TCP Connection - LAST_ACK

[Joanne Keegan]

Hi Everyone,

I'm looking for a bit of advice. We are running HP-UX 11.00 with UniVerse.

We stop and restart UniVerse when we split our EMC disks prior to backups every night. We have previously had problems restarting UniVerse due to a tcp connection in the FIN_WAIT_2 state. We have catered for this by modifying the UniVerse stop/start script to do a ndd tcp_disconnect. This works well - as long as the client is known to the host in the /etc/hosts file.

Last night, a tcp connection with the status of LAST_ACK caused a problem when restarting UniVerse. We had to reboot the system to get things going.

Has anyone else experienced something similar, and how did you get around the problem? Has anyone written a script to disconnect such a connection when the IP address is not in the hosts file?

Any assistance will be appreciated.

Please let me know if you need any further information.

Regards,

Jo

[rick jones · Accepted solution]

DON'T use the ndd disconnect kludge. INSIST that UniVers fix their app to set SO_REUSEADDR before trying to bind() to its well-known port number(s).

Once that bug in UniVers is fixed, it will be able to restart with TCP connections in any state other than LISTEN.

All this ndd disconnect stuff is simply a MASSIVE KLUDGE to workaround app bugs...

Unlike FIN_WAIT_2, LAST_ACK is an active retransmit state - there is a FIN outstanding for which the local system is awaiting an ACK. So, it should go away in no more than tcp_ip_abort_interval milliseconds

[Joanne Keegan]

Thanks for your timely response Rick. I will contact the UniVerse app vendors for them to resolve.

Regards,

Jo