Re: [RFC] log out-of-virtual-memory events

[Robin Holt <holt@sgi.com>]

On Fri, May 18, 2007 at 09:50:03AM +0200, Andrea Righi wrote:
> Rik van Riel wrote:
> > Andrea Righi wrote:
> >> I'm looking for a way to keep track of the processes that fail to
> >> allocate new
> >> virtual memory. What do you think about the following approach
> >> (untested)?
> >
> > Looks like an easy way for users to spam syslogd over and
> > over and over again.
> >
> > At the very least, shouldn't this be dependant on print_fatal_signals?
> >
> 
> Anyway, with print-fatal-signals enabled a user could spam syslogd too, simply
> with a (char *)0 = 0 program, but we could always identify the spam attempts
> logging the process uid...
> 
> In any case, I agree, it should depend on that patch...
> 
> What about adding a simple msleep_interruptible(SOME_MSECS) at the end of
> log_vm_enomem() or, at least, a might_sleep() to limit the potential spam/second
> rate?

An msleep will slow down this process, but do nothing about slowing
down the amount of logging.  Simply fork a few more processes and all
you are doing with msleep is polluting the pid space.

What about a throttling similar to what ia64 does for floating point
assist faults (handle_fpu_swa()).  There is a thread flag to not log
the events at all.  It is rate throttled globally, but uses per cpu
variables for early exits.  This algorithm scaled well to a thousand
cpus.

I think this may be a good fit.

Good Luck,
Robin
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/