Lighttpd

I like to preconfigure lighttpd with knowledge of several FCGI backend processes, and then rely on its load-balancing behaviour to detect which of those processes are available. This allows me to manually start/stop any number of FCGI backends without bouncing lighttpd.

With UNIX sockets, a missing socket file is like a TCP connection error, so when Lighttpd finds that a socket file is missing, I would expect it to wait "disable-time" before looking for the file again.

What actually happens is that Lighttpd spins without a delay, filling up the disk rather quickly with its error log file. I would be very happy if this "hard" error condition could be changed to a "soft" error.

For reference, my configuration looks something like this:

fastcgi.server =  (
    ".fcgi" =>
    ("fcgi-01" => ( "socket" => "../fcgi-01/fcgi.sock", "disable-time" => 5 ),
     "fcgi-02" => ( "socket" => "../fcgi-02/fcgi.sock", "disable-time" => 5 ),
     "fcgi-03" => ( "socket" => "../fcgi-03/fcgi.sock", "disable-time" => 5 ),
     "fcgi-04" => ( "socket" => "../fcgi-04/fcgi.sock", "disable-time" => 5 ),
     "fcgi-05" => ( "socket" => "../fcgi-05/fcgi.sock", "disable-time" => 5 )
    ))

-- stephen_purcell