Hi,

********
> Are you sure you're performing a close.
Yes, since I am now logging the calls (see below).

> What do you mean by "shutdown the server". ...
Server responds to Ctrl-C by close(2) ing the listening socket and exit cleanly.

>> run on a SunOS computer, ...
> This seems to indicate that there is a problem...
Both the client and server software has worked for years (I know this doesn't prove anything).
However if one end is constantly calling write(2) and the other end is blocked in read(2) there isn't a lot of things that can go wrong (is it ?).

********

Four instances of the server runs on ports 17395..17398, clients issues two connections alternating on 17395/17396 and 17397/17398.

* When communication has stopped:
server issues non-blocking write(2) calls, client is blocked in read(2).

server retransmits continuously:
"17206","79.425849","server-ip","client-ip","TCP","[TCP Retransmission] 17395 > 34386 [ACK] Seq=1295964 Ack=224 Win=6912 Len=1448 TSV=528357141 TSER=3610227624"

Why does the client not respond to this ?
(A tail on the client log at this point in time shows no new entries).
The last client entry is :
10:10:37.406609 IP (tos 0x0, ttl 64, id 58005, offset 0, flags [DF], proto: TCP (6), length: 80) client-ip:34386 > server-ip:17395: ., cksum 0xf106 (correct), 223:223(0) ack 1295964 win 5114 <nop,nop,timestamp 3610227622 528289197,nop,nop,sack 3 {1339404:1933084}{1330716:1332164}{1319132:1323476}>
(It is possible that the tcpdump I/O buffers has not been flushed so that the log is not complete here).


netstat after the communication has stopped:

server$ netstat:
tcp 0 936856 server-ip:17395 client-ip:34386 ESTABLISHED

client$ netstat:
tcp 0 0 client-ip:34386 server-ip:17395 ESTABLISHED

The next thing that happens is that write(2) fails,
write(2) -> -1, errno = 110(TimeOut)
shutdown(fd, 2) -> -1, errno = 107 (Not connected)
close(fd) -> 0

No packets are transmitted when this happens.
And it seems the retransmissions have stopped.

Now the connection does not show up in netstat on the server, but on the client it is the same:

client$ netstat
tcp 0 0 client:34386 server:17395 ESTABLISHED


**********
The server reacts to Ctrl-C by doing a close(2) on the filedescriptor that was passed to bind/listen, and then exits cleanly.

When this happens no packets is transmitted and the
server process no longer exists

*********
Finally I kill the client with Ctrl-C
That cause 9 FIN events to be sent from the client.

netstat on the client shows that the connection is gone.

*********
It seems the client is continuously sending ACK for packet 1295964, why ?
"6512","11.705225","10.21.33.125","10.30.33.154","TCP","[TCP Dup ACK 5345#428] 34386 > 17395 [ACK] Seq=224 Ack=1295964 Win=654592 Len=0 TSV=3610227624 TSER=528289197 SLE=1339404 SRE=1950460 SLE=1330716 SRE=1332164 SLE=1319132 SRE=1323476"

And packets with this id (or whatever it is called) seems to be passed between the server and the client.

(attached two more logs, not sure if they reveal anything else).


Regards,

Even