| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 8496.1 | | SMURF::MCCANN | | Mon Jan 20 1997 09:00 | 18 |
| 8496.2 | | TPOVC::SIMONLEE | | Fri Mar 21 1997 11:48 | 11 |
|
I got the same problem and checked the reference page of setsockopt but
did not see TCP_KEEPIDLE option. Is it in old version or does it belong
to different protocol level? I also checked the tcp_keepalive value of
kernal and it was 1440. If its unit is second, then it is 24 minutes
instead of 2 hours. How come? If I don't set SO_KEEPALIVE, will the
socket never disconnect? Please advise. Thanks in advance.
/Simon
|
| 8496.3 | | TPOVC::SIMONLEE | | Sun Mar 23 1997 20:32 | 3 |
|
Sorry, I made a mistake, the keepalive value in kernel is 14400, that
is 4 hours.
|
| 8496.4 | 2 hours is right | SMURF::DUSTIN | | Mon Mar 24 1997 10:24 | 5 |
| The keepalive value is specified in units of 1/2 seconds, so it's
actually 2 hours, not 4 hours.
John
|
| 8496.5 | SIGPIPE not sent for TCP_KEEPALIVE | TPOVC::SIMONLEE | | Mon Mar 24 1997 20:40 | 24 |
|
Thanks!
About the setsockopt(), just the other days, I wrote a program to test
the SO_KEEPALIVE option. I set the tcp_keepidle to 60 and tcp_keepintvl
to 5, so I didn't need to wait for two hours for the tCP to probe the
other party. I connected to my program (server) from a client and used
tcpdump to examine the traffic. I unplugged the network cable from the
client and after a moment (I believe is the time I set for
tcp_keepidle), the server started to probe the client. Everything
seemed good, after 10 probes, the connection is broken and the server
exited. This is what I didn't expect because I had captured the PIPE
signal and in the signal handler, I did something, but those things had
not been done. That meant the PIPE signal had not been sent. If I sent
PIPE signal explicitly with "kill -pipe", the handler did be called.
According to 0. and reference manual, system should send a pipe signal
to the process using the socket, but it didn't. What could be the
possible cause of it? Does anybody have the same experience?
I am using UNIX V3.2C. Please advise. Thanks.
/Simon
|
| 8496.6 | | VAXCPU::michaud | Jeff Michaud - ObjectBroker | Mon Mar 24 1997 21:10 | 10 |
| Re: .5
I'm not sure I understand what you are saying, but do note that
a SIGPIPE signal is only sent syncronously when you write(2) or
send(2) on a pipe (or socket, etc) that is not connected.
Ie. you will not get a SIGPIPE signal at the time the transport
has marked the socket as disconnected. You should however get
a SIGIO, if you've marked the socket for async io, and setup
a signal handler.
|
| 8496.7 | | TPOVC::SIMONLEE | | Thu Mar 27 1997 21:28 | 28 |
|
Re: .6
This is what the setsockopt() MAN page says about SO_KEEPALIVE option:
SO_KEEPALIVE
Keeps connections active. Enables the periodic transmission
of messages on a connected socket. If the connected socket
fails to respond to these messages, the connection is broken
and processes using that socket are notified with a SIGPIPE
signal.
It does not say wheather the processes using the socket is doing
read/receive or write/send will be notified with SIGPIPE. If, according
to you, the SIGPIPE is only sent to the processes that are doing
write/send on a connected socket, then the phenomenon is right because
my program is doing receive operation. If the socket will be
disconnected automatically after failed keepalive probes, that's fine
and it means I don't need to capture the SIGPIPE signal and close the
socket explicitly. Any comments?
/Simon
|