas Bekman            JAm_pH ------> Just Another mod_perl Hacker     mod_perl Guide --->

To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message