Multi-threaded FastCGI servers

Hello,

I'm in the process of modifying a single-threaded FastCGI server. I'm 
trying to make the server multi-threaded by using Posix threads and the 
FastCGI API's in one of the two ways, as follows:

pcreate_thread(&AcceptThread, 0, (void*)&AcceptThreadFunc, 0);
...
AcceptThreadFunc(...)
{
     FCGX_Request FcgiReq;
     FCGX_InitRequest(&FcgiReq, 0, 0);
     while (FCGX_Accept_r(&FcgiReq) >= 0)
     {
         // Do the work.
     }
}

Or:

FCGX_Request FcgiReq;
FCGX_InitRequest(&FcgiReq, 0, 0);
while (FCGX_Accept_r(&FcgiReq) >= 0)
{
     pcreate_thread(&WorkerThread, 0, (void*)&WorkerThreadFunc, 0);
}
...
WorkerThreadFunc(...)
{
     // Do the work.
}

I have read that it is possible when using multi-threading in FastCGI 
to:

a). Put the entire accept loop into the worker thread, as in the first 
code segment (or the threaded.c example), or
b). Have just a single accept loop and put just the code within the 
accept loop into a worker thread.

Is this correct? If it is, option b) would imply just a single instance 
of an FCGX_Request variable meaning, in turn, that only one FcgiReq.out 
would ever be used/accessed by, say, FCGX_FPrintF/PutS. How is this 
different from just using FCGI_stdout in a single-threaded FCGI_Accept 
loop?

I understand that FCGI_Accept is not thread-safe, whereas FCGI_Accept_r 
is, but in option b) the accept loop is outside of the worker thread, 
so does it matter which accept API is used? I guess, it might if 
FCGI_fwrite, say, is not thread-safe, since that would be used in an 
FCGI_Accept loop, but not in a FCGX_Accept_r accept loop...

Basically, is either a) or b) a preferable solution? Is one better in 
some circumstances, but not others?


Regards,
Peter Colson.

___________________________________
fastcgi-developers mailing list
http://fastcgi.com/fastcgi-developers/