[Mailman-Developers] gate_news gone mad
The Dragon De Monsyne
dragondm@delta.integral.org
Mon, 9 Nov 1998 09:33:33 -0600 (CST)
On Sun, 8 Nov 1998, Corbett J. Klempay wrote:
> First: thanks to Scott...he spent some time on the machine last night
> finding some oddness (I don't know that we're sure yet why it set a bunch
> of the symlinks wrong....but it works now, so hey :)
>
> Now: I got an email when I woke up (just recently :) that mailman was
> spiking the load like mad. Sure enough, watching top revealed that it was
> spawning ~70 mailman processes every few minutes, which drove the load up
> to ~30...the processes would run for <= 5 seconds or so max and be gone.
> I thought it must be a cron job issue, and gate_news is the only thing
> with an interval anywhere near that small (I didn't watch forever to make
> sure it was 5 minutes, but it looked to be in the ballpark). After
> reinstalling that crontab file with gate_news commented it, the spikes
> have stopped. Has anyone else seen weird gate_news behavior before? We
> don't need it here...we don't do any gating. I'm fairly sure I had it
> commented out in the previous (b5) installation, but I'm not positive.
Yup. I've seen this too. And it is gate_news, I've run top on a
very wide xwindow to read the full proc. name. I also don't use the
gateway, and just commented it out.
And I can see, looking at the gate_news code exactly why it will
do that, too.... The way it is currently coded, it will fork
<number-of-lists> proccesses, and _those_ will fork <number-of-lists>-1
proccesses, and _those_ will fork <number-of-lists>-2 proccesses, etc.
The last for loop in gate_news needs a sys._exit() at the end to
fix this.
-The Dragon De Monsyne