[Swift-devel] Coaster CPU-time consumption issue

[Mihael Hategan]

On Mon, 2009-07-13 at 14:11 -0500, Michael Wilde wrote:
> On 7/13/09 2:06 PM, Mihael Hategan wrote:
> > A while ago I committed a patch to run the service process with a lower
> > priority. Is that in use?
> 
> Looks like 22395 is running with a nice value of 10 which I think is 
> what you set in that patch: 22395 aespinos  25  10

Ok. Now, lower priority doesn't mean it won't use CPU. It means that
other processes with a higher priority will get preferential treatment,
and if there is CPU left and the coasters need it, it will be used.

In other words, near 100% CPU usage isn't in itself a problem. While it
shouldn't stay there according to my understanding of the code, if that
is the only problem observed, then I think it's an overreaction.

> > 
> > Also, is logging reduced or is it the default?
> > 
> > Is the 97% CPU usage a spike, or does it stay there on average?
> > 
> > Can I take a look at the coaster logs from skenny's run on ranger?
> > 
> > I'd also like to point out in as little offensive mode as I can, that
> > I'm working 100% on I2U2 and my lack of getting more than lightly
> > involved in this is a consequence of that.
> 
> Right, understood. Any pointers you can give are welcome, and Allan and 
> I are expecting to do the legwork. We'll at least try to find out where 
> the overhead is coming from.

I find it somewhat odd that there was a process with 1GB of virtual
memory use. Are you sure that wasn't a WSRF container from somebody
else?

Can we switch to exclusive evidence mode here (i.e. nothing is
considered unless there is clear proof of it, like a screen dump or log
output, or copy an paste of session from a terminal)?