NYCPHP Meetup

[Hans Zaunere]

Chris Shiflett wrote on Tuesday, November 15, 2005 9:44 AM:
> Dan Cech wrote:
> > I for one would be very interested in a more in-depth discussion on
> > this subject, as it sounds like there is a lot more to it than I was
> > aware.
> 
> You might find this useful (I did):
> 
> http://netevil.org/node.php?nid=280

That's a good link (although I'm not sure I would agree that select() is a
form of threading) but it's not really the same as trying to run a process
in the background (daemonize).

The crux of a good fork is the double-fork, in addition to destroying the
relationship between the Apache process and PHP (file descriptors).  For
mod_php, this is quite an infamous problem, which is related to things like:

http://www.securitytracker.com/alerts/2003/Dec/1008559.html

(and search for "Need to fork so apache doesn't kill us" for some C
examples).


And:
http://aspn.activestate.com/ASPN/Mail/Message/php-Dev/538021

Which as far as I know, is still an issue in Apache/PHP/etc (not CGI AFAIK).


Anyway, the basic best-practice to fork is something like:


function Daemonize() {

   $pid = pcntl_fork();

   if( $pid < 0 )
      return NULL;

   if( $pid )
      exit(1);

   posix_setsid();

   $pid = pcntl_fork();

   if( $pid < 0 )
      return NULL;

   if( $pid )
      exit(1);

   return posix_getpid();
}
(from http://www.pcomd.net/pfork)


That should completely detach a PHP process from it's Apache process... well
sort of.  Because you're basically now left with an Apache process, and
fork() dupes file descriptors (read a BSD manpage for fork(2)) the child
process (even though it's a session leader and all that) needs to disconnect
from Apache's descriptors (similar to the "controlling terminal" comment in
the C example in the security link).

So what has worked typically is:

fclose(0); fclose(1); fclose(2);

You should be ok in most cases after this.  However, from what I understand,
the Apache forked process inherits *all* file descriptors from Apache, which
is not good (yes, like database connections, etc).  So, in an ideal world -
and especially for very long running processes - you would cycle through all
open file descriptors in the forked PHP process, and close them all.  Then,
of course, reopen the ones you need.

Long story short: if you're running mod_php, don't expect to create a whole
other daemon out of Apache, but it's fine for long running batch processing
from what I've seen.  And, in terms of preventing a user's browser from
"timing out", the important file descriptors to close are the magic three
(stdin/out/err).


---
Hans Zaunere / President / New York PHP
   www.nyphp.org  /  www.nyphp.com