[syslog-ng]TCP/IP connection - server SIGHUP problem

[Balazs Scheidler]

On Wed, Oct 03, 2001 at 09:55:45AM +0200, Roman Ernst wrote:
> First of all:   thanks for your help with my little TCP/IP
> connection-drop-determination-problem. Of course the idea using a
> monitoring software would solve the problem, but I think many people don't
> have / use such a tool...
> 
> So... it should be possible to catch such an error by other means...
> 
> Somebody wrote that i could figure out via "netstat -an | grep tcp ...."
> whether connection is okay or not. Unfortenately when plugging out the
> network cable AIX does not recognize this, i.e. with netstat -an the
> connection seems to be still available. Only the client tries to close the
> connection when he tries to send a log message and can't reach the log
> server (after TCP timeout...). So... the client isn't the problem.
> 
> But the server never tries to send a package to the client... so he will
> never recognize whether the client is available or not.
> 
> Maybe it would be useful to introduce some sort of TCP heartbeat (server
> sends some packages to the client every xxx minutes) . This way the server
> take care of unreachable clients and logs such problems...
> 
> If I'm totally wrong with my opinion please let me know...

you are right. syslog-ng should enable TCP keep alives, so it recognizes
when the connection is broken. this is not currently done, however I add it
to my todo list.

> 
> 
> Okay... another problem:
> 
> Sending a SIGHUP to the syslog-ng client results in restarting syslog-ng on
> the client, dropping connection to server, reconnecting to the server and
> initializing the new configuration     (=> that's what the server log
> says...)
>    =>  is it really necessary to drop the connection?
> 
> Sending a SIGHUP to the syslog-ng server results in restarting the
> syslog-ng on the server and initializing the new configuration (that's what
> the server log says...)
>    => So far so good... now the problem.  If the client tries to send
> messages to the server the following happens   (CLIENT and SERVER LOG)
> 
>      AF_INET client connected from 10.x.x.1, port xxxxx   (SERVER LOG)
>      io.c: do_write: write () failed (errno 32), Broken pipe  (CLIENT +
> SERVER LOG)
>      pkt_buffer::do_flush(): Error flushing data (CLIENT + SERVER LOG)
>      Connection broken to AF_INET 10.x.x.2, reopening in 10 seconds (CLIENT
> + SERVER LOG)
> 
>      AND:   one log message is missing afterwards...
> 
> Okay...  that's it.  Maybe someone can help me. Thanks in advance!

the client always drops the connection, the server can keep it alive by
specifying the keep-alive(yes) option to your source statement.

-- 
Bazsi
PGP info: KeyID 9AF8D0A9 Fingerprint CD27 CFB0 802C 0944 9CFD 804E C82C 8EB1