I am currently using kill -9 to restart the fcgi process

Don't do that! Signal 9 is SIGKILL, and a process which receives this signal has no chance to

• flush buffers
• close filehandles properly
• clean up resources
• execute anything compiled into END blocks

since it is not trappable, and AFAIK not propagated at all to the signalled process, but an instruction to the kernel to reap the process immediately (although the parent of the reaped process gets noified).

Use signals 1, 2, 3 or 15, not 9. Signal 9 is last resort only. Try

perl -le '$SIG{KILL} = sub {print "boom!"}; print $$; sleep;'
[download]

and kill that process from another shell with SIGKILL: no output.

But my questions keep:

1. the '-M' test looks expensive if I put it in the FCGI loop and do it on every request
2. what are the differences between 'require' and the suggested "do" here?, given that I do need to import some functions from that modules
3. What should be considered differently between an OO-modules and non-OO-modules in this context..

Instead of using the rather obscure -M file test (which calls stat under the hood), I'd go with (stat $module)[9] i.e. mtime. Set up a $hash {$modulepath} = $timestamp at script start, preferrably using the key in %INC for easy comparison. Expensive? That's relative to what else your fcgi application is doing.

Also, you don't have to check the module files on every request. Set up a signal handler for e.g. SIGUSR1 (on the shell command line, kill -l will show you the available signals and their numbers) which sets a package global variable to some true value, and check that variable at every request. If it's true, reload the module (take care to undef $INC{'Some/Module.pm'} first) and clear the variable.

Differences between require and do:

• require propagates $@ from inside modules at compilation, whereas do does not.
• do evals a file's content regardless of the contents of %INC, whereas require just returns 1 if the module's path (relative to any directory in @INC if found in one of those, or an absolute path if not) exists as key in %INC.
• require dies on compilation errors whilst evaling the required file - or if the required file doesn't return a true value -, whereas do does not.

See perlfaq8 for any bits that I missed here (or perldoc -q require).

For the suggestion-(2), I am not sure which SIG should I trap to handle the file change

Any but ALRM, CONT, ILL, ABORT, KILL, SEGV, ... I'd go with USR1 or USR2. But as stated above, I'd just set a variable in that signal handler, but not process the file immediately from somewhere inside your request processing where you could be using that module; instead delay that to the next request cycle.

_($_=" "x(1<<5)."?\n".q·/)Oo.  G°\        /
                              /\_¯/(q    /
----------------------------  \__(m.====·.(_("always off the crowd"))."·
");sub _{s./.($e="'Itrs `mnsgdq Gdbj O`qkdq")=~y/"-y/#-z/;$e.e && print}

if ( -M "/path/to/lib/MyModule.pm" < 0 ) {
        require MyModule;
        MyModule->import(qw[ :mytag ]);
    }
[download]

You have to delete the corresponding entry in the %INC hash before you do a new require MyModule.

But do have a look at the source code of Apache::Reload. Other than the way to bind it into the webserver's request cycle, I think you can almost fully re-use Apache::Reload's code.

the '-M' test looks expensive if I put it in the FCGI loop and do it on every request

Hi, guys:

1. I will write a separate HTML file which contains lines like

   { 
     local $SIG{USR1} = \&myapp_reload_module($modulename);
     kill("USR1", -$$);
   }
   [download]

   where $modulename comes from a select-options box in this HTML file(or from a form submission). myapp_reload_module is a customized module function which reloads a module by following the ways in Apache::Reload(I will read that source though).
2. Each time I modified a module, I only need to open this webpage, select that module from a dropdown list(select-option box), and then submit the form, the above Perl code with kill(USR1,-$$) will trigger the reloading of the selected module, Right?

I will not use kill -9 for sure, thanks shmem for this nice tip.

Best regards,

Lihao(XC)

Updated:missed out using kill(USR1, -$$) to send the signal:)

Just a follow-up, I followed the above idea and have made it work for my FCGI process, pretty happy.. thanks again. :- )