Too many FastCGI servers with cgi-fcgi
Ben Engber (ben@benhur.condenast.com)
Tue, 29 Apr 1997 18:49:37 -0400
Date: Tue, 29 Apr 1997 18:49:37 -0400
From: Ben Engber <ben@benhur.condenast.com>
Message-Id: <199704292249.SAA10827@benhur.condenast.com>
To: fastcgi-developers@OpenMarket.com
Subject: Too many FastCGI servers with cgi-fcgi
We've been running our FastCGI programs through the cgi-fcgi bridge that came
with the FCGI libraries. As we increase the load on the application the FastCGI
servers start to multiply, until eventually the entire application ceases to
function.
Specifically if we set cgi-fcgi to start 1 server and hit the application with
10 clients, the application spawns 50 or more servers and then dies. I would
have thought it couldn't possibly have gone higher than 10 (each client does
nothing until the request completes, so there are at most 10 concurrent
requests).
If we set 10 servers it behaves fine for 10 clients.
But after a certain number of clients are added (maybe 20) no number of FastCGI
servers is enough to prevent the whole thing from dying.
Where are these processes coming from? How does cgi-fcgi balance across the
FCGI servers? Is this handled automatically by the way the sockets function?
Has anyone else had this problem?
We have reproduced this on both Netscape and Apache web servers on Solaris 2.5.
I believe we are using FastCGI version 1.5 (how can I tell from the binary?)
Can I upgrade to a new version of cgi-fcgi without recompiling the FCGI
application?
Any help would be greatly appreciated.
-Ben Engber
Senior Programmer
CondeNet