Subject inet_error
Author Scott Taylor

I am running FB/IB Super Server 0.9-4 on a Linux 2.2 server. Using
ibobects and Delphi, for the front end, on Win2K Pro workstations.

Every once in a while, Win2K will blow the TCP stack and the interbase.log
file shows these error messages:
intrbase.dct.UUCP (Server) Mon Dec 10 12:46:01 2001
INET/inet_error: read errno = 104

intrbase.dct.UUCP (Server) Mon Dec 10 12:46:01 2001
Super Server/main: Bad client socket, send() resulted in SIGPIPE,
caught by server
client exited improperly or crashed ????

the last error shows up multiple times (10-12). When that happens the
clients can not reconnect, even after restarting Win2K. Looks to me like
the connections are still on the server so the only way to recover is to
kill the PIDs and restart ibserver. "netstat -t" shows the clients still
connected even when the client machine is shut down.

Does anyone know what could be wrong, what I could look for, or perhaps a
work around? Maybe a trap for this SIGPIPE that could remove the bad
client socket?

I actually wrote a Perl script that the users could execute to restart the
ibserver, I really hate that. But, still a lot better than rebooting an NT
server, running MS SQL, at 2AM. :)

Thanks for any help.

Scott Taylor
Systems Administrator
DCT Chambers Trucking Ltd.