Re: *** POKED TIMER ***

[Florian Weimer <fw@deneb.enyo.de>]

* Marc Haber:

> May I ask why you pollute the general development mailing list for
> that?

The comment in the code is:

        /*
         * This is a temporary (probably) hack to fix a bug on tru64 5.1
         * and 5.1a.  Sometimes, pthread_cond_timedwait() doesn't actually
         * return when the time expires, so here, we check to see if
         * we're 15 seconds or more behind, and if we are, we signal
         * the dispatcher.  This isn't such a bad idea as a general purpose
         * watchdog, so perhaps we should just leave it in here.
         */

It could be a BIND bug (somewhat unlikely, it didn't appear on the
bind-users list), a libc or kernel bug, or simply an overloaded
system.  Asking on -devel for such cross-package issues should be
okay, I think.

Technically, the BIND code is wrong because pthread_cond_timedwait is
not guaranteed to return within 15 seconds after the requested
timeout.  But if this happens without extremely high load, something
fishy is going on.