Hello. I'm curious as to how one might limit the number of simultaneous spamassassin processes.
I run spamassassin as a procmail filter. It's a fine program, except when I have a bunch of messages waiting. In that case, here's what happens. 1. I type `fetchmail'. Fetchmail starts downloading ~50 messages. 2. As fetchmail runs, sendmail spawns procmail processes, which spawn spamassassin processes. 3. About halfway through the mail transfer, the number of processes hits a critical point and the system runs out of memory, starts thrashing the swap partition, and remains wedged for about 20 minutes. 4. Eventually, the spamassassins finish, and the system responds to input again. Is there any way to rate-limit spamassassin spawning? I thought about ulimiting the sendmail daemon and its children, but that seems heavy-handed. I could renice the processes, but this is a memory problem, not a CPU-time problem. I know that exim has a limit-simultaneous-deliveries setting, which, when cranked down to 10, prevented the thrashing, but meant that mail delivery was very slow, as messages after the first ten were dumped in the queue to wait for the next run (the same happened to the next batch, of which the first ten were delivered, and so on). I eventually gave up using Exim because I missed DSN support. Would spamc/spamd respond better to an inrush of mail? Or is there a switch to flip to make procmail or sendmail process the mail serially -- not by re-queuing it, but by using some form of locking such that when one spamassassin terminates, another is immediately started? Thanks, --Joe
msg07051/pgp00000.pgp
Description: PGP signature
