[Freeciv-Dev] Re: the core files

[Paul Zastoupil <paul@xxxxxxxxxxxxx>]

On Fri, May 23, 2003 at 09:46:14PM +0100, Vasco Alexandre Da Silva Costa wrote:
> On Fri, 23 May 2003, Reinier Post wrote:
> 
> > > I suspect that this is the result of an "ulimit -t" call. *searching*
> > > Yes there is such code in the start script:
> > >
> > >   cpulimit=5000
> > >   ulimit -S -t $cpulimit
> > >
> > > So a server is killed if it uses more then 5000s CPU time.
> >
> > I have no idea what a reasonable value would be.
> > We can wrap the civserver call in "time" to record it.
> 
> I guess the new AI or something else uses more CPU cycles than it used
> to and it wastes the allocated slice of CPU time.
> 
> But what is the point of adding limits to overall CPU consumption? That
> seems to be a bit hard to me.
> What if someone runs a game with many AI's or for a long time and wastes the
> slice anyway?
> Why not just 'nice' the civserver processes? IIRC pubserver runs in a
> separate machine so it is not like we risk losing essential services for
> a short time if a particular process starts hogging CPU.
> 
> In alternative civserver could catch SIGXCPU and exit(0) instead of with
> an error. No core dumps but you would get some angry players who got
> their server nuked in mid game. I do not like that solution much.
> 
> Comments?

Yes, the limit is removed.  I'm not going to argue that it should be
reinstated, however there is a very positive side to having a limit.

True, this is its own machine, however there is more than one person
using it.  Nice(1) wont really help us. Civservers are pubservers most
important job, and we wouldn't know which civserver to nice.  We don't
want 3 different people starting up 30 AI games while 7 people are trying
to play a normal game with each other.  We also don't want some bug to
cause a civserver to go nutty.

With these things in mind is the exact reason why I will never put
pubserver on a non smp machine.

-- 
Paul Zastoupil