We've written a custom AOLServer module to provide our software's
functionality directly via an HTTP interface. On Solaris, it runs as a
shared library. In front of this is a series of Perl scripts (yes, yes, I
know, can't rewrite this right now) also served by the same nsd instance
that query our module and rewrite the results for presentation to the user.
We use an Ns_MutexLock/Ns_MutexUnlock pair to ensure that we don't have
reentrancy in the module (again, yes, yes, I know). Our Ns_Mutex is static,
and is initialized using Ns_MutexInit. Here's a sample from a small
function:
int GetStatistics(Ns_OpContext context, Ns_Conn *conn)
{
Ns_MutexLock(&SpNebulaMutex);
//Get the statistics from the HTTP administrator
NQAdmin::Result statResult = SHTTPAdmin->GetStatistics();
//Send the resulting HTML to the server
int Result = Ns_ConnReturnHtml(conn, statResult.m_iResult,
const_cast<char *>(statResult.m_sResult.c_str()),
statResult.m_sResult.length());
Ns_MutexUnlock(&SpNebulaMutex);
return Result;
}
NOW, the problem:
We in some instances appear to have a reentrancy problem. Not 100% sure,
but in some heavy loading cases, a function that would fail if a re-entrant
call was made seems to fail with alarming regularity. The first thing we
did was change the limit (of CGIs) to 1. That helps, but not enough-it
still fails in the same way.
Any suggestions? Thanks in advance....
--
Bill Hofmann ph: 510-525-7637
VP Engineering mobile: 510-409-0395
ZNOW Inc., 2342 Shattuck Ave #314, Berkeley CA 94704
