On Wed, 2006-03-01 at 10:01 -0500, Andreoli, Tony A. USNUNK NAVAIR B1490, R215 wrote:
I'm using 1.6.9 (upgraded from 1.6.6 because I was seeing the same problem). I have it running on 8 different servers at different locations, some are SMP, some aren't. On these hosts, we have anywhere from 2 to 14 devices logging to the servers, some via 514/tcp, others via 514/udp. All of the loggers typically sit with a load average < 1 (usually not even registering), and a cpu idle of 99%. 7 of these remote loggers also log to our local machine, but only 5 lines every 2 minutes (for stats).
What I've noticed (and I've seen this on all of them at one time or another), is that syslog-ng just stops. ps shows it running, but the log file (/logs/messages) never changes. If I tcpdump on the interface that it's listening on, I see traffic, and it seems that the act of tcpdumping causes the log file to start to grow again, then a little while later, it may stop again. It's sporadic though, on one of my systems, it hasn't done it in over 2 months, on another, it's done it 3 times today.
I've pulled out my last hair and still haven't come any closer to a solution. I've recompiled the source, loaded 3 different versions, etc. The only thing common is that all of these systems are running RHEL3.
Don't you happen to read /proc/kmsg by both syslog-ng and klogd ? That is a known bad situation and the symptoms are exactly what you describe. (poll indicates readability but by the time syslog-ng gets to read the file the data has already been read) This is documented in the FAQ as well. -- Bazsi