Overloaded server

[Steffen Grunewald]

On Fri, Jun 22, 2007 at 05:08:14PM +0200, Thomas Lange wrote:
> >>>>> On Fri, 22 Jun 2007 16:42:40 +0200, Steffen Grunewald <steffen.grunewald at aei.mpg.de> said:
> 
>     > yesterday I tried to FAI setup about 600 nodes using one file/webserver.
> Wow. Did you start them all at once? Or with some time (how long)
> between each poweron? I would have expected more problems with TFTP.

Spread over about 20 minutes, in batches of about 100. :-)

>     > was no problem, I got lots of http download errors (resulting in that
>     > famous Broken package, and error 25600 messages).
> Maybe the webswerver only allows a certain amount of connections at a
> time? How was the CPU load?

don't know. standard apache (2?) of Sarge.
CPU load was not too high I guess ... I'm installing 50 nodes right now, and
some nfsds are busy, load <<5. I've seen loads up to 60.
Is there a similar limitation to apache threads as it is with nfsds?

> I don't understand that.

grep -q "Broken package" $LOGDIR/fai.log && reboot

>     > but that
>     > would deprive me of the correctly downloaded error messages... how much 
>     > effort would it be to run the savelog task before the chboot task? (and
>     > has there be a reason for the current order?)
> Just log into a machine and read /tmp/fai/*.log

I cannot if it reboots itself. Reboot by hand is counterintuitive with FAI.

The only (small) problem is that I cannot track machines which might have
different problems, cycling through the instell for ever...

So why does chboot precede savelog? Any reasonable reason?

S