On Thu, Mar 10, 2005 at 08:31:18AM -0500, Dan Mahoney, System Admin wrote: > Actually, my experience has been that the child in this case doesn't > respond to kill -TERM, even after larger delays like five minutes with > repeated kill'ing. Kill -9 seems to cause another process to pop up with > similar (instant) loads. Nuking the users' bayes/awl databases does not > seem to fix the problem.
Well, if the process is ignoring signals, only SIGKILL will do you any good. There aren't a lot of places we do that though: spamd child spawning, Bayes upgrading, Bayes expiry, Bayes Journal Sync, and Bayes Restore DB. So if a kill -INT or -TERM doesn't work, either you're in one of these places, or the process is stuck waiting for I/O somewhere (can't deliver signals when stuck in the kernel). Doing a SIGKILL will nuke the process, so if another child comes up and before getting a message gives "instant load", I'd guess there's an issue in the spawning area. If the new child gets a message to process and the problem starts occuring, that's a different issue. But the question I was responding to was how to get a "stuck" message through unmarked, which a SIGKILL ought to do for you. -- Randomly Generated Tagline: "Aiee!" - Linux kernel error message
pgpOfbEFQq19k.pgp
Description: PGP signature
