[Speedycgi-users] Apache/Speedy duo problem: running out of file descriptors

[PW <sub...@me...>]

Since I installed the latest SpeedyCGI package 2.21 a couple of months 
ago, running as combo with Apache/1.3.23 (Linux, approx. 70 virtual 
hosts - thereof 14 running Perl applications under SpeedyCGI), I 
experience never encountered problems before.

Problem description:
After around 3 days (probably depending on traffic) our Apache Server 
runs out of the hardcoded per user file descriptors (1024), bringing all 
Web services to a halt. Mentioned Perl applications are clean and close 
all of their file descriptors. However, when checking the Server logs 
and querying lsof/netstat -a, I always notice an aweful lot of open 
Socket connections, typically around 500+ by speedy_ba per process:

## 509 UDP port connections altogether for one process!
speedy_ba 14332 wwwrun  237u  IPv4     308290               UDP 
ns.megapublic.net:28952->ns.megapublic.net:domain
speedy_ba 14332 wwwrun  238u  IPv4     308414               UDP 
gateway.megapublic.net:28966->ns3.magnet.ch:domain
speedy_ba 14332 wwwrun  239u  IPv4     308415               UDP 
ns.megapublic.net:28967->ns.megapublic.net:domain
speedy_ba 14332 wwwrun  240u  IPv4     308431               UDP 
gateway.megapublic.net:28970->ns3.magnet.ch:domain
speedy_ba 14332 wwwrun  658u  IPv4     495600               UDP 
ns.megapublic.net:7764->ns.megapublic.net:domain

Once the "Too many open files" error appears, I have to kill all 
SpeedyCGI processes and restart Apache in order to get into business again.

Questions:
1. Why does SpeedyCGI listen on such a lot of ports?
2. Is there a possible bug in that SpeedyCGI forgets to close once 
opened sockets after exiting?

Thanks for any insight!

P. Wiede